ABSTRACT
Title of dissertation: THREE TRANSACTION COST
ECONOMICS ESSAYS
WHICH USE ROMANIAN DATA
Radu Adrian P?aun
Doctor of Philosophy, 2007
Dissertation directed by: Professor Peter Murrell
Department of Economics
The first chapter of the dissertation considers the complexity of contracts to be
determined by transacting partners based on their exposure to the opportunistic be-
havior of the other. A transaction cost economics model generates the hypothesis that
buyer and seller relationship-specific investments have opposite effects on the expo-
sure to opportunism, hence on contract complexity. The precise direction of the effect
is thought to depend on partners? relative vulnerability. The treatment-effects model
estimated by maximum likelihood indicates that sellers? relationship-specific invest-
ments increase contract complexity, while buyers? investments reduce it. This is the
first transaction cost economics analysis that simultaneously counters the problems
of unobserved heterogeneity, generates estimates of the effects of relationship-specific
investments that are opposite in sign on opposite sides of the agreement, and explains
the patterns in the biases of ordinary least squares estimates. The quality of court
services and the impact of buyers? prepayments are also investigated.
The second chapter presents a simple methodology for measuring transaction
costs at agreement level by using reports from business officials who supervise compa-
nies? buying and selling activities. In a practical implementation in Romania, those
transaction costs directly related to the buying and selling activities are assessed as
large, accounting for more than a fifth of value added. The recorded transaction
costs estimates correlate significantly with variables suggested by theory, indicating
validity. The quality of the data is also analyzed.
The third chapter of the dissertation investigates the determinants of trans-
action costs by using the information collected by the survey question proposed in
the second chapter. Given the limited nature of the data, the Tobit model is first
employed. The sample selection model is then adopted. However, combined evidence
indicates that a two-equation approach is more appropriate. Results show that the
existing theory is somewhat successful at predicting the size of transaction costs and
very successful at predicting the existence of these costs. The two-sided nature of
the decision to invest in relationship-specific assets is discussed, and the potential
endogeneity of several factors is investigated.
THREE TRANSACTION COST ECONOMICS
ESSAYS WHICH USE ROMANIAN DATA
by
Radu Adrian P?aun
Dissertation submitted to the Faculty of the Graduate School of the
University of Maryland, College Park in partial fulfillment
of the requirements for the degree of
Doctor of Philosophy
2007
Advisory Committee:
Professor Peter Murrell, Chair/Advisor
Professor Daniel R. Vincent
Professor Ginger Zhe Jin
Professor Christopher McKelvey
Professor Curt Grimm
c? Copyright by
Radu Adrian P?aun
2007
P?arint?ilor mei, Gheorghe ?si Ani?soara
ii
ACKNOWLEDGEMENTS
I would like to acknowledge and thank the many people who helped me bring
this dissertation to fruition. First and foremost, I extend my deepest gratitude to my
academic adviser, Professor Peter Murrell, for giving me the opportunity to pursue my
Ph.D. in USA and under his guidance. His understanding and expertise in the field
added greatly to my graduate experience, while his constant interest in the subject
matter made this dissertation possible.
I am deeply indebted to Mrs. Elizabeth Allehaut for giving me the chance
to be part of the IRIS Center Romania team. I must express my appreciation to
her and to my many friends in the Bucharest office for making my job there such a
pleasant experience. They have also offered the help Professor Murrell and I needed
in collecting the data this dissertation rests upon.
I also owe a special debt of gratitude to Dr. Cynthia Clement of the IRIS Center
for being such a good friend and for making things so much easier for me during the
graduate program.
Last but not least, I am grateful to have such a wonderful family, which has
provided continued and unwavering support throughout these years.
iii
TABLE OF CONTENTS
List of Tables vi
List of Figures viii
List of Abbreviations ix
An Overview of the Dissertation 1
1 The Contingent Effect of Partners? Relationship-Specific Investments: Evi-
dence from Romania on the Use of Complex Contracts 7
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2 An overview of the opportunistic behavior . . . . . . . . . . . . . . . 13
1.2.1 Opportunistic behavior: notion and forms . . . . . . . . . . . 13
1.2.2 Solutions to opportunistic behavior . . . . . . . . . . . . . . . 14
1.2.3 Complex contracts as an alternative to vertical integration . . 16
1.3 Some related existing studies . . . . . . . . . . . . . . . . . . . . . . . 19
1.3.1 Previous results on contract complexity . . . . . . . . . . . . . 20
1.3.2 Previous results on dependence balancing . . . . . . . . . . . . 23
1.3.3 Concerns with the existing studies . . . . . . . . . . . . . . . . 25
1.4 Theoretical support of our analysis . . . . . . . . . . . . . . . . . . . 28
1.4.1 A description of the buyer-seller relationship . . . . . . . . . . 28
1.4.2 Support for the vulnerability assumption . . . . . . . . . . . . 30
1.4.3 A simple theoretical model . . . . . . . . . . . . . . . . . . . . 33
1.5 The econometric details of our analysis . . . . . . . . . . . . . . . . . 36
1.5.1 The estimation approach . . . . . . . . . . . . . . . . . . . . . 36
1.5.2 Predictions which can be taken to data . . . . . . . . . . . . . 40
1.6 The survey and main variables . . . . . . . . . . . . . . . . . . . . . . 44
1.6.1 The survey generating our data . . . . . . . . . . . . . . . . . 45
1.6.2 The main variables we employ . . . . . . . . . . . . . . . . . . 47
1.7 A hint on the existence of hold-up . . . . . . . . . . . . . . . . . . . . 52
1.8 The results of our analysis . . . . . . . . . . . . . . . . . . . . . . . . 54
1.8.1 The effect of relationship-specific investment . . . . . . . . . . 54
1.8.1.1 Ordinary least squares estimations . . . . . . . . . . 56
1.8.1.2 Maximum likelihood treatment-effects estimations . . 58
1.8.1.3 Instrumental variables estimation . . . . . . . . . . . 64
1.8.2 Robustness tests of the main results . . . . . . . . . . . . . . . 66
1.8.3 Additional results: the effect of buyer prepayments . . . . . . 69
1.9 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
iv
2 A Proposed Way to Measure Transaction Costs Using Survey Data on Sales
Agreements 89
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
2.2 Transaction costs: notion and difficulty of measurement . . . . . . . . 94
2.3 A new way to measure transaction costs . . . . . . . . . . . . . . . . 98
2.4 The size and correlates of transaction costs . . . . . . . . . . . . . . . 103
2.4.1 The size of transaction costs . . . . . . . . . . . . . . . . . . . 103
2.4.2 The correlates of transaction costs . . . . . . . . . . . . . . . . 105
2.5 The quality of our transaction costs measure . . . . . . . . . . . . . . 108
2.5.1 Potential biases at respondent level . . . . . . . . . . . . . . . 109
2.5.1.1 A potential upward bias at respondent level . . . . . 110
2.5.1.2 A potential downward bias at respondent level . . . . 118
2.5.2 Potential biases at aggregate level . . . . . . . . . . . . . . . . 119
2.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
3 An Analysis of the Factors Determining the Genesis and Size of Transaction
Costs 135
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
3.2 Determinants of transaction costs proposed by theory . . . . . . . . . 139
3.3 Other variables we construct . . . . . . . . . . . . . . . . . . . . . . . 146
3.4 Econometric details and results . . . . . . . . . . . . . . . . . . . . . 150
3.4.1 Tobit results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
3.4.2 Sample selection results . . . . . . . . . . . . . . . . . . . . . 152
3.4.2.1 Maximum likelihood estimation . . . . . . . . . . . . 154
3.4.2.2 Two-step estimation . . . . . . . . . . . . . . . . . . 157
3.4.2.3 A new specification . . . . . . . . . . . . . . . . . . . 158
3.4.3 Distinct ordinarily least squares and probit results . . . . . . . 159
3.4.4 Additional results . . . . . . . . . . . . . . . . . . . . . . . . . 162
3.4.4.1 The two-sided nature of the relationship-specific in-
vestment . . . . . . . . . . . . . . . . . . . . . . . . 162
3.4.4.2 The potentially endogenous nature of some variables 164
3.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
Appendices to Chapter 1 177
A Further details on the survey . . . . . . . . . . . . . . . . . . . . . . . 177
B Further details on the variables . . . . . . . . . . . . . . . . . . . . . . 179
C A different use of the dependence variables . . . . . . . . . . . . . . . . 189
D Other results using the overall contract complexity score . . . . . . . . 197
Appendix to Chapter 3 201
E Further details on the variables . . . . . . . . . . . . . . . . . . . . . . 201
v
LIST OF TABLES
1.1 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
1.2 A hint on the existence of hold-up (I) . . . . . . . . . . . . . . . . . . 78
1.3 A hint on the existence of hold-up (II) . . . . . . . . . . . . . . . . . 78
1.4 OLS estimates of the determinants of contract complexity . . . . . . 79
1.5 ML TE estimates of the impact of seller specific investment . . . . . . 80
1.6 ML TE estimates of the impact of buyer specific investment . . . . . 81
1.7 IV estimates of the impact of buyer and seller specific investments . . 82
1.8 Comparing the estimates obtained by different methods . . . . . . . . 82
1.9 ML TE estimates with identification based only on non-linearities (I) 83
1.10 ML TE estimates with identification based only on non-linearities (II) 84
1.11 Heckman two-step estimates of the impact of specific investments . . 85
1.12 Robustness: Adding variables to the equation explaining seller contract
complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
1.13 Robustness: Adding variables to the equation explaining buyer con-
tract complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
1.14 ML TE estimates when the sample is split by significant prepayments 88
2.1 The size of transaction costs . . . . . . . . . . . . . . . . . . . . . . . 131
2.2 The correlates of transaction costs . . . . . . . . . . . . . . . . . . . . 132
2.3 Transaction costs information based on renegotiation requests and suc-
cess of agreements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
2.4 Investigation of biases at aggregate level . . . . . . . . . . . . . . . . 134
3.1 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
3.2 Censored regression estimates of two different specifications . . . . . . 171
vi
3.3 Sample selection estimates of the complete specification . . . . . . . . 172
3.4 Sample selection estimates of the reduced specification . . . . . . . . 173
3.5 OLS and probit estimates . . . . . . . . . . . . . . . . . . . . . . . . 174
3.6 Distinction between the two specific investments (I) . . . . . . . . . . 175
3.7 Distinction between the two specific investments (II) . . . . . . . . . 176
1.15 OLS estimates when dependence on partner is not used . . . . . . . . 190
1.16 Seller ML TE estimates when dependence on partner is not used . . . 191
1.17 Buyer ML TE estimates when dependence on partner is not used . . 192
1.18 Seller ML TE estimates when dependence on partner is used in both
equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
1.19 Buyer ML TE estimates when dependence on partner is used in both
equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
1.20 Seller ML TE estimates when dependence on partner is used only as
an instrument . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
1.21 Buyer ML TE estimates when dependence on partner is used only as
an instrument . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
1.22 OLS estimates of the determinants of overall contract complexity . . 197
1.23 ML TE estimates of the impact of seller specific investment on overall
contract complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
1.24 ML TE estimates of the impact of buyer specific investment on overall
contract complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
1.25 Comparing the estimates obtained by different methods . . . . . . . . 200
vii
LIST OF FIGURES
1.1 The distributions of the contract complexity scores . . . . . . . . . . 76
2.1 The distributions of the transaction costs responses . . . . . . . . . . 130
3.1 Distribution of answers to the transaction costs question . . . . . . . 169
viii
LIST OF ABBREVIATIONS
AQR Appropriable quasi-rents
IV Instrumental variables
LR Likelihood ratio
ML Maximum likelihood
OLS Ordinary least squares
PRT Property rights theory
TCE Transaction cost economics
TE Treatment-effects
VI Vertical integration
ix
An Overview of the Dissertation
We provide below an overview of this dissertation, interweaving the commonal-
ities with the particularities of the three essays it contains.
Part of the New Institutional Economics, Transaction Cost Economics consid-
ers transaction costs to arise from two inherent features of human nature (bounded
rationality and opportunistic behavior) and to take a wide array of forms. Follow-
ing Coase, firms are considered to conduct their activities and shape the governance
structure of their agreements in an attempt to save on transaction costs. The most
important common feature of the three essays which compose this dissertation is that
they all adhere to these ideas, each of them exploring particular topics within the
transaction cost area of research.
Another common element of the three essays in this dissertation is that they
all rely on data collected by a survey of Romanian companies. Implemented in the
middle of 2001, the survey targeted middle and large companies, and two hundred
fifty-four establishments were included in the sample. The goal of the survey was
to understand the way Romanian companies conduct business and, in particular, to
assess their reliance on the formal legal system. To achieve this goal, four different
respondents were interviewed in face-to-face meetings: the general manager, the legal
adviser, and the managers of the sales and procurement departments. Responses of
1
the latter two managers provide most of the information we use in the three essays.
The main feature of the survey, with direct implications on the types of analyses that
could be conducted, is the richness of information it provides. We will elaborate on
this below and in a later section of the dissertation.
The first chapter investigates the use of contracts, the complexity of which is
seen as a decision firms make. This decision is considered to be analogous to that
on vertical integration, hence the analysis is directly relevant to the current debates
on theories of the firm. Transaction cost economics theory is commonly viewed to
imply that the presence of specific assets in a transaction governed by an incomplete
contract leads to the existence of appropriable quasi-rents, the distribution of which
creates tensions between the trading partners.
To limit the scope of opportunism, firms design more integrated governance
structures for their agreements, hence more complex contracts are expected when
asset specificity is present. We investigate relationship-specific investments made
by both partners in an agreement, and a simple model based on the rent-seeking
branch of the transaction cost economics generates the hypothesis that buyer and
seller relationship-specific investments have opposite effects on the exposure to op-
portunism, hence on the complexity of the contract.
The precise direction of the effect is thought to depend on partners? relative
exposure to the opportunism of the other. Our results indicate that, at least in
Romania, sellers are vulnerable to buyers? opportunism even without undertaking
dedicated investments in physical assets.
2
The first essay also analyzes companies? reliance on the formal legal system in
a transition economy and investigates the effects the prepayments buyers make have
on the contract complexity.
The second chapter constitutes a methodological addition to the transaction cost
literature. Despite the importance of the transaction cost concept, the multitude of
forms this cost takes, combined with its hidden nature, led to the existence of very
few studies which analyze the size of transaction costs, in particular at the agreement
level. We aim at filling this gap by proposing a simple methodology for assessing the
size of transaction costs that firms incur in exchanges.
To achieve our goal we use reports from business officials who supervise com-
panies? buying and selling activities. The transaction cost estimates collected in the
survey correlate significantly with variables suggested by theory, indicating validity of
the method. In Romania, those transaction costs directly related to the firms? buying
and selling activities are assessed as large, accounting for more than a fifth of the
value added.
The third chapter of this dissertation is very much related to the second one
given that it uses the transaction costs information collected as indicated in the second
chapter. The objective of the third essay is to investigate the impact of the factors
theory proposes on both the probability of existence and the size of transaction costs.
To accomplish this objective, the first step we take is to survey the determinants of
transaction costs ? this step confirms our earlier comments on the scarcity of existing
empirical work using transaction cost information at agreement level. Our results
3
indicate that theory is very successful in predicting the existence of transaction costs,
and moderately so in predicting their size when such costs are incurred.
The multitude of questions contained by the survey instruments provides infor-
mation on various transaction and firm features, allowing for a deep analysis of the
theoretical predictions. For instance, researchers in the transaction cost area have fre-
quently encountered difficulties in measuring asset specificity directly, and so proxies
are often used instead. The choice of proxies is always open to debate. Our measure
of relationship-specific investment, despite being a binary variable, captures the pre-
cise information of interest and is consistently measured across firms and industries.
Also, the survey provides variables which can be used as instruments for the likely
endogenous decision to invest in specific assets.
One of the concerns economists raised with the existing empirical studies in
the transaction cost area is related to the particularities of the data they rely on.
Namely, given companies? reluctance to publicize their contracts, many studies have
focused on agreements in which the state is involved or have concentrated on agree-
ments from highly regulated areas of activity. Doubts arise given peculiarities of
these agreements, especially in what concerns partners? tendency to behave oppor-
tunistically. In response to this concern, our survey has targeted firms with various
ownership structure, which conduct business in various productive sectors of activity
of the Romanian economy.
An extremely useful feature of the questionnaires addressed to these sales and
procurement managers is a two-fold symmetry. We have called symmetry between
4
the two survey instruments the fact that they contain almost identical questions,
once addressed to the person most knowledgeable on firm?s sales, and once to the
person most knowledgeable on firm?s acquisitions. In addition, apart from being asked
about their own firm?s activity in a particular agreement, the sales and procurement
managers are also asked about their partner?s activity in the same transaction. We
have called this feature symmetry within the two questionnaires.
The two-fold symmetry has an important implication on the sample size: we
collect information about two agreements each company has entered into, hence the
dataset the three essays rely on consists of approximately five hundred observations
(agreements). This sample is larger than those used elsewhere, thus answering an-
other concern in the literature. For the first chapter in particular, the survey?s two-
fold symmetric feature is crucial. We could not have tested the predictions of the
theoretical model if we would not have collected information on both the buyer and
the seller in each agreement.
Lastly, this dissertation (in particular the first essay) constitutes advances in the
field also in terms of the precision of the empirical methods. We will briefly elaborate
on this below.
Unlike many other studies, the first chapter takes into account the endogenous
nature of the decision to invest in relationship-specific assets. Moreover, this is the
first transaction cost economics analysis which explains the patterns in the biases of
ordinary least squares estimates (given that a particular scenario is adopted).
The second essay analyzes the quality of the transaction cost information col-
5
lected by the proposed survey question. Various biases are investigated: potential
(upward and downward) biases at respondent level, which may arise from a different
understanding of the question, and potential biases at aggregate level, which may
primarily be the results of a sample selection process.
The third chapter of this dissertation uses the transaction cost information
gathered by the second one. Given the limited nature of the data, we first employ
Tobit, the model most commonly used in such situations. The sample selection
model is found to be more appropriate. However, combined evidence indicate that
sample selection bias would not affect the estimates obtained by distinct regressions:
an ordinarily least squares estimation using the positive transaction cost responses,
and a probit estimation using a transaction cost dummy variable. The potential
endogeneity of various factors is also discussed.
6
Chapter 1
The Contingent Effect of Partners? Relationship-Specific Investments:
Evidence from Romania on the Use of Complex Contracts
Abstract
The complexity of the contract governing an exchange is considered to be deter-
mined by transacting partners based on their exposure to the opportunistic behav-
ior of the other. Complex contracts are believed to play an analogous role to that
of vertical integration, making the present analysis directly relevant to the current
debates on theories of the firm. Investments in relationship-specific assets have a
direct impact on partners? vulnerability to opportunism. We investigate investments
in dedicated assets made by both partners in an agreement, and a transaction cost
economics model generates the hypothesis that buyer and seller relationship-specific
investments have opposite effects on the exposure to opportunistic behavior, hence
on the complexity of the contract. The precise direction of the effect is thought to
depend on partners? relative vulnerability to the opportunism of the other. Biases in
ordinary least squares estimates are investigated: unobserved heterogeneity is shown
to bias downward the difference between the effects of buyer and seller specific invest-
ment, thus making it difficult to find supporting evidence for the above hypothesis.
We use agreement-level data collected from a cross-section of Romanian firms, and
7
employ the treatment-effects model estimated by maximum likelihood. The results
indicate that sellers? relationship-specific investments increase contract complexity,
while buyers? investments reduce it, supporting our view that, even in the absence
of a concrete investment sellers are exposed to buyers? opportunistic behavior. Ordi-
nary least squares estimates do not generate this result. This is the first transaction
cost economics analysis that simultaneously counters the problems of unobserved het-
erogeneity, generates estimates of the effects of relationship-specific investment that
are opposite in sign on opposite sides of the agreement, and explains the patterns
in the biases of ordinary least squares estimates. Other significant results consider
the quality of court services and the impact of buyers? prepayments. The limited
regional variation in court quality is shown to affect the complexity of contracts,
suggesting even moderate amounts of legal reform can have appreciable effects in
Romania. Prepayments buyers make are shown to provide credible signals of com-
mitment, balancing partners? vulnerability to opportunism, with direct implications
on the complexity of contracts.
1.1 Introduction
This dissertation chapter examines microeconomic evidence and analyzes the
way in which companies in Romania (a typical transition, developing country) orga-
nize transactions between them.
Despite the numerous papers examining the determinants of the structure of
transactions, the continuing interest in this topic suggests that issues are far from
8
settled (Whinston, 2003; Garrouste and Saussier, 2005; Gibbons, 2005). The most
frequently discussed form of governance in the literature is vertical integration (VI),
while one of its alternatives, contracting, is much less analyzed. Even when contract-
ing is discussed, most papers deal with the length of contracts (expressed in time
units or number of discrete transactions), while very few authors have investigated
the complexity of contractual arrangements.
We will try to fill the gap between the work on VI and contracting by ana-
lyzing the latter. An important feature of this chapter is that we consider contract
complexity to be chosen by firms in accordance to their exposure to the opportunis-
tic behavior of their trading partners. We view the decision on contract complexity
to be analogous to the make-or-buy decision, meaning that our results are probably
relevant to debates on the determinants of VI.
Recent theoretical discussions of VI have focused on why the effect of relationship-
specific investment might vary (Whinston, 2003; Acemoglu et al., 2005; Gibbons,
2005). In particular, under the assumptions of the property-rights theory (PRT), the
specific investments made by the two partners in an exchange have opposite effects
on the likelihood to integrate. In contrast, transaction cost economics (TCE) has
usually been interpreted to indicate that relationship-specific investments made by
any party in a transaction increase VI chances. The comparison between PRT and
TCE is a good example of the existence of competing theories, which employ dif-
ferent methodologies and different conceptions of the core ingredients of a theory of
transactions (Whinston, 2003).
9
In this chapter we use a transparent model based on the rent-seeking branch of
TCE (Gibbons, 2005), and show that specific investments can have either positive or
negative effects on contract complexity. The variation in the sign of the effect appears
strongly in our empirical results: relationship-specific investments sometimes induce
the use of more costly legal mechanisms and sometimes reduce their use, depending
on whether the buyer or the seller is undertaking the investment. This finding is
inconsistent with the almost universal assumption in the literature that the chief
feature distinguishing TCE from PRT is variation in the sign of the effect of specific
investment (Whinston, 2003, p. 2; Acemoglu et al., 2005; Gibbons, 2005, p. 204)1.
Despite the impressive number of papers examining the determinants of transac-
tional characteristics, mostly from the TCE perspective, doubts about methodologies
remain (Chiappori and Salani?e, 2002, p. 27?28; Masten and Saussier, 2002, p. 13).
Many methodological problems stem from the difficulty to obtain data on transac-
tions and their determinants, which limits the econometric methods that can be used
and the precision with which theoretical constructs can be rendered empirically.
In this chapter we make progress on this front by using detailed survey data on
transactions conducted by a broad cross-section of Romanian firms. Also, we show
that it is crucial to address standard methodological problems in empirical work when
investigating whether there is a conditional effect of relationship-specific investment
on contract complexity. Collecting very detailed data on individual transactions is
1Whinston (2003, p. 21) remarks on the possibility of a negative effect of specific investment,
but he regards this as inconsistent with the TCE. Within the business economics literature, there
has been research on that negative effect in the context of the TCE.
10
a key step in being able to address these problems. Our data set has information
on both partners to a transaction, allowing investigation of the asymmetry between
seller and buyer effects.
A simple econometric framework, which nevertheless incorporates the insights
of the TCE theory, indicates why unobserved heterogeneity leads to biases in the
ordinary least squares (OLS) estimates, which blur the differences between estimated
effects of buyer and seller specific investment. This analysis is substantiated by the
Romanian data: OLS estimates of the effects of buyer and seller relationship-specific
investment are similar, while consistently estimated effects are opposite in sign. Bene-
fiting of detailed data, we are able to address the problem of unobserved heterogeneity.
Once this is addressed, the conditional effect of specific investment appears.
Another feature of this chapter is that is discusses companies? reliance on the
legal system in a developing country ? designing more complex contracts implies the
use of the formal legal system. We believe it is fruitful to discuss such reliance in the
context of Romania, despite the prevailing view of the courts as being ineffective in
developing countries. The essays in Murrell (2001) suggest that this view is misplaced,
while Johnson et al. (2000) provide evidence that Romanian companies use the formal
legal system more than other countries in the region.
This dissertation chapter shows that the use of the law does respond to varia-
tions in the nature of the transactions, with firms investing more in legal arrangements
when they are more vulnerable to the opportunism of their partners. Moreover, firms
use a more law-intensive form of transaction when courts offer services of a higher
11
quality. Since only small variations in court quality are possible in the centralized
Romanian legal system, this is evidence that even minor court reforms could have
significant effects on the use of law.
This chapter is organized as follows. Section 1.2 provides an exposition of the
core notion of opportunistic behavior and of the solutions firms adopt to counteract it.
We focus on contracting, regarded as an alternative to VI. Section 1.3 briefly presents
related existing studies and points out some concerns raised elsewhere. These two
sections do not present a comprehensive discussion of the theory or the empirical
findings. For this, the following survey studies are exhaustive: Shelanski and Klein
(1995), Masten (1996), Chiappori and Salani?e (2002), Masten and Saussier (2002),
Sykuta (2005), Macher and Richman (2006).
Section 1.4 provides the theoretical support of our analysis, including a simple
TCE model which shows how the conditional effect of relationship-specific investment
arises from standard assumptions of the theory. Section 1.5 presents the econometric
details of our analysis, exploring the consequences of unobserved heterogeneity for
estimates of the effect of specific investment on contract complexity. We show that
OLS estimates tend to blur the difference between estimated effects on different sides
of the transaction. Section 1.6 sets up the empirics, describing the data and vari-
ables, while Section 1.7 hints on the existence of hold-up in the Romanian business
environment.
Section 1.8 presents the core estimates, focusing on maximum likelihood (ML)
estimates of the determinants of contract complexity. Robustness of the results is
12
examined, followed by a discussion on how the effects of relationship-specific invest-
ment vary with prepayment by the buyer. This constitutes supporting evidence for
our proposition that the critical factor in determining the sign of the effect of specific
investment is which partner is naturally vulnerable to opportunistic behavior. Section
1.9 concludes this chapter.
1.2 An overview of the opportunistic behavior
It is commonly accepted that the causes of transaction costs can be reduced to
two inherent features of the human nature: bounded rationality and opportunistic
behavior, and the focus of this dissertation chapter is on the latter. After presenting
various sources and forms of opportunistic behavior, we briefly go over some of the
solutions presented in the literature. Our focus will be on contracting, which can be
regarded as an alternative to the often discussed VI.
1.2.1 Opportunistic behavior: notion and forms
Opportunistic behavior generally refers to any misbehavior that appears in the
course of a transaction. It is considered to be particularly acute in the presence of
asset specificity, which leads to the existence of appropriable quasi-rents (AQR)2.
Masten (1996) discusses two sources of opportunism: inevitable gaps and ambiguities
in contracts, and imperfect and costly nature of the judicial enforcement. Masten
2In their seminal 1978 article, Klein, Crawford, and Alchian provide a discussion of AQRs and
consider that ?after a specific investment is made and such quasi-rents are created, the possibility
of opportunistic behavior is very real? (Klein et al., 1978, p. 298)
13
(1996, p. 6?7) also presents two distinct forms the opportunistic behavior can take.
The first form is represented by deviations from joint-surplus maximizing behavior,
when the breaching party would like his actions to be undetected by his partner and
courts. Examples include shirking, cutting corners, degrading quality in undetectable
fashion or in dimensions that have not been specified in contracts. The second form of
opportunism is represented by those efforts to hold-up a trading partner, when actions
are taken to set new terms rather than in reaction to existing ones, and the goal is
reaching renegotiation of the contract. During contract execution, such behavior
includes suing for trivial deviations and making false claims of dissatisfaction.
Rogerson (1992) defines hold-up more precisely. According to him, hold-up is
the ex post request for renegotiation after a relationship-specific investment has been
made in a transaction governed by an incomplete contract. A common element in the
definitions of opportunistic behavior and its particular form of hold-up is therefore the
presence of asset specificity3, which usually takes the particular shape of relationship-
specific investment. However, despite being used interchangeably sometimes, the
terms ?hold-up? and ?opportunistic behavior? bear distinctions, as hold-up is just an
element in the set of opportunistic measures a partner in a transaction may adopt.
1.2.2 Solutions to opportunistic behavior
Following Coase, firms are considered to adopt different governance structures
in an effort to reduce transaction costs. As these costs are triggered by the exposure
3Williamson (1983) identifies four different types of asset specificity: dedicated assets and site,
physical asset, and human-capital specificity. Masten (1996) adds a fifth type ? temporal specificity.
14
to partner?s opportunism, firms will try to employ the least costly deterrents of such
behavior (Klein et al., 1978; Gibbons, 2005).
Integration (the situation when trading partners act under unified ownership
and control and their actions are guided by a single objective: profit maximization)
is the most prominent solution found in the economics literature. Apart from eco-
nomics, the business area of research is also abundant with articles investigating VI
from a TCE perspective. For example, Anderson and Schmittlein (1984) and John
and Weitz (1988) are two of the papers incorporating TCE concepts into marketing
and distribution strategy4. Given the extent to which VI has been debated in the
literature, it will not constitute the main interest of the present chapter.
VI is a form of governance placed at one end of a spectrum. At the other end are
spot market exchanges, when one-time agreements take place between independent
economic actors. In this case, all relevant information is contained in the price and
there is no deterrent of partners? opportunistic behavior, hence this governance mode
is used mainly for simple transactions. According to Shelanski and Klein (1995, p.
337), ?[b]etween the two poles of market and hierarchy are a variety of ?hybrid? modes,
such as complex contracts and partial ownership arrangements.? This view is shared
by others, such as Williamson (1991a), Hubbard (2001), and Sykuta (2005).
A distinct body of research is therefore dedicated to intermediary forms of gov-
ernance (placed on the spectrum between VI and spot market transactions) which
4Despite providing a critique of TCE, Ghoshal and Moran (1996) list other strategic and orga-
nizational areas of great importance to managers where TCE principles have been implemented:
international expansion, strategic alliances, optimum financial structure, and design of internal in-
centive systems.
15
are employed in an attempt to limit the scope of partners? opportunism. Examples
include long-term contracts5, complex contracts, quasi-vertical integration6, option
contracts7, hybrids8, and informal agreements9. Since our focus is on complex con-
tracts, in what follows we will examine in detail this possible deterrent of opportunism.
1.2.3 Complex contracts as an alternative to vertical integration
Rogerson (1992) considers that if arbitrarily complex contracts (that will remove
the need for renegotiations) can be signed, problems caused by opportunistic behav-
ior in general and hold-up in particular would be resolved. Nevertheless, bounded
rationality, limited foresight, skill, knowledge and time, combined with increasing
costs of designing more complex contracts, imply that, regardless their complexity,
all complex contracts are unavoidably incomplete (Williamson, 2000, p. 599).
Since contracts usually imply written agreements, designed based on a clear set
of rules, an important characteristic of contracting is that it takes partners beyond
5The fact that long-term contracts provide a viable solution to the opportunistic behavior pf
the partners has been long recognized: Klein et al. (1978) view these contracts as the primary
alternative to integration.
6As Monteverde and Teece (1982) have defined it, quasi-vertical integration represents a situation
in which the downstream firm owns the specialized tools and the patterns used in the fabrication of
components, but parties use contracts to govern the production process.
7An option contract is one that gives sellers the right (but not the obligation) to deliver a fixed
quantity of the good and makes buyers? contractual payment contingent on the sellers? delivery
decision ? see Noldeke and Schmidt (1995) for details.
8Some examples from are franchise agreements, joint ventures, exclusive dealing contracts, the
prime-contractor/subcontractor organizational unit, reciprocity agreements, or equity linkages.
9Some informal agreements imply the use of outside connections (with the state, other companies,
or even criminal groups), while others imply an exchange of hostages. A hostage can be any asset that
has some value for partners, is exchanged to show commitment, thus leading to balanced exposure
to opportunism. Hostages may take various forms: Shelanski and Klein (1995) discuss reciprocal
buying arrangements, Williamson (1983) considers the use of security bonds buyers present to cover
part of relationship-specific investment producers undertake, while others have investigated offsetting
specific investments.
16
simple promises. Partners in a contract agree to follow certain procedures when
disputes arise, and to present their case before arbitrators (which usually take the
form of commercial courts). When entering a contract, partners are discouraged to
behave opportunistically by the penalties imposed by the legal system. Hence, we
see reliance on the formal legal system to be the main feature of both long-term and
complex contracting.
However, differences exist. Long-term contractual agreements also act as de-
terrents of opportunism by relying on economic incentives arising from repeated ex-
changes: the prospect of missing future gains from trade reduces one?s current propen-
sity to such behavior. Complex contracts however rely solely on the use of the formal
legal system. A direct implication of this slight difference between long-term and
complex contracts is that the latter are probably equally efficient at any stage of the
transaction, while the efficiency of the former is expected to decrease as transaction
progresses and approaches its closure.
Another important feature of contracting is that the costs of designing proper
contracts is assumed to increase with the level of appropriable rents. The main
implication for this chapter is that VI is more likely to be the governing structure
when large AQRs are at play, while complex contracts probably constitute a less
costly alternative for moderate levels of specialized quasi-rents. Complex contracts
may also be the preferred choice when financial resources are scarce so as to limit the
possibility of integration, or when integration is not possible for other reasons, such
as government regulation.
17
A main obstacle in organizing any transaction is the delicate balance between
the possibility of ex-post adjustments (necessary in any complicated interaction) and
the safeguards that encourage partners to undertake relationship-specific investments.
VI avoids this obstacle. We believe that complex contracts can also do so, as they
can be used to anticipate the types of adjustments that have to be made and how
they are to be accomplished.
This view is by no means novel. In the TCE area of research, it has been
expressed by others, such as Joskow (1985, p. 38?39), who notes that ?[c]onsiderable
emphasis has been placed on the proposition that vertical integration is more likely to
emerge when cost-minimizing (in the neoclassical sense) transactions involve durable
transaction-specific sunk investments. Williamson?s recent work and a great deal of
Klein?s work, however, consider contractual alternatives to vertical integration where
transactions costs are important. [...] The structure of these market contracts will
reflect efforts to create incentives and restrictions that reflect anticipated performance
problems so that agents will perform as initially promised when different contingencies
arise.? Crocker and Masten (1991) also share this view. In the business literature, the
existence of contractual alternatives to the make-or-buy decision has been underlined
by Adler et al. (1998) and others.
Hence, the parallels with VI become apparent as both complex contracts and in-
tegration can be viewed as ways to facilitate adjustment without consequent increases
in vulnerability. In addition, one could easily imagine detailed contracts mimicking
18
the set of arrangements ascribed to VI in many formal models10. To conclude, we
believe that VI and complex contracts can be seen as alternative measures against
opportunism, alternatives which bear many features in common.
In this context the following comment, made by Gibbons (2005, p. 234), seems
all too relevant: ?the adaptation theory applies not only to the make-or-buy problem,
but also (and at least as well) to the class of contracting problems where two firms
with fixed boundaries pass decision rights across their boundaries by contract. [...]
In short, the make-or-buy theories described here may have as much to say about
contracts between firms as they do about the boundaries of firms.?11
1.3 Some related existing studies
This dissertation chapter investigates the use of complex contracts as a solution
to the potential opportunistic behavior of partners. Before presenting our insights,
we consider useful to display some of the findings in the existing studies which have
also analyzed the complexity of contracts. In addition, the model we will present
in Subection 1.4.3 bears some features found elsewhere in the literature, resembling
work on dependence balancing between partners in a transaction. In what follows,
we will briefly go over some of these finding as well.
10In an early contribution, Kessler and Stern (1959) compared ?contract integration? to VI. Also,
in his analysis of long-term contracts, Joskow (1985, p. 50) considers that very long and tight
contracts can ?simulate the opportunism-mitigating features of vertical integration.?
11Gibbons (2005) also cites the work of Maskin and Tirole (1999), who show the formal equivalence
between the property-rights theory and certain contracting models.
19
1.3.1 Previous results on contract complexity
The option researchers frequently adopt is to construct a measure of contract
characteristics, denoted C below, representing contract length, complexity, choice
of pricing scheme, or such. This is usually regressed on a set of independent vari-
ables, of which the most important is considered to be specific investment, denoted I.
The other explanatory variables represent transaction and firm characteristics, and
encapsulated into the set Z1. The most common regression equation is therefore:
C = ?1 + ?1I + ?1Z1 + ?1, (1.1)
where the unit of observation is a particular transaction between two firms12.
One point worth noting is that C in this study plays the same role as the
decision to vertically integrate does in other studies. Another point is that, in the
context of this chapter, where C denotes the complexity of contracts, the meaning
of ?1 is straightforward: it captures the increase in contract complexity given that a
relationship-specific investment is made.
In the TCE literature the usual reasoning is that asset specificity raises the
threat of opportunistic behavior. In particular, specific investments undertaken in
incomplete contracts are considered to increase the likelihood of hold-up. To counter
such a behavior, partners rely on more integrated governance structures ? longer-
term or more complex contracts are signed, stricter pricing schemes are adopted,
12For presentation purposes we choose not to display a subscript for C, I, Z1, and ?1.
20
etc. Hence, the common expectation is that ?1 is positive. Joskow (1988, p. 105)
mentions that ?we expect the parties to more frequently choose vertical integration or
a long-term contract as the quasi-rents associated with specific investments become
more important.?, while Gibbons (2005, p. 204) states that the unifying theme in
empirical work following the rent-seeking version of transaction cost theory is that
larger AQRs make integration more likely, where relationship-specific investment is
the source of these rents.
The literature on contracting is rather young, especially when compared to
the vast VI literature (Chiappori and Salani?e, 2002). Moreover, within the area of
contracting, many authors have investigated the implications various factors have on
contract length, while contract complexity has been discussed by only a handful of
papers. Some of these papers will be briefly presented below.
Crocker and Reynolds (1993) analyze pricing procedures used in U.S. Air Force
engine procurement contracts. The authors expect that contracts will be more com-
plex when opportunistic behavior is more likely, and when the environment is less
intricate and uncertain. They use a panel dataset of forty-four contracts signed be-
tween 1970 and 1987, and run an ordered probit for the type of the contract (a
discrete value ranging from one to eight) on variables which proxy for uncertainty,
and on the reputation of the supplier to behave opportunistically. An exogeneity
test is employed, suggesting that estimated coefficients are not ridden by endogeneity
bias. The results support the theoretical prediction13.
13A possible drawback of this paper is that authors use pricing schemes as a measure of contract
complexity, while they are more an indicator of contract rigidity. For example, if the contract sets a
21
Gompers and Lerner (1996) use a sample of one hundred-forty venture partner-
ship agreements and consider the number of contractual covenant classes included in
these agreements (out of a total of fourteen) to be the dependent variable. Part of
authors? endeavor is to test whether more restrictive contracts (which include more
covenants) are employed when the potential for opportunistic behavior is greater.
The authors use variables suggested by the standard literature (they focus on high-
technology investments) and conclude that results support their hypothesis.
Saussier (2000) considers the complexity of contracts to be the result of parties?
willingness to save transaction costs, a view we also adopt in the present chapter.
However, unlike the contract complexity measure we use, Saussier constructs an dia-
metrically opposite one, called ?contract incompleteness?, and discusses the degree of
incompleteness as a result of transaction characteristics. The data the author relies on
represents the twenty-nine contracts signed between 1977 and 1997 by El?ectricit?e de
France for transportation of coal to its power plants. The author uses the discrete con-
tract incompleteness score (ranging from zero to six) to tests two basic propositions:
an increase in the level of asset specificity will lead to more complete contracts, while
an increase in the uncertainty level will lead to less complete contractual agreements.
Results, which take into consideration the possible endogeneity of asset specificity,
match author?s predictions.
The three studies mentioned above lead to two remarks. The first is that they
fixed price the authors consider this contract to be most restrictive (which it is) but also to be most
complete (which it is not necessarily the case). Another limitation is that asset specificity is not
directly observed and instead the authors use a proxy for seller?s tendency to behave opportunistically
? number of disputes a seller was involved into in the previous 5 years. As always, the choice of
proxies is open to debate.
22
all follow the TCE approach, thus predicting a positive ?1. The second relates to
the parallel between VI and complex contracting, which appears strongly in these
empirical studies; the independent variables used to explain the make-or-buy decision
are those employed as determinants of the choice between simpler and more elaborated
contract forms.
To conclude, the main link, following the TCE theory, is between the AQRs at
play, the partners? opportunistic behavior, and the chosen governance structure of the
transaction. As it appears from both theory and empirical applications, the existence
of AQRs has become almost synonymous with the amount of specific investment. The
relation mostly investigated in TCE is between relationship-specific investment and
the decision to integrate (Whinston, 2003), hence, in focusing on the link between
the degree of contract complexity and the presence of relationship-specific investment,
this chapter?s results are directly relevant to the voluminous literature on VI.
1.3.2 Previous results on dependence balancing
At this stage, we believe it is worth mentioning a particular area of transac-
tion costs research focusing on the effects of partners? bilateral exposure (especially
through relationship-specific investments), which follows Oliver Williamson?s insights
on hostages.
In his 1983 paper modeling the use of hostages, Williamson discusses reciprocity,
which transforms a unilateral supply relation into a bilateral one. He notes that
?reciprocity can serve to equalize the exposure of the parties, thereby reducing the
23
incentive of the buyer to defect from the exchange? Williamson (1983, p. 531).
Williamson believes that when accepting reciprocal exposure of specialized assets,
the buyer shows commitment to exchange and so defection hazards are mitigated.
The argument is therefore that specific investments might have advantages purely for
transactional reasons.
Several empirical papers followed this line of reasoning: Anderson and Weitz
(1992) use questionnaires addressed to managers and show that relationship-specific
investments are positively related to ?commitment? between manufacturers and dis-
tributors, where commitment is probably inversely related to the formal rigor of
contracts. Fein and Anderson (1997) show that the specific investments of manufac-
turers and distributors affect the degree to which each is willing to become tied into
a relationship with the other, suggesting that there is reciprocity in the willingness to
make specific commitments. Joshi and Stump (1999) suggest that the presence of two-
sided relationship-specific investments might diminish the tendency for the creation
of formal cooperative agreements between manufacturers and their suppliers, while
obtaining the standard result that specific investments lead to more formal arrange-
ments. Koss (1999) shows that buyers of fresh fish in British Columbia are more likely
to invest in seller?s boats if the seller has undertaken relationship-specific investments,
a standard result if one views the buyer?s investment as a partial VI. Ahmadjian and
Oxley (2006) examine how the equity stakes taken by Japanese auto assemblers in
their suppliers vary with the dependence of the suppliers on the assemblers.
Regarding the particular aspect we are interested in, none of the empirical stud-
24
ies above focuses directly on the issue of whether it is possible that increases in specific
investments actually reduce the likelihood of VI or the complexity of contracts. Even
though the discussions of theory in these studies would certainly be consistent with
a negative ?1 (the model we present in the next section also shows this possibility),
none of the corresponding empirical results is clearly inconsistent with a positive ?1,
that is, with the basic hypothesis that relationship-specific investments always induce
more complex contracts or VI.
1.3.3 Concerns with the existing studies
To conclude this section, we present below some concerns with the existing TCE
empirical studies, which are mainly related to data availability and measurement of
concepts of interest. Additional concerns, regarding econometric details, are presented
in Section 1.5.
Since contracts among firms are often confidential, companies are generally
reluctant to allow them to be studied, hence the areas of research are limited to
those where contracts are available. As a consequence, researchers have frequently
investigated contracts between state and private entities (as governments publicize
their contracts), or contracts between private entities, which are public because the
sector of activity is strictly regulated. Chiappori and Salani?e (2002, p. 34) note
that ?many papers in this field use similar data and/or focus on similar problems,
as shown by the number of papers on sharecropping or natural gas we surveyed. We
would certainly want to see wider-ranging empirical work in the future.? Despite
25
the meaningful insights such contracts uncover, most of the agreements are reached
between private companies acting in diverse, unregulated sectors.
Focusing on contracts involving the state (either directly or through strict reg-
ulation of a sector) may lead to several drawbacks. First, it is plausible that state
interference has particular implications on partners? incentives and behavior, espe-
cially regarding opportunism. Second, the low number of economic agents usually
acting in regulated sectors (e.g. coal transportation, gas extraction) may also af-
fect the propensity of opportunistic behavior. Chiappori and Salani?e (2002, p. 27)
also note that ?[m]ost of the empirical tests of transaction costs theory have been
implemented on data from relatively thin markets, where quasi-rents are large. An
interesting question is whether these intuitions extend to thicker markets.? Lastly,
the TCE papers tend to use samples of reduced size, with direct implications on the
quality of the estimates. Sykuta (2005, p. 20) comments that ?[g]iven the costs
associated with these data collection efforts and the proprietary nature of contract
documents, sample sizes in this field tend to be small ? so small as to preclude effective
use of most of the econometric techniques described above.?
Another concern comes from the measurement of the primary concepts. As fre-
quently argued, opportunistic behavior is driven by the presence of asset specificity,
but researchers have encountered difficulties in measuring specificity directly, and so
proxies are often used instead. The choice of proxies is always open to debate. Re-
garding asset specificity, Shelanski and Klein (1995, p. 340) comment that ?[f]urther
refinement and analysis need to be done [...] particularly in the area of measure-
26
ment. Proxies such as capital intensity of fixed costs are very imperfect.? Masten and
Saussier (2002, p. 13) note that ?[a] variety of issues should temper our confidence
in the findings to date. Probably chief among those is the quality of proxies used for
the explanatory variables identified by the theory. Often, these proxies are crude and
imprecise stand-ins for the variables of true interest.? Macher and Richman (2006)
express a similar concern.
Sykuta (2005, p. 26) raises an additional question regarding the measure-
ment of asset specificity. According to him, the theory predicts that investments
in relationship-specific assets give rise to potential quasi-rents and associated be-
havioral ills, while it is less clear whether, or when, assets that are firm-specific or
industry-specific are necessarily relationship-specific14.
Since the survey and variables will be presented in a later section, we only men-
tion here that our data address many of the limitations found elsewhere. In particular,
the survey we use targeted Romanian companies which undertook activities in various
productive sectors of activity, and our sample consists of four hundred twenty-three
agreements signed between firms with different ownership structures. In addition,
our measure of the relationship-specific investment (despite being a binary variable)
captures the precise information of interest and is consistently measured across firms
and industries.
14For instance, Acemoglu et al. (2005) use a firm-specific measure such as technology intensity.
27
1.4 Theoretical support of our analysis
When examining VI, those who contrast TCE and PRT focus on the sign of
?1. PRT implies that the sign of ?1 depends on transaction circumstances and which
party is undertaking the relationship-specific investment. Namely, if ?1 is positive for
specific investment undertaken by one party to the transaction, it will be negative for
the other side. In contrast, the existing TCE theory predicts that any investment in
relationship-specific assets increases the likelihood of integration (Whinston, 2003 p.
2, 21; Acemoglu et al., 2005, p. 1)15.
Even though the focus of this dissertation chapter is on TCE issues, and not on
PRT concepts, the distinction between the two views is of great interest to us ? this
chapter may help reconcile TCE and PRT in what concerns the expected sign of ?1.
1.4.1 A description of the buyer-seller relationship
In what follows we will try to briefly present our view on the relationship between
the buyer and the seller in a simple transaction, including the consequences this may
have on the complexity of contracts governing these transactions.
Some simplifying assumptions will be made. The first is that one of the two
parties is exposed to the opportunism of the other. The main cause of exposure
is considered to be the presence of asset specificity, which is mainly triggered by
partners? decisions to undertake investments in dedicated physical assets. However,
15Gibbons (2005, p. 226) cites Williamson (1979) and Masten (1986) and claims that transaction
costs of non-integration increase with asset specificity. Whinston (2003, p. 21) states the prevailing
consensus but also notes counter-examples such as Koss and Eaton (1997).
28
asset specificity can have other sources, such as the precise role each party has in the
agreement, combined with macroeconomic conditions. In particular, we believe that
at least in Romania, even in the absence of a concrete investment in a dedicated asset,
the seller is usually the vulnerable party in an exchange (the following subsection will
provide the arguments for this second assumption).
The third assumption we make is that the exposed party in a transaction has
more to say about contract terms than the other party does. Let us consider an overly
simplified contracting process, in which only two types of clauses can be discussed:
technical and safeguarding. In a given agreement it is plausible that parties have
equal bargaining power in negotiations over the technical aspects of the transactions.
However, the prospects of opportunism may make the vulnerable partner reluctant to
enter an agreement which does not stipulate satisfying safety terms, thus this party
may have greater bargaining power in negotiations over the safeguarding clauses.
The possibility of opposing impacts of buyer?s and seller?s specific investments
on contract complexity appears naturally. When the seller undertakes a relationship-
specific investment this increases his initial exposure to buyer?s opportunism, and he
will probably argue for additional safeguarding clauses to be included, thus leading to
a more complex contract. A relationship-specific investment by the buyer however is
likely to lead to a balanced exposure of the parties. In this case the net exposure (of
the buyer or seller) is lower than seller?s initial vulnerability, unless buyer?s exposure
from specific investment is tremendous (more than double compared to seller?s initial
vulnerability). Hence, the complexity of the contract governing such a transaction is
29
likely to be lower. In addition, the exchange would now be governed by a different
type of agreement: a transaction which has both parties exposed to the opportunistic
behavior of the other has self-enforcing features. For such transactions extended
written clauses are no longer needed.
1.4.2 Support for the vulnerability assumption
The second assumption we have made above is crucial for our argument. We
believe that in the simplest situation the partners can face, with no contract governing
the exchange, no prepayment offered by buyer, and no investment in a dedicated
asset, asset specificity can still be present. In particular, at least in Romania, it is
likely that the seller is the vulnerable party even without undertaking a relationship-
specific investment. Since our predictions depend on the validity of this assumption
we present below evidence supporting the background vulnerability of the seller.
For a typical industrial sector, where market power is the same on both sides
of a transaction, this assumption is standard. It is common practice in Romania
and elsewhere for buyers to pay on or after delivery, which means that sellers have
to commit financial resources to contract fulfillment before buyers do (McMillan and
Woodruff, 2002, p. 163).
In addition, as in most other transition countries, external financing in Romania
is hard to obtain. A recent study (National Agency for Small and Medium Sized
Enterprises and Cooperatives, 2004) shows that bank loans were inaccessible for small
enterprises in 2004, 60% of them being started with owners? private savings. The
30
study also indicates that 87.5% of small companies use their own resources to make
investments. A similar argument is provided by Rizov (2004, p. 725), who uses a
panel of medium and large Romanian manufacturing firms for the 1995 ? 1999 period.
The survey provinding our data was conducted in the middle of 2001, and was focused
on medium and large establishments. Therefore, we infer that at the time of our data
collection the business environment was harsh in what concerns external financing.
As McMillan and Woodruff (2002) document, transition is a time when sellers
are scrambling for financial resources to invest in inputs and production. The scarcity
of external financing increases the pressure on sellers? shoulders: they have to find
other sources of finance since buyers are not cooperating in this process, especially
when there are no contractual guarantees.
Our data show that over half of the buyers do not pay the seller anything before
delivery. It also appears that when buyers pay something before delivery, they do
this reticently ? only 13% of buyers pay at least half of total bill before getting the
ordered goods, while a mere 3.5% pay the bill in full before delivery. Hence, most
of the sellers have to rely exclusively on internal financial resources to produce the
goods, meaning that they commit resources before the buyers do and hope to recover
them following the successful completion of transactions.
The macroeconomic environment also has major implications on sellers? vul-
nerability. Despite being in transition for ten years at the time data were collected,
Romania was still confronting high levels of inflation: 45.7% in 2000 and 34.5% in
2001. Thus, any delay of payment for the goods already produced is equivalent to an
31
interest-free loan for the buyer, which must impose real losses for those sellers who
have not protected themselved through tight contractual provisions.
Our data indicate that, in addition to receiving most of the payments after
delivery, 22% of sellers have to wait more than a month after provision of goods for
the buyer to pay its obligations. Therefore, for an important share of sellers, it takes
a long time from making the good until receiving the payment. Even though we do
not have clear information whether some of these waiting periods are actually agreed
upon, it must be that some customers delay their sellers.
The data also indicate that there are more buyers who are late with payments
than sellers who are late with delivery. Precisely, the prepayments buyers make are
lower than initially agreed in 16.78% of the investigated agreements16. Since a seller
faced with a delay in prepayment is likely to react by postponing the delivery of the
product, one would expect the percentage of sellers delaying their customers to be
at least as high. However, in only 12.53% of transactions the delivery of the product
took more than parties agreed initially.
Lastly, the inter-enterprise debt problem, a symptom of this interaction between
the state of the macroeconomy and the relative vulnerability of seller, was one of the
most discussed phenomena in the first decade of transition. In Romania at the time
of the survey, firms did report the presence of large inter-enterprise debts. The con-
nection between tight credit, inflation, seller relative vulnerability, and opportunistic
behavior is also documented in a Slovakian case study by Gow et al. (2000).
16Our survey instruments only permit the investigation of the delays in payments made before
delivery of the product, not of the delays in other payments buyers make.
32
In the above paragraphs we have tried to argue that the scarcity of external
financing, combined with macroeconomic instability, induces many buyers to refrain
from offering significant prepayments, which makes sellers advance (own) resources
needed in the production process, thus leading to vulnerability. Then, the question
which arises is why do suppliers accept such an exposure? The answer may be that
there is a differential access to finance for buyers and sellers.
This view is supported by at least two factors. First, following the production
process, suppliers own a good which is still valuable in itself since it can be sold for a
fraction of the agreement value (or can be used as a collateral). In contrast, even if
they decide to finance part of the production process by offering prepayments, buyers
do not hold anything in their possession if the transaction fails. Second, one can see
suppliers as being partially financed by their own workforce since it is common prac-
tice for employees (managers, in particular) to be paid later, following the successful
completion of an agreement. This means that of the two parties suppliers are the
ones most likely to be able to advance the resources for the production of the good
(with direct implications on vulnerability).
1.4.3 A simple theoretical model
Based on TCE principles and bearing many common aspects to that presented
by Koss and Eaton (1997), the model below hints on the possibility of a negative ?1.
A transaction occurs between a seller (S) and a buyer (B) of an intermediate
good. The production process requires both parties to make investments, which we
33
denote by KS and KB. Once these investments are sunk their values in the next-best
use are denoted by OS and OB, with OS ? KS and OB ? KB. Thus, Ki ?Oi, where
i = S, B, is a measure of the specificity of the investments made by the two parties
in the transaction.
Similarly as before, we consider the seller to be exposed to buyer?s opportunistic
behavior even without undertaking an investment in a specific asset. This background
vulnerability is modeled as a cost V the seller incurs for engaging in the exchange.
After purchasing the good, the buyer sells it to an end-user, obtaining a gross
return of R. If KS + KB + V < R, then on efficiency grounds the good should be
produced (the transaction should take place).
We consider throughout that Ki, Oi, and V are common knowledge between
the two parties. An important assumption we make is that one party is exposed to
the opportunistic behavior of the other, hence we rule out the case KS ?OS + V =
KB?OB. In particular, we assume that it is the seller who is more exposed to hold-up
than is the buyer, hence that KS ?OS +V > KB ?OB.
In the initial stages of the exchange partners take the investment decision and,
at the same time, they decide on the governance structure of the agreement. In
particular, we assume there are two types of contracts parties may adopt. One (?no
contract?) costs nothing to implement, but cannot be enforced by the courts, leaving
one party possibly vulnerable to the opportunistic behavior of the other. Another,
costing C, is an ex ante perfectly enforceable contract17.
17C can be viewed as a proxy for the dependent variable in (1.1).
34
As it will appear below, given the assumptions we have made so far it is the
seller who has the incentive to ask for a contract to govern the transaction. Without
such a contract the seller does not enter the exchange and realized payoffs are zero
for both parties. Hence, in this framework, the vulnerable sellers can be seen to
impose on their partners some of the contracting terms (at least in what concerns the
safeguarding clauses we have mentioned before).
If we consider the seller and the buyer to engage in Nash cooperative bargaining
and to have equal bargaining strengths, then with no contract, but assuming no
opportunistic behavior, the agreed-upon ex ante gross returns to each party are:
RaS = KS + V + 12(R?KS ?KB ?V)
RaB = KB + 12(R?KS ?KB ?V).
If there is no contract, once KS, KB, and V are sunk, one of the parties might ask
for the initial agreement to be renegotiated. Nash bargaining leads to the following
ex post gross returns:
RpS = OS + 12(R?OS ?OB)
RpB = OB + 12(R?OS ?OB).
For the seller, the difference between ex post and the ex ante returns is:
RpS ?RaS = ?12[(KS ?OS) +V ?(KB ?OB)] < 0,
35
so it is the seller who might have an incentive to ask for a contract (to protect against
buyer?s opportunism). Then, the seller?s choice is between the net return when there
is no contract and hold-up occurs, which is OS + 12(R?OS ?OB)?KS ?V, and the
net return when a contract is used and the buyer is constrained by law to follow it,
which is 12(R?KS ?KB ?V ?C).
The seller will opt to have the transaction governed by a contract if:
OS + 12(R?OS ?OB)?KS ?V < 12(R?KS ?KB ?V ?C),
which upon simplification implies (KS ?OS) +V ?(KB ?OB) > C.
When the seller is the more vulnerable of the two parties, the model has three
simple predictions. Most importantly, it shows that greater relationship-specific in-
vestment by the seller (higher KS ?OS) makes adopting a contract more likely, while
greater relationship-specific investment by the buyer (higher KB ? OB) makes the
use of a contract less likely. Greater seller background vulnerability (higher V) also
makes adopting a contract more likely.
1.5 The econometric details of our analysis
1.5.1 The estimation approach
The transparent model we have presented above shows that the possibility of a
negative ?1 is very plausible. However, the empirical studies on contracting relying
on a regression equation similar to (1.1) have usually led to positive estimates for
36
?1. One reason for this might lie in the difficulty of obtaining consistent estimates
of parameters such as ?1, a fact well recognized in the literature. Chiappori and
Salani?e (2002, p. 27?28) consider that ?it is hard to feel completely satisfied with the
methodology of [transaction costs theory] studies. [...] A problem with this literature
is that it usually does not control for the possible endogeneity of the right-hand side
variables.? Masten and Saussier (2002, p. 13) comment that ?the specificity of assets
and the level of investment in those assets, which are treated as exogenous variables
in much of the research, are themselves decision variables [...and] should, by rights,
be treated as endogenous variables. Only a few studies have made tentative steps in
that direction.? Sykuta (2005) expresses a similar view.
As mentioned above, Crocker and Reynolds (1993) use exogeneity tests to sup-
port their results. Our data fail such tests. The endogenous nature of the decision
to invest in relationship-specific assets leads us to employ the instrumental variables
(IV) technique when estimating an equation such as (1.1). We will show below that
IV is the proper approach under various scenarios since it avoids several potential
problems a researcher may encounter.
A first scenario is that the decisions on contract characteristics and investment
are taken at the same time, each motivated by transaction and firm features. Then,
the true model is represented by a system of two equations of the following form:
C = ?2 + ?2Z2 +?2 (1.2)
I = ?3 + ?3Z3 +?3, (1.3)
37
where Z2 and Z3 may contain similar variables, but need not necessarily be the same.
Under this scenario, using OLS to estimate equation (1.1) alone would be problematic,
in particular given the focus on ?1 while I does not in fact belong there. Since ?2 and
?3 may be correlated (because of unobserved variables which may affect both C and
I), I is correlated with the error term of equation (1.1), and this endogeneity leads
to biases. Hence, instead of obtaining a zero estimate for ?1, one would end up with
different result, leading to spurious conclusions. Nevertheless, in case equations (1.2)
and (1.3) represent the true model, if proper instruments are used for the decision to
invest in specific assets, the IV technique will correctly estimate equation (1.1) and
will return a zero estimate for ?1.
A second (more plausible) scenario considers the following system of equations:
C = ?1 +?1I + ?1Z1 + ?1 (1.1)
I = ?4 +?4Z4 + ?4 (1.4)
That is, contract characteristics depend on the specific investment and other
variables, while the decision to invest in specific assets depends on the set Z4 of
exogenous variables. Same as above, if error terms ?1 and ?4 are correlated, then an
OLS estimate of equation (1.1) alone will be ridden by endogeneity bias. Also as
above, employing the IV method for equation (1.1) will lead to unbiased estimates,
provided that valid instruments are used for the relationship-specific investment.
The third scenario we may consider has the decisions on contract characteristics
38
and investment taken at the same time, depending on each other and on transaction
and firm features. This would imply the following system of equations:
C = ?1 + ?1I + ?1Z1 + ?1 (1.1)
I = ?5 + ?5C + ?5Z5 + ?5 (1.5)
In this framework it is likely that an increase in I leads to an increase in con-
tract complexity, followed by a subsequent increase in I. Hence, an OLS regression of
equation (1.1) alone is problematic, and in this case the result would be an overesti-
mate of ?1. Note that for this to happen, the error terms of equations (1.1) and (1.5)
need not be correlated. As mentioned, our main interested in the effect of specific
investment on contract complexity, and in this setup an IV approach on equation
(1.1) serves us fine: it returns an unbiased estimate of ?1. As before, the choice of
instruments is fundamental.
A fourth scenario may be discussed. The specification of equation (1.1) may be
erroneous and the story reversed, hence equation (1.5) alone may represent the true
relationship between C and I. In this case, running OLS on equation (1.1) would
lead to spurious conclusions, and the same would happen with the IV approach.
However, this scenario has not been a part of the TCE literature since it implies that,
despite varying across firms, contract features are exogenous (predetermined), and
that investment decisions have no impact on contract characteristics.
Going back to the theoretical model we have presented in Subsection 1.4.3, one
39
of its important features is that the choice of contract complexity is shown to be con-
tingent. That is, contract complexity depends on the relationship-specific investment
of one party, but also on other factors which determine that party?s vulnerability to
opportunistic behavior. The model also uncovers fundamental differences between
buyers and sellers in a transaction, with direct implications on the governance struc-
ture of the agreement. For these reasons, we decide to estimate equation (1.1) using
data on each side of the transaction: we investigate separately the impact of buyer?s
specific investment on contract complexity from that of seller?s. This approach has
several advantages. One is that it allows us to relate to existing studies in the TCE
literature, most of which have investigated the impact on governance structure of one
partner?s decision to invest in specific assets. Another advantage of this approach
is related to the particular estimation method we employ ? details on this will be
presented further in the chapter.
1.5.2 Predictions which can be taken to data
Even though IV is the proper approach when estimating equation (1.1), as
Chiappori and Salani?e (2002) and Masten and Saussier (2002) clearly point out, most
of the existing empirical studies rely on OLS to estimate equations similar to (1.1).
In what follows, we will try to show the consequences of employing IV, rather than
OLS, on the estimates of main interest. For this, it is useful to consider an exogenous
measure of vulnerability to partner?s opportunism when no enforceable contract is
in place. We denote this by V ? it can be interpreted as background vulnerability,
40
capturing all factors not explicitly named as explanatory variables in equation (1.1).
We continue to make the same two assumptions on partners? vulnerability. The
first is that at any specific instance in a transaction one and only one party will be
vulnerable to the opportunistic behavior of the other. The second is that, at least
in Romania, the seller is the vulnerable party, even in the absence of a concrete
relationship-specific investment (evidence supporting this assumption was provided
in Subsection 1.4.2).
The following system of equations is considered to represent the link between the
decision to invest in relationship-specific assets and the choice of governance structure
(this model resembles the second scenario presented above):
C = ?6 + ?6I + ?6Z6 + ?6V + ?6 (1.6)
I = ?7 + ?7Z7 +?7V + ?7, (1.7)
In this model V represents a measure of seller background vulnerability (when the
seller does not undertake any relationship-specific investment). Thus, we expects that
?6 > 0. As before, subscripts indicating transactions have been omitted.
The theoretical model indicates that specific investment by the seller increases
the hazards of exchange for that party, requiring greater contract complexity (higher
C). Hence, if seller data is used to estimate equation (1.6), ?6 is expected to be
positive. On the other hand, the model indicates that relationship-specific investment
by the buyer decreases the likelihood of opportunistic behavior, leading to lower
41
contract complexity (lower C). Thus, if buyer data is used, the estimate of ?6 is
expected to be negative. In respect to equation (1.7), it is reasonable to assume that
?7 < 0 when the equation is applied to seller data (because more vulnerability on the
part of the seller increases the transaction-related costs of seller specific investment),
while ?7 > 0 when using buyer data (because increased seller vulnerability implies
that buyer relationship-specific investment has greater transactional advantages).
Since V is unmeasured, it is part of a composite error term. When equation
(1.6) is estimated by OLS, this unobserved heterogeneity leads to biased estimates of
?6. Given the above arguments, we can predict the direction of these biases.
When using seller data for C, I, and Z6, the negative relationship between I
and the composite error term (of which V is part of) leads to the OLS estimate of ?6
being biased toward zero relative to the consistent positive estimate of ?6. When using
buyer data for C, I, and Z6, the positive relationship between I and the composite
error term also leads to the OLS estimate of ?6 being biased toward zero, relative
to the consistent negative estimate. Hence, the directions of bias are such that the
difference between the OLS estimates on the buyer and seller sides will be smaller
than the difference between the consistent estimates.
The above offers several predictions that can be taken to data:
? if consistently estimated, ?6 is positive using seller data for C, I, Z6, and Z7;
? if consistently estimated, ?6 is negative using buyer data for C, I, Z6, and Z7;
? using seller data for C, I, Z6, and Z7, the OLS estimate of ?6 is biased downward
42
(toward zero) relative to the consistent estimate;
? using buyer data for C, I, Z6, and Z7, the OLS estimate of ?6 is biased upward
(toward zero) relative to the consistent estimate;
? using seller data for C, I, Z6, and Z7, the correlation between the error terms
of equations (1.6) and (1.7) (?6V +?6, ?7V +?7) is negative;
? using buyer data for C, I, Z6, and Z7, the correlation between the error terms
of equations (1.6) and (1.7) is positive.
One thing to underline regarding these predictions is that they all depend on
the form of the model being investigated. Precisely, when stating the above we
have considered the relationship between investment and governance structure to be
represented by equations (1.6) and (1.7), which constitute an extension of the second
scenario presented in Subsection 1.5.1. Should a different relationship be investigated,
such as the last two scenarios discussed in that subsection, the use of OLS combined
with the endogeneity issue would lead to different predictions. Such alternative forms
are possible, however, as we have pointed out, they have not been equally successful
in the transaction costs literature.
It is also important to emphasize once again that all our empirical predictions
depend on the assumption that, even in the absence of a concrete relationship-specific
investment, the sellers are vulnerable to buyers? opportunism. Interestingly, the only
other empirical study of which we are aware that contrasts the signs of the estimated
coefficients for asset specificity on the two sides of the transaction, Acemoglu et al.
43
(2005), generates signs that are the opposite of ours. Namely, if equation (1.1) is
applied to their theory of the make-or-buy decision based on the PRT framework,
with C representing the decision to integrate, then ?1 is negative for the seller and
positive for the buyer. But just as our prediction on signs depends on an assumption
of whether buyers or sellers are vulnerable in Romania, the prediction of Acemoglu
et al. (2005) rests on an assumption of whether backward or forward integration is
most relevant. Precisely, to obtain their results, Acemoglu et al. (2005) compare the
default situation of non-integrated partners with that of backward integrated firms,
and acknowledge the fact that the opposite results apply when the relevant margin
is forward integration.
Hence, in both TCE and PRT, it is the prediction of opposite signs on the two
sides of the transaction that is likely to be a feature of empirical studies, not which
sign is positive and which is negative. The latter issue depends on the background
environment of the country under examination and the particular sectors that are
being investigated.
1.6 The survey and main variables
In what follows, we discuss those features of the survey, data, and variables
essential for understanding our argument. For presentation purposes, we choose to
relegate to Appendix A details about the survey, and to Appendix B at the end of
this dissertation details on variables? construction.
44
1.6.1 The survey generating our data
We use a survey of two hundred fifty-four Romanian companies conducted in
the second quarter of 2001. The companies were medium- and large-sized, undertook
activities in various productive sectors of activity, and were located in and closely
around the twelve largest cities of the country.
Using different survey instruments, four different persons in each company were
questioned in face-to-face interviews: the general manager, the managers of the sales
and procurement departments, and the head of the legal department. For those firms
that did not have a legal department constituted, the lawyer most frequently dealing
with the company was interviewed instead.
The general manager questionnaire asked mostly about the organization of the
company and the way it interacts with its partners, while the legal relations ques-
tionnaire focused on companies? approach on designing the contracts and on their
reliance on the commercial court system. The other two questionnaires, administered
to the sales and procurement managers, are the ones providing most of the data for
this dissertation chapter. In one particular section of each of these two survey in-
struments the respondent is asked a large set of questions about a specific agreement
its company entered into (a different agreement for each manager). By this we mean
an agreement the respondent chooses freely and with which he is familiar, that pro-
vided for a sale or purchase in the past six months. The questionnaire instructed
the respondent to choose a transaction which involved either a traditional or a new
product, and to not choose the agreement on the basis of its success or failure.
45
An important feature of the survey instruments addressed to the sales and
procurement managers is symmetry between and symmetry within them. Symmetry
between the two questionnaires is given by the fact that they ask almost identical
questions, but while one asks about firm?s experience as seller, the other asks about
the experience as buyer. Symmetry within the two survey instruments means that
each of them contains two sets of similar questions in the specific agreement section;
in addition to questions about his own side of the transaction, each respondent is
asked his opinion on similar issues concerning the trading partner. Some examples of
symmetric questions are provided in Appendix A at the end of this dissertation.
As theory suggests, transaction characteristics are the main cause for a particu-
lar governance structure, so our approach in using the data is to have the agreement,
rather that the firm, as the unit of observation. Given the twofold symmetry men-
tioned above, we have information about sellers and buyers in two agreements for
each interviewed company. We can therefore put together a larger data set by com-
bining information on buyers from the two main survey instruments, on one hand,
and information on sellers from the same two sources, on the other hand, leading to
a sample of potential five hundred eight observations. What makes this feature of
our data possible is that we questioned company officials who had specialized jobs in
charge of transactions and who had knowledge of the characteristics of the partner
firm?s operations.
Lastly, we consider agreements between Romanian and foreign companies to be
particular because they are governed by a different body of law and because of the
46
particular nature of the opportunistic incentives. We therefore exclude these transac-
tions from our sample, and use four hundred twenty-three observations (agreements).
1.6.2 The main variables we employ
We now turn to a brief description of the main variables, while those variables
we use as instruments for the possible endogenous specific investment are revealed
in a subsequent section. Additional details on all variables we employ (including
those used to test the robustness of the results) are presented in Appendix B at the
end of this dissertation. The reader will also find there some of the actual questions
contained by the survey instruments. Descriptive statistics for all variables we have
constructed are presented in Table 1.1 at the end of this chapter.
In this chapter, contract complexity is an index created by assigning scores to
various features of a contract. We construct such an index because contracts include
many clauses which are highly dependent on each other, hence focusing on individual
contract provisions ignores the important interactions between them. This view is
expressed by Masten and Saussier (2002, p. 14) and is also adopted by Sykuta (2005).
The information we collect on contracts refers to their form and to a particular
clause each partner may include. Regarding the form of the contract, we have asked
whether the agreement was oral or written. If written, we were interested in contract?s
number of pages, whether it was authenticated18 or not, whether it was specifically
18Authentication is done by a notary public and is helpful if disputes arise, since such contracts
are easier to enforce. The reason not all contracts are authenticated is lack of legal knowledge and
the high fess notary publics charge for this service.
47
designed for the transaction, as opposed to being based on a form contract. We have
also obtained information on the length of the discussions with the legal adviser for
crafting the contract. Lastly, we have asked whether particular clauses were included
in the written contract: a penalty clause for late delivery (protecting the buyer in the
agreement) and a penalty clause for late payment (protecting the seller).
We use the above information to construct three contract complexity variables:
buyer, seller, and overall, with the latter including all the information contained in
the other two (there is a very high correlation between all three indexes). One reason
we do this is to match the existing empirical work. Since other authors only had data
on one side of the agreement, we also focus on just one side of the transaction and run
distinct regressions, one for the buyer, and one for the seller. For example, we use the
buyer contract complexity index in regressions that relate contract features desired by
the buyer to buyer characteristics. As it will become apparent further in the chapter,
this approach also allows us to employ more powerful estimating procedures. Finally,
we also run a single regression, using the overall contract complexity index together
with variables representing both sides of the transaction.
The presentation of the main independentvariables begins with therelationship-
specific investment dummy variables, whose sources of information are the man-
agers of the sales and procurement departments. We have constructed a buyer and a
seller specific investment dummy variable. The former equals one if the buyer under-
took any special investment in dealing with the seller in the specific agreement (rather
than an alternative supplier), zero otherwise. For the seller relationship-specific in-
48
vestment we use information provided by a question focusing on customization in the
production process. Details on these variables? construction are provided in Appendix
B. As previously mentioned, although we use a simple dummy variable, even this is
an advance on the information often used in empirical studies, who frequently rely
on imprecise proxies for the variables of interest (Shelanski and Klein, 1995; Masten
and Saussier, 2002; Sykuta, 2005; Macher and Richman, 2006). It is worth noting
that these variables capture the investment into the process of using or producing the
good, investment which is made prior to the exchange, and which is something very
tangible and observable.
Quality of the courts is a dummy variable, equal to one if the commercial
section of the court in respondent?s region is considered to offer services of a higher
than average quality, zero otherwise. Since complex contracts would only be useful if
the relevant court can be trusted to enforce them, the inclusion of this variable tests
whether the characteristics of courts affect the organization of transactions. This is
a conservative test, since variations in court quality across regions must be relatively
small within Romania, which has a highly centralized legal system. The use of this
variable follows a small but growing literature on transition countries suggesting that
variations in the quality of the legal system do influence firm behavior (Murrell, 2001;
Johnson et al., 2002). Lastly, since the information on court services comes from the
legal adviser, it is used twice in the sample of agreements and so interpreted as one
party?s view on court system, regardless that party is the buyer or the seller in the
agreement.
49
Information dissemination is a variable which captures possible gains from
an informal relationship. Namely, information dissemination among buyers is a
dummy variable equal to one if other buyers of the good would learn about the seller
in the specific agreement defaulting its obligations, zero otherwise. Constructed in a
similar fashion, information dissemination among sellers is a dummy variable equal to
one if other sellers of the good would learn about the buyer in the specific agreement
not paying its obligations, zero otherwise. By using these variables we try to test
whether dissemination of relevant information is an effective deterrent of opportunis-
tic behavior in Romania, as generally assumed in the literature. Joskow (1985) and
McMillan and Woodruff (1999) are only two of the papers providing a similar dis-
cussion. To conclude, the theory indicates we should observe less complex contracts
when information spreads on the market.
First agreement is a dummy variable, equal to one if the specific agreement is
actually the first agreement between the parties, zero otherwise. We use this variable
to test the assumption that personal trust builds slowly, hence firms rely on more
complex contracts with first-time partners. In a similar vein, Heide and John (1990)
show that repeated transactions may develop a sentiment of continuity and trust
between partners, leading to reduced threat of opportunism. Also, past interactions
may lead to smoother current transactions due to improved coordination between the
parties ? for a similar argument on the effect of repeated interactions see Banerjee
and Duflo (2000). To conclude, the development of trust and improved coordination
imply that we should observe less complex contracts when partners in the specific
50
agreement have traded before.
Exogenous market uncertainty is a dummy variable equal to one if unpre-
dictable changes in weather or transportation links would induce important variations
in the demand or supply of transacted good, zero otherwise. We use this informa-
tion, which is provided by the general manager, to investigate the effect of market
uncertainty on the use of complex contracts. Expectations on the effects of market
uncertainty on governance form of the agreement are mixed. In our framework, one
may argue that more complex contracts will be designed when market exogenous un-
certainty is acute, as more clauses would be included so that solutions are specified
for the various possible situations. On the other hand, limited foresight and increas-
ing costs of designing contracts could make transacting partners rely on more flexible
(at the same time less complex) arrangements when market uncertainty plays an im-
portant role. Such contracts would consist of clauses that set up general rules for
dealing with various situations, and rules to guide the renegotiations likely needed in
such environments. For discussions on the effect of market uncertainty on transaction
governance structures see Crocker and Masten (1991), Crocker and Reynolds (1993),
and Chiappori and Salani?e (2002).
Dependence on partner dummy variables are constructed for both buyers
and sellers. A buyer is considered to depend on the seller from the specific agreement
if, should the supplier fail to deliver the contracted goods, it would take a month or
more to find an alternative supplier. The dummy capturing seller?s dependence on
his customers is constructed in a similar fashion. We believe that the goods for which
51
and alternative buyer/supplier cannot be easily found constitute, in a sense, a specific
asset. Then, firms which are dependent on the current trading partner are expected
to behave accordingly, much in the same way firms which invest in relationship-
specific assets do, and to take precautionary measures against opportunistic behavior
of their partners. The similarity of the two concepts implies that the dependence
variables will be used as explanatory variables (as the relationship-specific dummy
variables are used) and that their effect on contract complexity should match that of
the specific investment variables. Apart from their obvious relevance as explanatory
variables, the dependence on partner dummy variables play an important role in
the estimation procedure; as it will become apparent at a later stage, the inclusion
of the dependence variables supports the choice of instruments used to counter the
unobserved heterogeneity problem we have mentioned before.
1.7 A hint on the existence of hold-up
Before getting into a detailed analysis of the phenomenon, one question worth
raising is whether opportunistic behavior is really present in the Romanian business
environment, thus constituting a threat for transacting partners. One could consider
that Romanian companies have not yet experienced the opportunism (at least in the
forms discussed in the literature). This could be the case given the Romanian eco-
nomic environment?s relatively new development from a command-and-control system
and given the arguably limited occurrence of this behavior under the former regime.
A different view is that the human nature is fundamentally opportunistic (Lyons,
52
1994) and therefore such a behavior should be present in all economic environments,
including the highly centralized and controlled ones. Our data tend to support the
latter belief.
For each agreement in our sample, we have information on whether partners
have made a relationship-specific investment and also whether they have asked for
agreement renegotiation during its implementation. When we put this information
together some interesting insights emerge, which we present in Table 1.2 at the end
of this dissertation chapter.
Table 1.2 tends to indicate that when sellers undertake specific investments,
buyers are more likely to ask for renegotiation of agreement. As one would mainly
ask for renegotiations in order to get better terms for himself, we may in fact be
faced with the definition of hold-up presented before. The ?2 test of independence
indicates we should reject the null hypothesis of no relationship between row and
column frequencies and conclude that the change in percentages is significant (p-
value = 0.062).
Given the symmetric nature of our study, Table 1.3 shown at the end of the
chapter presents the reverse relationship, between buyers? relationship-specific invest-
ments and sellers? requests for renegotiation. Even though the row percentages look
as before, the ?2 test returns p-value = 0.395, which means the null hypothesis can
no longer be rejected. At this point it remains an open question whether this re-
sult comes from the low number of buyers undertaking specific investments or it is
an indication of the fundamental differences between buyers and sellers mentioned
53
above. However, without claiming this is solid proof of hold-up (since, for instance,
the direction of causality may be debated) the two tables hint that opportunism may
be present in the Romanian business environment. Thus, it would be fruitful to
investigate whether companies rely on complex contracts to defend against such a
behavior.
1.8 The results of our analysis
1.8.1 The effect of relationship-specific investment
We begin our presentation of the main results with the OLS estimates, and con-
tinue with the consistent ones. OLS is the typical approach found in the transaction
costs literature, which, however, has been shown to not be the best practice. Our
previous arguments have suggested that even though OLS estimates could be unsatis-
factory, they may provide important interpretive evidence, especially when compared
with the consistent estimates.
An important decision we made was to estimate the model separately for buyer?s
and seller?s side of the agreement. Namely, we investigate the impact on seller (de-
sired) contract complexity of seller?s relationship-specific investment and of other
variables assumed to influence the decision on contract?s complexity. Conversely,
we investigate the effect of buyer?s specific investment, along with a similar set of
explanatory variables, on the contract complexity a buyer would demand for the
agreement.
54
As already noted, there are at least two reasons behind this approach. First, this
approach allows a clearer diagnosis of causes of differences between our results and
those previously obtained (comparison with previous empirical findings is sought, and
the authors who only had information on one side of agreements have done exactly
the same). Second, adopting this route avoids some econometric obstacles: due to
the fact that our main independent variables (relationship-specific investment) are
dummies, combined with their endogenous nature, we employ the treatment-effects
(TE) model. Dealing with a single treatment is more accessible and allows us to
use the full ML estimator19. Employing this powerful statistical technique is also
preferred as it will permit testing the quality of our instruments. An approach less
commonly used in the literature is also taken, and a single regression is used ? it relies
simultaneously on data from both sides of transaction and uses as dependent variable
the overall contract complexity score we have constructed.
Another decision we made was to use the two dependence on partner dum-
mies as explanatory variables for contract complexity. As pointed out when we have
presented the construction of our variables (Subsection 1.6.2), this is natural given
the similarities between the concepts captured by the dependence on partner and
the specific investments variables. Moreover, this use of the dependence variables
supports the choice of instruments for the decision to invest in relationship-specific
assets. Should the reader not concur with our view, additional results coming from
different uses of the dependence on partner variables are presented in Appendix C.
19To our knowledge, ML with two treatments has not been implemented yet.
55
1.8.1.1 Ordinary least squares estimations
The first two columns of Table 1.4 at the end of this chapter present the OLS
estimates of equation (1.1) for the two sides of the transaction. Both the seller and
the buyer relationship-specific investment coefficients are positive, with the former
highly statistically significant but the latter not significant. These results match the
overwhelming majority in the literature: greater specific investment leads to more
elaborated methods of transactional governance.
The last column of Table 1.4 shows the OLS estimates of the single regression,
which uses data on both sides of the transaction simultaneously. The general picture
does not change, suggesting that the choice of approach (either separate regressions
or a single one) is not critical, a conclusion that is useful when interpreting estimates
that are constrained by the data collection process.
The results in Table 1.4 indicate there is a positive relationship between contract
complexity and the quality of services offered by commercial courts. This matches
our expectations, indicating that firms do take into consideration court features when
designing their contracts (one would construct a more complex contract only if would
expect that contract to be enforced in court). This is evidence that contract com-
plexity, as we measure it, is an element of a legal strategy that places emphasis on use
of the formal legal system. This result is also interesting as it adds to the expanding
literature on the importance of the formal legal system in transition and developing
economies. In particular, this positive relationship indicates that, even in a legal
system where regional variation in quality of court services is very limited, there is a
56
significant effect of court quality.
The estimated coefficients on information dissemination and first agreement are
non-significant. One interpretation of these results is that choice between very differ-
ent modes of transactional governance (between reliance on the formal legal system
and reliance on relational agreements) is not based on the types of variables suggested
by TCE analysis of developed countries. There is no inconsistency here. Choice be-
tween closely related alternatives, such as between more or less complex contracts,
might be dominated by efficiency considerations, while choice between radically differ-
ent alternatives might reflect a different calculus (Nelson and Winter, 1982). Murrell
(2003) suggests that choice between formal law and informal relationships is domi-
nated by the peculiarities of transition, such as firm history, ownership, and relations
with the state.
While some of the results on relationship-specific investment change signifi-
cantly when different estimation techniques are adopted in the next subsection, the
results on other explanatory variables remain constant throughout this chapter. For
example, exogenous market uncertainty never has a significant effect. Hence, no fur-
ther comments are made on market uncertainty, reputation effects, and the quality
of court variables. In contrast, comments on the seller and buyer dependence on
partner variables are reserved for the next subsection, given their particular role in
the regressions.
57
1.8.1.2 Maximum likelihood treatment-effects estimations
The econometric section has shown that, despite its frequent use, OLS is not the
most suitable estimation procedure. Given specific investment?s endogenous nature,
other estimation methods should be employed instead, and we rely on the IV approach
to obtain the estimates for the two sides of the transaction.
To counter the probable biases caused by unobserved factors that can simul-
taneously affect contract complexity and relationship-specific investment, proper in-
struments will be used. We rely on technological and market characteristics, which
are exogenous determinants of specific investment, and which do not directly affect
the complexity of contracts. We present below our choice of instruments, and start
with those determining buyer?s decision to invest in specific assets.
A plausible assumption is that the existence of many potential sellers of a good
is an indicator that a standardized production technology is available, while the ex-
istence of few potential sellers is an indicator that a specialized production technol-
ogy in needed for that particular good. This assumption has direct implications on
the investments in specific assets the buyers may need to make. For a good with
many potential sellers (and standardized production technology) it may be more effi-
cient for buyers to undertake any desired customization themselves (by undertaking
relationship-specific investments). Following the same line of thought, for a different
good in that sector, with few potential sellers (and specialized production technol-
ogy), the buyers may be less likely to undertake customization (through specific in-
vestment) as they probably lack the knowledge to do so. Hence, we believe that the
58
number of potential sellers in a sector is a proxy for those technological factors influ-
encing whether the buyer undertakes customization, ultimately influencing whether
the buyer makes a relationship-specific investment.
Alternatively, another plausible assumption is that the existence of many po-
tential sellers of a good is an indicator of much niche production, meaning that
the buyer does not need to undertake customization (does not need to undertake a
relationship-specific investment). Conversely, the existence of few potential sellers of
a good could mean there are large mass manufacturers, who do not adjust production
to the particular needs of their customers, hence buyers have to undertake their own
customization (through specific investment). Therefore, under this alternative sce-
nario also, the number of potential sellers in a sector is a proxy for those technological
factors influencing whether the buyer undertakes relationship-specific investments.
The two paragraphs above argue that the number of potential sellers is a proxy
for technological factors that determine buyers? specific investment decisions. How-
ever, they provide competing arguments and opposite predictions on the direction of
influence, suggesting that the proxy function can have different signs, depending on
which type of technology exists, which can vary across sectors. Therefore, we use as
explanatory variables for buyer?s relationship-specific investment the interaction of
two sets of dummy variables; one indicating whether there are few or many potential
sellers, and a set of sector dummies. This leads to eight dummy variables (details on
the way they are constructed are presented in Appendix B).
The above argument suggests that predicting the sign of the relationship be-
59
tween these dummies and buyer specific investment entails knowing features of the
production technology, hence no predictions are explicitly made.
A similar reasoning holds for seller?s decision to invest in specific assets. In one
type of sector, many potential buyers implies a relatively homogeneous good that
sellers do not customize (sellers do not undertake relationship-specific investments),
while few potential buyers implies niche usage, leading to specific investments by
sellers needed to deliver the particular good. The competing scenario considers a dif-
ferent sector, where many potential buyers indicates a differentiated product where
there are many niche sellers each customizing the product to a different set of buyers
(by undertaking relationship-specific investments), while few buyers indicates little
demand for customization (hence fewer seller specific investments). The implementa-
tion is the same as on the buyer side: the determinants of seller relationship-specific
investment are eight dummies representing the interaction between a dummy for the
number of potential sellers and four dummies for the different sectors of activity
considered.
We have tried to show above that the interaction of number of potential partners
with the sector of activity has a direct impact on the relationship-specific investments
made by buyers and sellers. However, in order to constitute proper instruments, our
sets of dummies must not influence contract complexity directly and to be exogenously
determined.
Given that Romania is a transition economy, we consider that old enterprises
have simply continued the production processes initiated in the former regime, while
60
the new companies have been established to supply for the new needs, and only subse-
quently choose the transactional strategies. For a similar argument on the exogeneity
of the sector of activity in transition see Murrell (2003). For the analysis in this
chapter the implications are direct: since the sector of activity is predetermined, it
is unlikely that it and the contract complexity are chosen subject to the same set
of determinants, hence the sector of activity alone is not likely to affect the chosen
complexity of contracts. In addition, contract complexity is chosen not on the basis
of the potential number of buyers and sellers before a contract is signed but rather on
the basis of the actual number of buyers or sellers that would exist after the contract
is signed and relationship-specific investment is undertaken. Also, the number of po-
tential partners in a particular sector is exogenously determined since it is a feature
of the market, which cannot be influenced by firms individually.
To conclude, our instruments are technologically induced characteristics of the
markets in which the firms are placed, rather than indicators of the firm?s actual
vulnerability to partners? opportunistic behavior.
Given the binary form of our specific investment variables, we employ the TE
model and estimate the system made of equations (1.6) ? the second stage equation ?
and (1.7) ? the first stage equation ? by ML. We use seller and buyer data separately
and rely on the instruments presented above for sellers? and buyers? decision to invest
in relationship-specific assets. As already argued, the sets of dummy variables we use
constitute proper instruments, hence they do not appear in the second stage equation.
This is where the use of variables indicating dependence on the current trading
61
partner becomes crucial. Their inclusion in the second stage equation counters the
possibility that our instruments proxy some post-specific investment market thinness
and enhance the acceptability of the exclusion restriction for our instruments20.
As mentioned at the beginning of this section, for various reasons, we estimate
the system of equations made of (1.6) and (1.7) on each side of the contractual
relationship. The main results of this chapter emerge, and they are presented in
Tables 1.5 and 1.6 for the seller?s and buyer?s side, respectively.
When endogeneity of relationship-specific investment is considered, the results
differ significantly from the OLS estimates. Namely, there is a large increase in the
size of the estimated coefficient for seller?s specific investment variable, while the
sign on the buyer?s side is reversed. The two coefficients are now of roughly equal
magnitude but of opposite signs, matching our expectations from Subsection 1.5.2.
The effects of relationship-specific investment on contracting behavior are quan-
titatively significant. The decision to undertake such investment produces a one stan-
dard deviation change in contract complexity. Specific investment by the seller ne-
cessitates a change equivalent to moving from no written contract to a contract that
specifies all relevant details, whereas relationship-specific investment by the buyer
would cause a change of equivalent magnitude in the reverse direction.
Moreover, the differences between the OLS and the ML TE estimates are also as
predicted, and the signs of ? (the correlations between the error terms of the first and
20Moreover, while this exclusion restriction is useful if it is correct, making for more powerful
tests, it is not absolutely necessary. When robustness are described, we show that dropping this
exclusion restriction has no appreciable affect on the results.
62
second stage equations) follow the analysis in Section 1.5. In addition, the statistical
significance of ? indicates the necessity of employing estimation procedures which
account for unobserved heterogeneity.
Regarding the dependence on trading partners, Tables 1.5 and 1.6 contain a re-
sult which may look troubling: the estimated coefficients for the dependence variables
are positive for both sides of transaction. This contrasts the opposing signs we have
predicted and obtained for the specific investment variables. We have argued above
that, even in the absence of relationship-specific investment, sellers are more exposed
buyers? opportunistic behavior. The sellers are therefore inclined to use more complex
contracts in order to counteract buyers? possible opportunism: the positive coefficient
of seller?s dependence on partner variable should therefore not be a surprise for the
reader. What may be surprising is the similarly positive (and significantly different
from zero) coefficient of the buyer dependence on partner variable since, given the
arguments in Section 1.5, one might expect a negative coefficient.
There are two possible reasons why this result does not appear. First, buyers
may be unable to properly assure sellers of their own dependence: unlike some hard
evidence of commitment such as specific investment (for which the parties must dis-
cuss technical specifications and particularities of the good), buyers? dependence on
sellers may represent a vague notion which cannot be properly demonstrated. It may
be that a buyer being dependent on its partner is not enough and instead it takes a
strong sign of commitment, such as a relationship-specific investment, to move the
relationship from one governed strictly by contractual clauses to one based on infor-
63
mal understanding of the partner. However, when buyers do depend on their sellers
(even though they cannot demonstrate this dependence) they will accept the proposed
complex contracts, and may even include additional clauses to safeguard them, thus
leading to the positive coefficient we observe in our regression. The second possible
reason behind this result is that we do not counter problems of endogeneity when
obtaining the results for the dependence on partner variables. Hence, the results for
specific investment and dependence on partners that are truly comparable are the
OLS estimates in Table 1.4, which show the same direction of influence on contract
complexity for the two pairs of variables21.
These results confirm the observation that there are two important problems
within the previous empirical literature that implements TCE to study the effect of
relationship-specific investment. The first is that unobserved heterogeneity must be
taken into account and must be countered. Second, the two sides of the transaction
might face very different circumstances, with specific investments affecting the two
sides in different ways, on one increasing transactional hazards and on the other
mitigating them.
1.8.1.3 Instrumental variables estimation
After presenting the results on each side of the agreement, we not turn to the
estimation which relies on all available data, and which uses as dependent variable
21Whether dependence on partner is endogenous or not is debatable. One argument could be that
dependence, as we measure it, is a feature of firm?s sector of activity, hence exogenously determined
(at least for firms in transition economies).
64
the overall contract complexity. Table 1.7 displays these results.
Since we now have two binary endogenous variables in the right hand side of the
regression, we implement an IV procedure suggested by Wooldridge (2002, p. 642).
In brief, the procedure is the following: a separate probit regression is run for each of
our dummy endogenous variable using all relevant explanatory variables (both those
included in the second stage regression and the instruments). The predicted proba-
bilities thus obtained are used as instruments for the relationship-specific investment
variables in the regression explaining the overall contract complexity. This procedure
is asymptotically efficient among those procedures for which the instruments are func-
tions of those variables included in the probit, and is also robust to misspecification
of the choice of the functional form of first stage equations.
Despite the lower statistical significance of the estimated coefficients, which is
probably due to the lower power of the procedure adopted, the results in Table 1.7
match the corresponding ones in the distinct specifications. The estimates we are
mostly interested in are again those for seller and buyer specific investment variables,
for which the contrasting signs are preserved. Given the lack of statistical significance
of these coefficients, it is useful to check whether the difference between them is
significant. For a Wald test of the null that the coefficients of the seller and buyer
relationship-specific investment variables are equal, the probability level is 0.064.
The size of these estimates is also very close to the corresponding ones obtained
relying only on data from one side of the transaction; Table 1.8 summarizes the esti-
mated coefficients obtained from the separate and the single regression by employing
65
the different procedures mentioned so far. This clearly indicates that the differences
between the impact of buyers? and sellers? specific investments on contract complex-
ity arise from accounting the unobserved heterogeneity, and not from our decision on
how to use the data (separately for each side of the transaction, or combined).
One note should be made on the estimates presented in Table 1.8. Since we
regress each party?s demand for contract complexity on that party?s features (the
most important being the decision to invest in specific assets), the coefficients in Table
1.8 are obtained from sets of three regressions which use the seller, buyer, and overall
contract complexity scores. Should the reader feel skeptical about the possibility to
compare the estimates from regressions with different dependent variables, we have
also run the regressions when only the overall contract complexity score is used. These
results are displayed in Appendix D at the end of the dissertation and are very similar
to those we have already presented.
To conclude, results indicate there are major changes as one moves from incon-
sistent estimates (OLS) to consistent ones, while there is little change as one moves
from separate buyer and seller regressions to a single regression. Hence, while the
latter move entails a real loss (giving up ML) there seems to be little gain in added
perspective on the signs and sizes of the coefficients.
1.8.2 Robustness tests of the main results
As the above paragraphs have shown, using data on both sides of the agreement
and relying on the overall contract complexity variable does not lead to additional
66
insights, but rather returns weaker results (probably since lower power procedures are
employed). We proceed and test the results from Tables 1.5 and 1.6, which constitute
the main findings of this chapter, by examining variations from the basic setup used
to generate them.
The first test we employ is the Hansen-Sargan overidentification test of instru-
ments, which is implemented in the standard IV context (ignoring the fact that our
endogenous variables are dummies). The null hypothesis of this test of overidentifying
restrictions is that instruments are valid (uncorrelated with the error term) and that
excluded instruments are correctly left out the estimated equation22. A rejection of
the null hypothesis casts doubt on the validity of the instruments.
We employ this test of instruments using data on each side of the agreement23.
When the buyer data is used, the test returns a Hansen J statistic equal to 7.621, lead-
ing to p-value = 0.267. Hence, the test confirms that the instruments we use to obtain
our most important economic result (the negative impact the buyer?s relationship-
specific investment has on contract complexity) are valid. For the seller?s side, the
test returns Hansen J statistic = 11.957 and p-value = 0.063. Given the borderline
p-value we obtain in this case we choose to employ the same test for the single regres-
sion as well. When data on both sides of the agreement are used the test returns a test
statistic of 9.820 and p-value = 0.365, hence the null hypothesis cannot be rejected.
An additional test is presented next, which also supports our choice of instruments
22In this terminology, the included instruments are our usual independent variables, while the
excluded instruments are the dummies obtained by interacting the number of potential partners
with the sector of activity.
23The estimated coefficients are not reported given that they are similar to those previously
obtained, and since the focus is now on the test statistic this procedure returns.
67
and the main findings thus obtained.
Two basic ingredients drive identification of the estimates in Tables 1.5 and 1.6.
First, there is the omission of the IVs in the first-stage relative to the second-stage
equation. Second, there is the non-linearity of the probit first-stage equation. This
means that the ML approach has a very important advantage: we can drop the first
element of the identification strategy and use the second, in order to evaluate the
acceptability of that first element. The relevant results appear in Tables 1.9 and
1.10, and differ from those in Tables 1.5 and 1.6 by the inclusion of the first-stage
explanatory variables in the second-stage equation.
Results indicate that the IVs maintain their significance in the first-stage and
have low explanatory power in the main equation. Hence, the impact on the vari-
ables of interest from the addition of excluded instruments in the main equation is
very limited ? nothing in Tables 1.9 and 1.10 suggests the necessity of revising the
conclusions that have been drawn so far.
As another robustness test of our main results we employ an alternative to the
TE model estimated by ML and rely on the two-step consistent estimator proposed
by Heckman. Table 1.11 presents the results for each of the two sides of the agreement
separately using this popular two-step procedure. The estimates thus obtained closely
match those discussed previously, therefore supporting our earlier comments.
Finally, the robustness of the main results is tested by the addition of some
further explanatory variables in the second-stage equation. We choose to (individ-
ually) add variables that have sometimes been suggested to impact on contractual
68
relationships between companies in transition economies: there are three variables
representing the past or present role of the state in the activity of one of the two
firms24, a variable indicating the location of the two partners, and two variables cap-
turing information on the age and size of one of the two partners.
Tables 1.12 and 1.13 show these results, indicating that some of these newly
added variables are significant. However, it is worth pointing out that none of these
additional results alter any of the previous conclusions on specific investment. The
results in Tables 1.12 and 1.13 indicate that one of the firms? ties with the state seem
to have an impact on complexity of contracts: current ties (through shares owned
by state) or past connections (stemming from state origins) lead to more complex
contracts governing the transactions these companies enter into. Firm size also seems
to matter: contracts which involve at least one large firm are more complex, probably
indicating recourse to other methods of contract enforcement or economies of scale
in using the courts. Finally, whether firms are located in the same region (indicating
locational specificities) or one of the firms is old (established under the previous
regime, indicating reputation) does not seem to influence the complexity of contracts.
1.8.3 Additional results: the effect of buyer prepayments
The theoretical model in Subsection 1.4.3 has shown that opposing effects on
contract complexity are expected from buyers? and sellers? relationship-specific in-
vestments. However, the actual direction of influence was based on the particular
24Since this is information gathered from the general manager of each interviewed firm, it is used
twice in the sample of agreements, and hence needs to be interpreted accordingly.
69
assumption that sellers are exposed to buyers? opportunism even in the absence of
an investment in a relationship-specific asset. Romanian evidence supported this as-
sumption; some of the key factors assumed to lead to sellers? vulnerability were lack
of external financing combined with the likely scarcity of internal financial resources,
and the fact that sellers usually need to commit resources to produce the good and
satisfy the agreement.
But the presence of this intrinsic (background) vulnerability is dependent on
buyers not doing anything to relieve it. The easiest way for the buyer to do so would
be to offer prepayments. As only one example, McMillan and Woodruff (1999) find
evidence that Vietnamese firms use prepayments as a hostage exchange, to balance
the risk between buyers and sellers.
Even though our data show that only 3.5% of buyers pay the bill in full be-
fore delivery, the data also show that some prepayments do happen: in a third of
the contracts under study buyers made a payment of 30% or more of the total bill
before delivery. We believe that when the buyers make important payments before
goods are delivered, this constitutes a clear signal of commitment, which should con-
vince sellers of their partners? good intentions. Since prepayments expose buyers to
sellers? opportunistic behavior, the transactions are likely to become self-enforcing,
which can be governed by less complex contracts. Therefore, there should be a large
difference between the results for those agreements where buyers do not make signif-
icant prepayments and where they do. Namely, without significant prepayments the
main results (presented in Tables 1.5 and 1.6) should hold. But in the opposite case,
70
the reduced sellers vulnerability (or even shifted vulnerability, to the buyers? side)
would lead to significant changes of the impact of buyers? and sellers? investments in
relationship-specific assets.
To formally test the above claims in a completely satisfying way would be an
extremely challenging task. Difficulties arise from the likely endogenous nature of the
decision to offer prepayment, combined with the relatively small number of significant
prepayments, and the possible sample selection problem arising from buyers? decision
to offer prepayment. Hence, the following results are merely suggestive.
To better highlight the impact of prepayments, we present the results after split-
ting the sample into two, depending on whether buyers made a significant prepayment
or not. The analysis which led to our main results (TE estimated by ML) will be
repeated for each subsample. This will lead to four sets of results: buyer and seller
estimates with and without a significant prepayment. For presentation purposes, Ta-
ble 1.14 at the end of this dissertation chapter does not include the first stage results
and displays the main ones for all the four regressions together.
The estimates are suggestive for our analysis. When no significant prepayments
are made the results are very similar to those previously obtained, the only difference
being that the coefficients of the specific investment variables are greater in magni-
tude. This should have been expected. The current working sample excludes those
agreements involving buyer prepayments, while our previous analysis uses data on all
transactions, which were attenuating the effects of relationship-specific investments
on contract complexity.
71
When buyers make significant prepayments the results change significantly from
the main ones. The estimated coefficients presented in the third column of Table 1.14
seem to indicate that prepayments of 30% or more of the total bill act as a major
sign of commitment from buyers? side. These prepayments appear to reverse the
vulnerability, in the sense that buyers seem to be exposed to sellers? opportunism.
Then, sellers? investments in relationship-specific assets attenuate this exposure, thus
leading to less complex contracts (the positive and highly significant ? supports this
view). However, the last column of Table 1.14 could be seen to indicate that sellers
are still the exposed party, with direct implications on the effect of buyers? specific
investments on the complexity of contracts (the estimated coefficient of interest is
negative, and the correlation between the error terms is positive, even though only
marginally significant).
It is hard to conclude which is the true situation in this case. In fact, we are
probably faced with a new setup, different from that we have considered before. Our
theoretical model from Subsection 1.4.3 has assumed that one party is vulnerable
to the opportunistic behavior of the other. However, when buyers make significant
prepayments, it is likely that both parties are simultaneously exposed to each other.
What are the implications on contract complexity of this bilateral exposure to op-
portunism? Our model does not provide an answer, hence this situation needs to be
analyzed further.
To conclude on buyers? prepayments, the results must be treated with great
caution given the obvious econometric issues that have been ignored. Nevertheless,
72
the results do provide support for the previous discussion. In particular, they show
that the effect of relationship-specific investment on contract complexity is highly
dependent on which side of the relationship is more likely to be vulnerable.
1.9 Conclusions
It is often mentioned that the main difference between TCE and PRT is the
prediction on the impact of partners? specific investments on the likelihood of integra-
tion. While in TCE any asset specificity is considered to increase the level of AQRs
at play, thus leading to more integrated forms of governance, in PRT the impact of
buyer?s and seller?s relationship-specific investments on the probability of integration
is considered to have opposite signs.
This dissertation chapter adheres to the view that contracting is an alterna-
tive to VI, and considers the choice of contract complexity as a measure firms take
to counteract the exposure to the opportunism of their trading partners. Hence,
the decision on contract complexity is similar to the make-or-buy decision which is
frequently discussed in the literature.
A simple theoretical model based on the rent-seeking branch of TCE is pro-
posed. The model indicates that, even under the assumptions of this theory, the
impact of buyer?s and seller?s specific investments on contract complexity has oppo-
site signs: while one increases, the other diminishes complexity of contracts. The
particular signs of the two effects is shown to depend on the initial vulnerability of
partners to the opportunistic behavior of the other. In the case of the Romanian
73
business environment, and probably in that of other transition, developing countries,
the sellers are shown to be vulnerable to buyers? opportunism, even without making a
specific investment. Then, the model predicts that sellers? investments in specific as-
sets lead to greater exposure, and ultimately to more complex contracts, while buyers?
investments balance the threats and lead to less complex contracts being adopted.
The chapter represents a significant improvement in what concerns data quality.
In particular, detailed data is collected, which allows for the analysis to be conducted
on both sides of agreements spanning a large variety of the Romanian economy. Also,
the concept of specific investment is measured directly and consistently across firms
and sectors.
But probably the main contribution regards the econometric analysis being
conducted. It addresses important concerns raised with existing studies, most of
which have relied on OLS as the estimation method, thus neglecting the potential
endogenous nature of the decision to invest in relationship-specific assets. Unobserved
heterogeneity is shown to blur the differences between the OLS estimates for buyer?s
and seller?s relationship-specific investments. Precise predictions are made, all of
which are confirmed by the data.
Apart from specific investment, other factors are investigated. Most significant
among them is the quality of services provided by commercial courts, which is shown
to impact on contract complexity: parties design more elaborated contracts when the
quality of services is higher than average. This finding has two major implications.
First, it corroborates with those on contract complexity, indicating that Romanian
74
companies do use the formal legal system in the way theory predicts. Second, it shows
that in the highly centralized Romanian legal system, where only small variation in
quality of court services is possible, even minor court reforms could have significant
effects on the use of law.
Last but not least, the impact of buyer prepayments is investigated. Even
though in this case the econometric analysis does not match the standards from the
rest of the chapter, the insights are relevant. Buyer prepayments are shown to act
as signs of commitment, thus balancing the initial seller vulnerability, with direct
impact on the relationship between contract complexity and relationship-specific in-
vestments. This provides additional evidence for the contingent effect of partners?
specific investments on contract complexity.
75
Figure 1.1: The distributions of the contract complexity scores
76
Table 1.1: Descriptive statistics
Number Mean Standard Min Max
Variables: of obs. deviation
Overall contract complexity 423 4.101 1.763 0 8
Seller contract complexity 423 3.416 1.508 0 7
Buyer contract complexity 423 3.296 1.536 0 7
Seller specific investment 418 0.246 0.431 0 1
Buyer specific investment 421 0.119 0.324 0 1
Quality of the courts 413 0.569 0.496 0 1
Information dissemination among sellers 415 0.800 0.400 0 1
Information dissemination among buyers 408 0.841 0.366 0 1
First agreement 423 0.286 0.452 0 1
Exogenous market uncertainty 421 0.247 0.432 0 1
Seller dependence on partner 411 0.440 0.497 0 1
Buyer dependence on partner 413 0.378 0.485 0 1
Construction 423 0.234 0.424 0 1
Heavy industry 423 0.303 0.460 0 1
Light industry 423 0.201 0.401 0 1
Other sectors 423 0.262 0.440 0 1
Few potential buyers 400 0.170 0.376 0 1
Few potential sellers 403 0.489 0.500 0 1
Few potential buyers in construction 400 0.038 0.190 0 1
Few potential buyers in heavy industry 400 0.068 0.251 0 1
Few potential buyers in light industry 400 0.020 0.140 0 1
Few potential buyers in other sectors 400 0.045 0.208 0 1
Many potential buyers in construction 400 0.195 0.397 0 1
Many potential buyers in heavy industry 400 0.240 0.428 0 1
Many potential buyers in light industry 400 0.178 0.382 0 1
Many potential buyers in other sectors 400 0.218 0.413 0 1
Few potential sellers in construction 403 0.079 0.271 0 1
Few potential sellers in heavy industry 403 0.194 0.396 0 1
Few potential sellers in light industry 403 0.092 0.289 0 1
Few potential sellers in other sectors 403 0.124 0.330 0 1
Many potential sellers in construction 403 0.149 0.356 0 1
Many potential sellers in heavy industry 403 0.112 0.315 0 1
Many potential sellers in light industry 403 0.112 0.315 0 1
Many potential sellers in other sectors 403 0.139 0.346 0 1
Significant buyer prepayment 421 0.335 0.472 0 1
State has control 423 0.165 0.372 0 1
State owns shares 423 0.188 0.390 0 1
Some state origins 423 0.662 0.474 0 1
In same region 423 0.482 0.500 0 1
Firm founded before 1990 419 0.558 0.497 0 1
Large firm 422 0.287 0.453 0 1
77
Table 1.2: A hint on the existence of hold-up (I)
Did the seller make a Did the buyer ask for
relationship-specific contract renegotiation? Total
investment Yes No
Yes 24 23.30% 79 76.70% 103 100%
No 48 15.29% 266 84.71% 314 100%
Total 72 17.27% 354 82.73% 417 100%
Notes:
1) Each cell of the table contains the number of observations and the row percentage.
2) The ?2 test of independence returns p-value = 0.062.
Table 1.3: A hint on the existence of hold-up (II)
Did the buyer make a Did the seller ask for
relationship-specific contract renegotiation? Total
investment Yes No
Yes 12 24% 38 76% 50 100%
No 70 18.92% 300 81.08% 370 100%
Total 82 19.52% 338 80.48% 420 100%
Notes:
1) Each cell of the table contains the number of observations and the row percentage.
2) The ?2 test of independence returns p-value = 0.395.
78
Table 1.4: OLS estimates of the determinants of contract complexity
Dependent variable:
Seller Buyer Overall
Independent variables: contract contract contract
complexity complexity complexity
Seller specific investment 0.652 0.651
(0.158)*** (0.191)***
Buyer specific investment 0.276 0.247
(0.236) (0.284)
Quality of the courts 0.284 0.340 0.314
(0.155)* (0.164)** (0.188)*
Information dissemination among sellers 0.163 0.210
(0.178) (0.241)
Information dissemination among buyers 0.228 0.106
(0.192) (0.271)
First agreement 0.120 0.213 0.133
(0.167) (0.173) (0.204)
Exogenous market uncertainty -0.085 -0.142 -0.097
(0.162) (0.173) (0.196)
Seller dependence on partner 0.502 0.400
(0.148)*** (0.197)**
Buyer dependence on partner 0.602 0.376
(0.153)*** (0.197)*
Constant 2.733 2.635 3.149
(0.190)*** (0.203)*** (0.270)***
Observations 390 390 371
Clusters 231 230 227
R-squared 0.091 0.066 0.092
Clustered standard errors displayed in parentheses, where each cluster is a firm.
* significant at 10%; ** significant at 5%; *** significant at 1%
79
Table 1.5: ML TE estimates of the impact of seller specific investment
Main equation results First stage results
Seller Seller
Independent variables: contract Instrumental variables: specific
complexity investment
Seller specific investment 1.806 Few potential buyers in heavy industry 0.694
(0.516)*** (0.275)**
Quality of the courts 0.327 Few potential buyers in light industry -0.045
(0.158)** (0.580)
Information dissemination among sellers 0.075 Few potential buyers in construction 1.166
(0.182) (0.364)***
First agreement 0.105 Few potential buyers in other sectors 0.758
(0.168) (0.323)**
Market exogenous uncertainty -0.070 Many potential buyers in heavy industry 0.110
(0.166) (0.219)
Seller dependence on partner 0.439 Many potential buyers in light industry 0.063
(0.148)*** (0.246)
Constant 2.560 Many potential buyers in construction 0.515
(0.232)*** (0.245)**
Constant -1.018
(0.167)***
Observations 373
Clusters 227
? (correlation between the error terms) -0.474
(0.153)***
Log-likelihood -853.797
Clustered standard errors displayed in parentheses, where each cluster is a firm. * significant at 10%; ** significant at 5%; *** significant at 1%
80
Table 1.6: ML TE estimates of the impact of buyer specific investment
Main equation results First stage results
Buyer Buyer
Independent variables: contract Instrumental variables: specific
complexity investment
Buyer specific investment -1.911 Few potential sellers in heavy industry -0.223
(0.375)*** (0.244)
Quality of the courts 0.368 Few potential sellers in light industry -0.535
(0.160)** (0.281)*
Information dissemination among buyers 0.246 Few potential sellers in construction -1.036
(0.191) (0.397)***
First agreement 0.176 Few potential sellers in other sectors -0.607
(0.167) (0.263)**
Market exogenous uncertainty -0.154 Many potential sellers in heavy industry -0.791
(0.168) (0.275)***
Buyer dependence on partner 0.526 Many potential sellers in light industry -0.347
(0.152)*** (0.282)
Constant 2.925 Many potential sellers in construction -0.626
(0.196)*** (0.263)**
Constant -0.723
(0.187)***
Observations 374
Clusters 224
? (correlation between the error terms) 0.726
(0.097)***
Log-likelihood -794.296
Clustered standard errors displayed in parentheses, where each cluster is a firm. * significant at 10%; ** significant at 5%; *** significant at 1%
81
Table 1.7: IV estimates of the impact of buyer and seller specific investments
Independent variables: Overall contract complexity
Seller specific investment 1.963
(1.264)
Buyer specific investment -2.022
(1.645)
Quality of the courts 0.266
(0.227)
Information dissemination among sellers 0.061
(0.315)
Information dissemination among buyers 0.228
(0.339)
First agreement 0.126
(0.261)
Market exogenous uncertainty -0.269
(0.231)
Seller dependence on partner 0.090
(0.274)
Buyer dependence on partner 0.386
(0.245)
Constant 3.372
(0.331)***
Observations 350
Clusters 218
Notes:
1) Clustered standard errors in parentheses, where each cluster is a firm.
* significant at 10%; ** significant at 5%; *** significant at 1%
2) F-test of excluded instruments for seller specific investment: F(2,217) = 5.32, p-value = 0.006
F-test of excluded instruments for buyer specific investment: F(2,216) = 5.27, p-value = 0.006
3) Wald test of equality of seller and buyer specific investment coefficients: p-value = 0.064
Table 1.8: Comparing the estimates obtained by different methods
Separate buyer and Regressions with both
seller regressions buyer and seller variables
OLS ML OLS IV
Seller specific investment 0.652 1.806 0.651 1.963
Buyer specific investment 0.276 -1.911 0.247 -2.022
82
Table 1.9: ML TE estimates with identification based only on non-linearities (I)
Main equation results First stage results
Seller contract Seller specific
Independent variables: complexity Instrumental variables: investment
Seller specific investment 1.627 Few potential buyers in heavy industry 0.785
(0.416)*** (0.280)***
Quality of the courts 0.295 Few potential buyers in light industry -0.075
(0.159)* (0.646)
Information dissemination among sellers 0.094 Few potential buyers in construction 1.032
(0.182) (0.373)***
First agreement 0.076 Few potential buyers in other sectors 0.660
(0.162) (0.347)*
Market exogenous uncertainty -0.049 Many potential buyers in heavy industry -0.014
(0.164) (0.217)
Seller dependence on partner 0.414 Many potential buyers in light industry -0.008
(0.145)*** (0.247)
Few potential buyers in heavy industry -0.224 Many potential buyers in construction 0.328
(0.347) (0.226)
Few potential buyers in light industry 0.095 Constant -0.936
(0.668) (0.162)***
Few potential buyers in construction 0.503
(0.403)
Few potential buyers in other sectors 0.340
(0.363)
Many potential buyers in heavy industry 0.409
(0.264)
Many potential buyers in light industry 0.215 Observations 373
(0.260) Clusters 227
Many potential buyers in construction 0.595 ? (correlation between the error terms) -0.409
(0.229)*** (0.118)***
Constant 2.349 Log-likelihood -848.698
(0.280)***
Clustered standard errors displayed in parentheses, where each cluster is a firm. * significant at 10%; ** significant at 5%; *** significant at 1%
83
Table 1.10: ML TE estimates with identification based only on non-linearities (II)
Main equation results First stage results
Buyer contract Buyer specific
Independent variables: complexity Instrumental variables: investment
Buyer specific investment -1.730 Few potential sellers in heavy industry 0.017
(0.461)*** (0.261)
Quality of the courts 0.349 Few potential sellers in light industry -0.396
(0.163)** (0.329)
Information dissemination among buyers 0.255 Few potential sellers in construction -0.775
(0.190) (0.433)*
First agreement 0.146 Few potential sellers in other sectors -0.491
(0.165) (0.306)
Market exogenous uncertainty -0.116 Many potential sellers in heavy industry -0.759
(0.174) (0.329)**
Buyer dependence on partner 0.424 Many potential sellers in light industry -0.314
(0.158)*** (0.310)
Few potential sellers in heavy industry 0.570 Many potential sellers in construction -0.348
(0.322)* (0.283)
Few potential sellers in light industry 0.366 Constant -0.879
(0.331) (0.191)***
Few potential sellers in construction 0.687
(0.360)*
Few potential sellers in other sectors 0.317
(0.336)
Many potential sellers in heavy industry 0.113
(0.400)
Many potential sellers in light industry 0.069 Observations 374
(0.353) Clusters 224
Many potential sellers in construction 0.701 ? (correlation between the error terms) 0.687
(0.295)** (0.132)***
Constant 2.579 Log-likelihood -789.412
(0.300)***
Clustered standard errors displayed in parentheses, where each cluster is a firm. * significant at 10%; ** significant at 5%; *** significant at 1%
84
Table 1.11: Heckman two-step estimates of the impact of specific investments
Dependent variable:
Seller Buyer
Independent variables: contract contract
complexity complexity
Seller specific investment 1.596
(0.792)**
Buyer specific investment -2.443
(1.501)*
Quality of the courts 0.324 0.326
(0.150)** (0.152)**
Information dissemination among sellers 0.080
(0.186)
Information dissemination among buyers 0.244
(0.204)
First agreement 0.122 0.187
(0.163) (0.169)
Market exogenous uncertainty -0.069 -0.149
(0.171) (0.174)
Seller dependence on partner 0.460
(0.158)***
Buyer dependence on partner 0.582
(0.159)***
? (inverse Mills ratio) -0.589 1.438
(0.467) (0.800)*
Constant 2.593 2.988
(0.265)*** (0.293)***
Observations 373 374
Standard errors displayed in parentheses.
* significant at 10%; ** significant at 5%; *** significant at 1%
85
Table 1.12: Robustness: Adding variables to the equation explaining seller contract complexity
Independent variables: Seller contract complexity
Seller specific investment 1.881 1.855 1.808 1.793 1.821 1.934
(0.509)*** (0.502)*** (0.539)*** (0.508)*** (0.530)*** (0.536)***
Quality of the courts 0.322 0.316 0.302 0.331 0.341 0.285
(0.157)** (0.157)** (0.157)* (0.158)** (0.160)** (0.154)*
Information dissemination among sellers 0.059 0.049 0.083 0.073 0.057 0.088
(0.182) (0.183) (0.180) (0.182) (0.181) (0.180)
First agreement 0.124 0.125 0.136 0.102 0.117 0.179
(0.171) (0.171) (0.169) (0.167) (0.169) (0.166)
Market exogenous uncertainty -0.082 -0.097 -0.075 -0.086 -0.055 -0.058
(0.166) (0.166) (0.166) (0.170) (0.168) (0.163)
Seller dependence on partner 0.433 0.434 0.444 0.430 0.445 0.465
(0.147)*** (0.146)*** (0.146)*** (0.148)*** (0.148)*** (0.147)***
State has control 0.467
(0.205)**
State owns shares 0.473
(0.192)**
Some state origins 0.334
(0.172)*
In same region 0.099
(0.151)
Firm founded before 1990 0.107
(0.165)
Large firm 0.511
(0.170)***
Constant 2.479 2.489 2.330 2.525 2.490 2.368
(0.232)*** (0.230)*** (0.261)*** (0.234)*** (0.257)*** (0.243)***
Observations 373 373 373 373 369 372
Clusters 227 227 227 227 225 226
? (correlation between the error terms) -0.515 -0.509 -0.480 -0.465 -0.475 -0.530
(0.146)*** (0.144)*** (0.161)** (0.152)*** (0.158)** (0.153)***
Clustered standard errors in parentheses, where each cluster is a firm. * significant at 10%; ** significant at 5%; *** significant at 1%
86
Table 1.13: Robustness: Adding variables to the equation explaining buyer contract complexity
Independent variables: Buyer contract complexity
Buyer specific investment -1.855 -1.833 -1.897 -1.821 -1.940 -1.956
(0.374)*** (0.383)*** (0.392)*** (0.412)*** (0.359)*** (0.367)***
Quality of the courts 0.355 0.354 0.347 0.380 0.375 0.323
(0.159)** (0.159)** (0.161)** (0.160)** (0.162)** (0.157)**
Information dissemination among buyers 0.218 0.204 0.244 0.231 0.235 0.222
(0.190) (0.190) (0.187) (0.193) (0.188) (0.189)
First agreement 0.191 0.187 0.187 0.168 0.192 0.255
(0.171) (0.170) (0.169) (0.167) (0.168) (0.166)
Market exogenous uncertainty -0.180 -0.187 -0.159 -0.176 -0.142 -0.162
(0.168) (0.168) (0.168) (0.170) (0.168) (0.165)
Buyer dependence on partner 0.507 0.514 0.516 0.544 0.518 0.495
(0.152)*** (0.151)*** (0.150)*** (0.151)*** (0.153)*** (0.151)***
State has control 0.442
(0.209)**
State owns shares 0.466
(0.191)**
Some state origins 0.185
(0.176)
In same region 0.178
(0.150)
Firm founded before 1990 0.157
(0.164)
Large firm 0.529
(0.159)***
Constant 2.886 2.880 2.812 2.837 2.839 2.816
(0.199)*** (0.199)*** (0.236)*** (0.205)*** (0.223)*** (0.200)***
Observations 374 374 374 374 370 373
Clusters 224 224 224 224 222 223
? (correlation between the error terms) 0.719 0.713 0.720 0.707 0.736 0.747
(0.095)*** (0.098)*** (0.101)*** (0.111)*** (0.091)*** (0.088)***
Clustered standard errors in parentheses, where each cluster is a firm. * significant at 10%; ** significant at 5%; *** significant at 1%
87
Table 1.14: ML TE estimates when the sample is split by significant prepayments
No significant buyer prepayment Significant buyer prepayment
Independent variables: Seller contract Buyer contract Seller contract Buyer contract
complexity complexity complexity complexity
Seller specific investment 2.778 -1.725
(1.001)*** (0.374)***
Buyer specific investment -2.135 -1.403
(0.512)*** (0.622)***
Quality of the courts 0.341 0.345 0.335 0.545
(0.211) (0.206)* (0.198)* (0.218)**
Information dissemination among sellers 0.088 0.006
(0.243) (0.243)
Information dissemination among buyers 0.339 -0.090
(0.243) (0.229)
First agreement 0.052 0.161 0.094 0.068
(0.228) (0.225) (0.182) (0.213)
Market exogenous uncertainty -0.163 -0.319 -0.133 0.146
(0.217) (0.223) (0.211) (0.190)
Seller dependence on partner 0.341 0.260
(0.236) (0.207)
Buyer dependence on partner 0.424 0.673
(0.192)** (0.207)***
Constant 2.354 2.817 3.967 3.308
(0.304)*** (0.249)*** (0.350)*** (0.251)***
Observations 248 246 125 126
Clusters 185 184 104 104
? (correlation between the error terms) -0.739 0.796 0.938 0.586
(0.197)** (0.098)*** (0.038)*** (0.279)
Clustered standard errors in parentheses, where each cluster is a firm. * significant at 10%; ** significant at 5%; *** significant at 1%
88
Chapter 2
A Proposed Way to Measure Transaction Costs Using Survey Data on
Sales Agreements
Abstract
The measurement of transaction costs, in particular at transaction level, has long been
an important goal. This chapter of the dissertation presents a simple methodology
for doing so, using reports from business officials who supervise companies? buying
and selling activities. These are the economic agents who would possess quantitative
knowledge of transaction costs at agreement level if such costs affected decisions. In
a practical implementation in Romania, those transaction costs directly related to
the buying and selling activities are assessed as large, accounting for more than 20%
of value added. The recorded transaction costs estimates correlate significantly with
variables suggested by theory, indicating validity. The quality of the data collected
by the proposed survey question is also investigated. Potential upward or downward
biases at respondent level, arising from a different understanding of the question, and
potential biases at aggregate level, arising mainly from sample selection, are shown
to not affect the transaction costs estimate.
89
2.1 Introduction
This chapter?s goals are to describe a methodology for measurement of trans-
action costs and to present an example of this method?s implementation, thus filling
a gap in the existing empirical literature. Part of the new institutional economics,
transaction cost economics (TCE) abounds with definitions of the ?transaction costs?
concept and with hypotheses on which institutions firms rely on in order to reduce
such costs. By contrast, the measurement of transaction costs is still in its infancy
(Allen, 2000; Benham and Benham, 2000; Wang, 2003).
The existing studies focusing on transaction costs? measurement generally fall
into two categories. Some, such as Wallis and North (1986), examine the economy
at the aggregate level, assessing the magnitude of those activities specifically focused
on transactions. Others, such as Kuperan et al. (1998) and Lesmond et al. (1999),
analyze niche sectors where transaction costs can be isolated from production and
transportation costs. The literature reveals that there is still no accepted method
of assessing how large transaction costs are in the operations of the broad spectrum
of producers, where transactional and productive activities are inextricably bound.
In this specific case, the following commentary seems all too relevant: ?The main
obstacle faced by researchers in industrial organization is the lack of available data
on contracts and the activities of firms? Coase (1992, p. 718)
The measurement of transaction costs is unusually difficult for many reasons
? Benham and Benham (2000) discuss why better measures do not currently exist.
In particular, when these costs are explicitly incurred, they are usually mixed with
90
other costs from which they cannot be separated, hence accounting statements do
not properly identify them. More frequently, transaction costs are hidden, reflecting
missed opportunities (e.g. investments not implemented because of lack of confidence
in trading partners).
Researchers have long recognized the difficulty of consistently measuring trans-
action costs across firms, which adds to the difficulty of obtaining transaction data in
general. Hence, empirical studies have relied on imperfect proxies instead and have
focused on specialized markets subject to government scrutiny (which includes the
filing of contracts). According to Chiappori and Salani?e (2002), these markets tend to
be fairly thin and are characterized by large appropriable quasi-rents1, which leaves
doubts about the generality of the results.
Transaction costs? implicit nature indicates that many relevant facts reside only
in the memories of company officials. But a central tenet of economics is that eco-
nomic agents act on the basis of such information. If such costs do help to determine
behavior, company officials must have some idea on their magnitude. Why not di-
rectly ask them about transaction costs? size?2
We propose a new way to measure transaction costs, by relying on survey ques-
tionnaires. While the actual question will be presented and discussed in detail in
a subsequent section, we mention here its essence: interviewed sales and procure-
1For an extensive discussion on appropriable quasi-rents see Klein et al. (1978).
2Blinder (1994, p. 118): ?the theories share one aspect in common: each traces a chain of
reasoning that allegedly leads the decision maker to conclude that [doing some particular thing
is in the] firm?s best interests. It struck me that if a particular [decision-maker] actually follows
such a chain of reasoning, he might just know it.? Hendley and Murrell (2003) follow this logic in
investigating the mixture of dispute-resolution mechanisms used by firms.
91
ment managers were asked to choose one particular agreement their company was
part of and, after answering a large set of questions about that agreement, they were
questioned how much it would have been worth to reverse history and instead have
reached an agreement with a non-opportunistic partner (one who would share busi-
ness information, would keep his promises, and would negotiate equitably to solve
problems). Alternatively, our question can be reinterpreted as asking how much the
respondent would be willing to pay for what first-best institutions could do. While it
is commonly accepted in the literature that transaction costs arise from two human
features ? bounded rationality and opportunism ? the survey question proposed in
this chapter places more weight on the latter.
We realize that the information extracted from company officials will be deemed
imprecise and subjective. Nonetheless, information is obtained from exactly those
respondents assumed to possess it and, given the present extent of ignorance in the
field, even imprecise estimates would add to the sum of knowledge3. Also, despite
being subjective, the data may still be meaningful. Researchers in other social science
fields have also relied on survey questions to collect highly subjective information, and
the collected data have been shown to possess appropriate features. For example,
information on happiness and well-being is gathered through survey questions ? see
http://www1.eur.nl/fsw/happiness/ and Kahneman and Krueger (2006). Di Tella
and MacCulloch (2006) defend the measures of happiness thus gathered and their use.
3A simple though experiment might persuade the reader. Suppose that accountants could obtain
perfect data on the transaction costs incurred in trading the typical goods and services offered by
a typical firm in a median income country. Would these costs be 0.5%, 5%, 25%, or 50% of value
added? An extensive search of existing literature suggests there is no basis for picking between these
numbers.
92
The proposed method provides a coherent and practical procedure for mea-
surement of a phenomenon of central interest to economists, and we show that the
transaction costs measure we obtain has appealing characteristics. First, since re-
spondents are required to express their answers as a percentage of the sale price, the
estimates we collect are comparable across sectors of activity, regions, and countries.
Second, we show that the survey question is not capturing noise but relevant infor-
mation, and for this we display correlates of our measure with many firm features
and factors theory predicts to affect transaction costs. Lastly, we analyze upward
and downward biases at respondent level as well as biases at aggregate level and show
that our data is not affected by such problems.
Despite only focusing on part of the concept (inter-firm transactional impedi-
ments are only part of all transaction costs firms incur in their activity), the present
dissertation chapter takes a much needed first step toward a consistent estimation of
transaction costs at firm level for a cross-section of business activities. The imple-
mentation of the proposed method indicates that, at least in Romania, transaction
costs directly related to buying and selling goods and services are large, accounting
for more than 20% of value added.
This chapter of the dissertation proceeds as follows. Section 2.2 presents some
of the various definitions the ?transaction costs? concept has received in the liter-
ature, the sources of these costs, and the efforts that have been made to measure
this concept. Section 2.3 describes the proposed methodology in the context of a
sample application in Romania. Section 2.4 discusses the size of transaction costs
93
associated with exchange activities, and also debates on the validity of the proposed
methodology. Section 2.5 analyzes the quality of our transaction costs measure by
investigating various potential biases that may affect the proposed methodology, both
at respondent and at aggregate level. Section 2.6 concludes the chapter.
2.2 Transaction costs: notion and difficulty of measurement
Simply put, the transaction costs paradox is that the wealth of definitions makes
the concept less defined: TCE abounds with proposed definitions, the multitude of
which leads to an abundance of forms transaction costs are considered to take, and
ultimately to a broad and rather fuzzy concept. Consequently, one reason this concept
is not consistently measured may reside in the lack of an unitary approach.
Since a review of the transaction costs literature exceeds the goals of this dis-
sertation chapter, we will limit below to only a few definitions and some other aspects
related to transaction costs, with a focus on existing measures of the concept. For
two extensive expositions on the history, use, and meaning of the term the reader is
guided toward Allen (2000) and Klaes (2000).
Arrow (1969, p. 48) referred to transaction costs as to ?the costs of running
the economic system?, while Cheung (1998, p. 515) called them ?the costs which
do not exist in a Robinson Crusoe economy?. The broad ?transaction costs? notion
covers all those expenses associated with an exchange, which do not directly stem
from the physical processes of production or transportation. As mentioned in the
introduction, this chapter?s main goal is to describe a methodology for measurement
94
of transaction costs at microeconomic level, hence we will adhere to the slightly more
precise definition presented by Masten (1996) and refer to transaction costs as to
those costs arising from impediments to reaching and enforcing agreements.
In an attempt to categorize transaction costs, North and Thomas (1973, p. 93)
consider them to take three forms: search costs (costs of locating information about
opportunities for exchange), costs of negotiating the terms of the exchange, and costs
of enforcing the contract. Furubotn and Richter (1997, p. 40) believe transaction
costs to include ?the costs of defining and measuring resources or claims, the costs of
utilizing and enforcing the rights specified, and the costs of information, negotiation,
and enforcement?. In a similar vein, Masten (1996, p. 6) argues that transaction
costs ?are associated with such activities as bargaining, contracting, and monitoring
performance, activities that are not directly productive but which are engaged in only
as a consequence of the need to coordinate activities among transactors.?
Regarding transaction costs? sources, Macher and Richman (2006) consider
these costs to stem from individuals? limited ability to accurately predict and plan
for all the various contingencies that may arise, from parties? difficulty to negotiate
and develop a common language to describe situations with which they have little
prior experience, and from the obstacles parties face to communicate their plans so
that an uninformed party, such as a court, could reasonably enforce them. Referring
to work of Oliver E. Williamson (the main contributor to the TCE field) and to other
authors, Scott E. Masten states that the existence of transaction costs ca be traced
95
to two attributes of human nature: bounded rationality and opportunism4.
Given their broad definition, wide array of forms, and significant impact on
the decisions of economic agents, researchers were eager to assess the importance of
transaction costs as a share of economy?s expenses. Yet, there are few estimates of
the size of transaction costs at macroeconomic level. Wallis and North (1986) have
estimated that transaction costs accounted for 26.1% of U.S. GDP5 in 1870, a share
which has constantly increased to reach 54.7% in 1970 (the authors also maintain that
throughout time an overwhelming proportion of these costs comes from the private
sector). Steven N.S. Cheung provides a rough estimate of transaction costs level for
1998 Hong Kong ? at least 80% of GDP ? and argues that ?in the modern world, it
would be difficult to find a rich country where transaction costs sum to less than half
of national income? Cheung (1998, p. 515).
Despite the importance of the transaction costs notion, empirical studies have
long been hindered by lack of micro-level data since economists have found difficult to
consistently measure this concept at firm or transaction level. Allen (2000) and Wang
(2003) have expressed concern in this respect, while Benham and Benham (2000) have
4Masten (1996, p. 6): ?Bounded rationality refers to fact that, although individuals intend to act
in a rational manner, their ability to realize those intentions is hampered by their limited knowledge,
foresight, skill, and time. [...] Opportunism refers to the willingness of transactors to renege on
promises, cheat on agreements, shirk responsibilities, circumvent rules, search out loopholes, or
otherwise exploit the vulnerabilities of a trading partner in hopes of eliciting a more favorable
distribution of the rents accruing to exchange.?
5Wallis and North (1986, p. 121): ?The method used to calculate [the transaction costs estimate]
is roughly this: For each sector in the national account the number of workers in different occupations
and a percentage of transactional work in each occupation is estimated. Then the salaries of these
workers have been used to calculate the transactional part of the GDP in each sector. The weighted
average of these figures is [the transaction costs estimate]. To illustrate, the following occupations
have been deemed to be 100% transactional: accountants, lawyers and judges, personal and labor
relations, farm managers, managers, clerical, sales workers, foremen, inspectors, guards, police,
military, and postal service. Industries with many non-transactional jobs are: agriculture, mining,
construction, manufacturing, and yes even transportation.?
96
presented several reasons why few empirical estimates of transaction costs exist (of
which, there are very few comparative estimates). Yet, some breakthroughs have
been made in narrow sectors of activity, where researchers have found easier to sepa-
rate transaction costs from the costs of production and transportation. For instance,
Kuperan et al. (1998) have measured transaction costs of fisheries co-management in
San Salvador Island, Philippines, Lesmond et al. (1999) have estimated transaction
costs on financial markets, while Dietrich (2003) has provided an innovative approach
by estimating production functions and relying on specific functional forms to isolate
transaction costs from production costs. Unfortunately, such attempts represent ex-
ceptions in the field, since many researchers have considered the direct measurement
of transaction costs to be an unachievable goal, and have instead relied on the use of
proxies. The appropriateness of the chosen proxies has always been subject to debate,
a view expressed by many researchers, such as Shelanski and Klein (1995), Chiappori
and Salani?e (2002), and Masten and Saussier (2002).
In his 1998 Presidential Address to the Western Economic Association, Steven
N.S. Cheung has forcefully pushed for a change of attitude in this area of research:
?It has been argued that it is fruitless to study transaction costs, because it is fre-
quently impossible to measure them. This view is wrong. Fundamentally, measure-
ment involves an assignment of numbers for the purposes of ranking, and precision
in measurement can only be judged by the extent of agreement among different ob-
servers. To say that cost is measurable, or measurable precisely, does not necessarily
mean it is measurable in dollars and cents. If we are able to say, ceteris paribus,
97
that a particular type of transaction cost is higher in Situation A than in Situation
B, and that different individuals consistently specify the same ranking whenever the
two situations are observed, it would follow that transaction costs are measurable,
at least at the margin. Testable propositions may then be obtained, and that is the
important thing.? Cheung (1998, p. 516)
Following the comments above, we propose a new way to measure (part of)
transaction costs which arise at agreement level, by relying on a survey question
addressed to company officials who supervise the buying and selling activities. The
next section presents our proposed survey question.
2.3 A new way to measure transaction costs
Probably chief among difficulties researchers face in their attempt to measure
transaction costs is the predominant implicit nature of the concept. That is, many
of the transaction costs components are not directly incurred (are hidden) or are in-
curred along with costs of other activities, and from which cannot be disentangled.
For example, costs from missed trade and investments due to lack of confidence in
potential partners are real expenses for which records can hardly be kept. On the
other hand, search costs for locating potential trading partners or payments for con-
sulting services are explicit costs. However, for those costs incurred at firm level,
it is frequently impossible in practice to clearly assess which parts of the expenses
relate to which precise transactions. To conclude, even when some transaction costs
are explicitly incurred, one?s attempt to investigate companies? accounting records to
98
obtain a transaction costs estimates may prove futile.
Despite their implicit nature, what matters for us is that information on transac-
tion costs is available in the minds of company officials, and it is of prime importance
that decision-makers use such information in conducting their business. Then, it
seems natural to try and inquire company officials about the size of transaction costs
faced in the course of their activity.
Our data are collected by means of a questionnaire. We acknowledge the fact
that relying on survey instruments and respondents? memory regarding past trans-
actions will lead to imprecise information. Nonetheless, given the above-mentioned
paucity of attempts to measure the concept, we believe that any micro-level, consis-
tent transaction costs estimate obtained for a cross-section of sectors of activity will
represent a valuable addition to the sum of knowledge in the field.
Before discussing the actual question soliciting information on transaction costs,
let us briefly present some details on the study which incorporated the question. In
the middle of 2001, two hundred and fifty-four Romanian companies in country?s
largest twelve cities were surveyed in an attempt to understand the role of formal
legal system in firms? activity. Given the goal of the study, only medium and large
companies were included in the sample, but they covered a wide range of sectors of
activity and represented various types of ownership structures. Four decision-makers
were questioned in separate face-to-face interviews in each company: the company?s
general manager, the head of the legal department (when the company did not have
such a department established, the lawyer most frequently dealing with the company
99
was questioned instead), and the managers of the sales and procurement departments.
The survey instruments addressed to the last two respondents contain a large
section soliciting information on a specific agreement the company has been involved
into (once as a seller, and once as a buyer)6. This agreement is one the respondent has
chosen, which could have been written or oral, successful or unsuccessful, regarding
the sale/acquisition of a new or traditional product. Our only constraints on the
choice of agreement were that it provided for the exchange to have taken place within
the last six months, and that the respondent had to be thoroughly familiar with the
agreement and its implementation. Boxes 2.1 and 2.2 displayed at the end of this
chapter contain the instructions for sales and procurement managers on how to choose
the specific agreement.
The survey instruments addressed to the sales and procurement managers are
almost identical, the only difference residing in questions? point of view: while sales
managers were asked about companies? activity as sellers, procurement managers were
questioned about companies? behavior as buyers. In the first chapter of the disser-
tation, this crucial feature was called symmetry between the two survey instruments.
The first chapter has also pointed out another important feature of the questionnaires
? the symmetry within the survey instruments addressed to the sales and procurement
managers. That is, apart from being questioned on particular aspects on their side
of the specific agreement, our respondents are asked about similar aspects on their
partners? side. As just one example of the two-fold symmetry, sales and procurement
6The section regarding the specific agreement is the largest of the questionnaire and includes
sixty-three different questions on various features of this agreement.
100
managers were asked whether their company has requested the renegotiation of the
specific agreement during its implementation. Also, the same respondents were asked
whether their partners have asked for renegotiation of the agreement. Following the
same approach as in the first chapter, the two features above allow us to construct a
larger dataset of agreements. Even though we have interviewed two hundred and fifty-
four companies, we have information on both sides of five hundred eight particular
agreements these companies have entered into.
?Transaction costs? is not a term of art in the Romanian business world. There-
fore, obtaining the pertinent information necessitated a question with wording to
which the respondent could relate to, but which would be driven by economic con-
cepts. The compromise entailed a focus on opportunism, which is central in all dis-
cussions of transaction costs. The question asks the respondent how much it would
have been worth to reverse history and instead have reached an agreement with a
non-opportunistic partner. Alternatively, it can be reinterpreted as asking how much
the respondent would be willing to pay for what first-best institutions could do.
With slight customizations depending on respondent (see below), the question
used to capture the transaction costs information was placed at the end of the section
soliciting information on the specific agreement. It is crucial to note that the transac-
tion costs question was asked in the context of a particular agreement, on which the
sales and procurement managers had already answered many questions. It was asked
only after respondents had recalled many of the details of the specific agreement, to
ensure informative focus on problems and successes. Boxes 2.3 and 2.4 displayed at
101
the end of the chapter show the questions addressed to the two managers, and the
conjecture is that if firms do make decisions based on transaction costs, then sales
and procurement managers can provide the elicited estimates.
The two questions are highly similar and the only difference resides in their
point of view: one is addressed to a seller and one to a buyer. Therefore, for brevity,
we will further comment on that addressed to the sales manager (when the company
acted as a seller in the specific transaction). Its essence is as follows: what percentage
of the sale price would the firm be willing to pay to reverse history and deal with a
different customer, an imaginary one, who would have the same characteristics as the
actual one, but would be frank, trustworthy, and fair.
It should be noted that negative answers were not allowed. We believe that this
restriction guides respondents to only think of their possible reduction of costs when
dealing with an ideal partner, and not of the possibility of taking advantage of the
desirable qualities of such a partner. This particular aspect of the question will be
further discussed later in the chapter.
The survey question we propose obviously causes respondents to focus most
closely on the costs of haggling, bargaining, and mis-alignment emphasized especially
by Williamson (1985), but also present in agency and contract theories (Holmstr?om,
1999; Salani?e, 1997). The reason for posing a question with this focus is that respon-
dents can only be expected to report on activities within their domain of competence.
For the sales and procurement managers, this means the costs incurred after commit-
ments had been made to a specific line of production activity and after basic decisions
102
on investment had been taken. They are the costs incurred as the firm chooses its
trading partners, negotiates agreements, and proceeds from agreement to implemen-
tation. Since sales and procurement managers would probably not fully appreciate
the costs of the investment distortions emphasized in incomplete contract theory, see
Grossman and Hart (1986), the information gathered probably does not capture such
costs completely.
2.4 The size and correlates of transaction costs
2.4.1 The size of transaction costs
After discussing the question soliciting the transaction costs information, we
present below data on the magnitude of the phenomenon, as captured by our survey
of Romanian companies.
Figure 2.1 at the end of the chapter displays the relevant information. The
transaction costs responses we record are presented by means of three histograms:
one for the answers provided by sales managers, one for those provided by procure-
ment managers, and one for the data we obtain from both these sources. The figure
indicates that many transactions go smoothly: 127 agreements (50.80%) in which re-
sponding firms have acted as sellers are characterized by zero transaction costs, and
the figures are higher (164 agreements, representing 65.34%) when the responding
firms were buyers in the specific agreement. Overall (when the responding firms were
either sellers or buyers in the specific agreement), no transaction costs are incurred
103
in 291 agreements, representing 58.08% of the sample. Figure 2.1 also shows that
transaction costs range from zero to a staggering 60% of the sale price, with many of
the positive answers concentrated at the 5 and 10% values.
Table 2.1 at the end of this chapter uses the same data. The second row indicates
that the average transaction costs as percentage of sale price, when the mean in
computed for all agreements (with either positive or zero transaction costs), is 3.87%
when the responding firms were sellers in the specific agreement, 3.11% when they
were buyers, and 3.49% overall. Obviously, these means are higher when computed
for only those agreements with positive transaction costs ? the third row of Table 2.1
presents the relevant figures.
Are these figures suggestive of the true significance of transaction costs for
Romanian companies? To appreciate the magnitude of these costs more fully, they
must be placed in the context of the firm?s operations as a whole. Firms incur costs
as both a buyer and a seller in a transaction: the two should be aggregated. Then,
assuming that transaction costs of purchasing labor and capital are commensurate
with those incurred in purchasing intermediate inputs, and that profit is a negligible
component of the value added, the transaction costs of a typical firm equal 6.98%
of revenues. Moreover, comparing costs to prices is not the best guide to economic
significance, even though it is the most useful approach for a survey question. The
firm?s production is best measured by value added, and transaction costs are incurred
in producing this. A reasonable estimate would place value added at 30% of sales7.
7In Romania, firms are guarded on financial issues and are anxious to hide information pertinent
to taxes (the value added tax has a rate of 19%). Survey information on the value added to sales
104
We need to multiply the previous results by a factor of 3.33. This way we obtain
that the mean transaction costs (when we consider all agreements, with positive or
zero reported transaction costs) as a share of firm?s output are 12.90% for sellers and
10.37% for buyers, totaling an impressive 23.27% for the company as a whole.
Some points need to be considered. First, the figures above are mere estimates,
and rest on some precise assumptions to provide a realistic image of the investigated
phenomenon. If we alter one of those assumptions and consider the ideal situation of
zero transaction costs in purchasing labor and capital (while using the same estimate
of value added, 30% of sales), computations indicate that transaction costs equal
6.05% of revenues, or 20.16% of value added. These lower bounds are themselves
a good indicator on the importance of transaction costs at firm level. Second, one
should recall that we have restricted attention to the costs of haggling, bargaining, and
mis-alignment, which are only part of the transaction costs incurred in agreements
between firms. Third, we have to keep in mind that transaction costs may also be
present within firms, as integration reduces but may not completely eliminate this
problem (Gibbons, 2005, p. 220). Given all these, we must conclude that Romanian
enterprises do face large transaction costs in the course of their activity.
2.4.2 The correlates of transaction costs
After presenting the impressive size of transaction costs in basic exchange agree-
ments, the next step is to show that the question soliciting the transaction costs infor-
ratio was unreliable. Value added as a percentage of sales in the 800 largest UK companies is 30.1%
and 32.8% in the 600 largest European companies (Department of Trade and Indsutry, 2004).
105
mation is in fact capturing the sought concept. Validity can be indirectly assessed by
examining whether our transaction costs measure correlates with variables suggested
by the theoretical literature. If there is a number of such correlations that are sta-
tistically significant, then this suggests that the estimates represent something very
real. Nevertheless, one should not expect significant correlations with all variables
suggested by theory as some theories might not be applicable in Romania or might
even be incorrect.
The questionnaires elicit information on characteristics of the firm and specific
agreement, and twenty-three of these features are chosen for analysis in the present
context. The use of some simply follows common suppositions (for example, larger
firms are different from smaller ones). Some reflect the peculiar history of Romania
and transition (for example, firms founded after 1989 behave differently from the
firms inherited from communism). But most are based on standard predictions from
commonly accepted economic theories (for example, agreements are easier to con-
struct for the sale of standardized products rather than for goods subject to quality
variation, the size of transaction costs correlates with market uncertainty and with
the frequency partners transact).
For each characteristic separately, the sample of agreements was divided into
two groups, one with the characteristic and one without. Then the difference between
the transaction costs means of the two sub-samples was examined. Some details are
worth noting. First, since some variables were categorical and some continuous, this
uniform procedure allowed symmetrical treatment of all variables. The second note
106
is related to the way we use our data, and is particularly useful to remember that
we investigate two agreements from each surveyed company. When our sample of
agreements is split into two groups (one with and one without a certain characteristic),
it is possible that one agreement the firm has entered into has the characteristic (hence
belongs to one group), while the other agreement does not have it (hence belongs to
the other group). Then, the statistical tests need to take into account the fact that
errors might be correlated across the two agreements of one firm.
Table 2.2 at the end of the chapter presents the basic results. They are not
intended to portray particular causal mechanisms, but rather correlations between
variables that would naturally occur under some theory about the process that gen-
erates transaction costs. Results indicate that higher transaction costs occur in newer
firms, when exogenous uncertainty is a very important factor, when exchanges occur
less frequently, and when the transacted good is more complex. Table 2.2 also in-
dicates that lower transaction costs characterize those agreements in which larger
companies are involved.
In addition, transaction costs are related to firms? decisions tocease old activities
or start new ones. Namely, lower transaction costs are reported when one of the firms
has abandoned some operations in the last two years (since those activities were
probably generating such costs), and higher transaction costs are observed when one
of the firms has recently invested in building or equipment (since these investments
are probably associated with new activities and the firm may lack the experience in
conducting such new operations). Whether one of the firms in the specific agreement
107
has state origins seems to have a limited impact on reported transaction costs.
Regarding firms? ownership structure, despite their low statistical significance,
results in Table 2.2 tend to indicate that state ownership is associated with lower
transaction costs, while employees ownership is associated with higher levels of such
costs. Last but not least, when distinction between buyers and sellers is made, results
seem to confirm the first chapter?s findings, which argue that sellers are more exposed
to buyers? opportunism. For example, the mean transaction costs is significantly
higher when the seller makes a relationship-specific investment than when the seller
does not undertake such an investment, which is not the case for buyers.
To conclude, there are no contradictory findings: the direction of the relation-
ship between our transaction costs measure and the characteristics does align with
standard assumptions. Also, eight of the twenty-three test statistics are significant at
the 10% level (of these, five are significant at the 5% level). Therefore, in sum, Table
2.2 presents a strong case that the empirical measure corresponds to what scholars
have in mind when using the ?transaction costs? concept to understand the nature
of contract and firm behavior.
2.5 The quality of our transaction costs measure
In this section we investigate biases that may affect our transaction costs mea-
sure, and which may be encountered at two different levels:
? respondent level bias ? arises if the interviewed managers have understood and
answered our question in a manner different from our intentions. If so, then
108
some respondents? answers may constitute under-estimates, while others may be
over-estimates of the true level of transaction costs. Even though these potential
biases may cancel out at aggregate level, an unbiased transaction costs measure
at respondent level would be ideal since it could be further used in agreement-
level regressions.
? aggregate level bias ? the overall picture we get on transaction costs may be
different from reality even if our respondents have provided accurate estimates
of the concept. Mainly, our data may be ridden by a sample selection bias given
the non-random way investigated agreements have entered our sample. Apart
from this, other sources of aggregate level bias will be discussed below.
2.5.1 Potential biases at respondent level
We begin by commenting the possibility of bias at respondent level, that is, the
situation in which the interviewed managers have understood our question differently
from our intentions, and consequently have provided under- or over-estimates of the
transaction costs actually incurred in the specific agreement. This discussion requires
us to present an additional theoretical argument regarding transaction costs.
Section 2.2 has tried to shed some light on the various definitions the ?transac-
tion costs? concept has received in the literature. Despite the impressive number of
definitions, it is commonly agreed that transaction costs arise because of individuals?
bounded rationality and opportunism. The latter refers to any misbehavior that ap-
pears in the course of a transaction when partners lack outside options, is triggered by
109
the existence of appropriable quasi-rents, and frequently takes the form of hold-up8.
In general, one partner?s opportunism is considered to lead to the renegotiation
of the agreement. There is a multitude of new terms the opportunistic partner may
impose on the exposed one, ranging from better trading conditions to in-kind or
monetary transfers. For simplicity, we?ll use the term ?transfer? to denote any kind
of amenity the opportunistic partner may obtain. According to theory, the distortions
these transfers may lead to, and the negotiation costs incurred to reach the settlement,
are included in the ?transaction costs? concept, but the transfers themselves are not
(one reason for this is that transfers are not in fact costs to society but can be seen
as monetary amounts exchanged between trading partners, whose sum at aggregate
level is zero).
The fine distinction between transfers and transaction costs is what may cause
our measure of the latter to be biased: respondents who were held-up may include
transfers to their opportunistic partners, hence biasing the transaction costs response
upward, while opportunistic respondents may provide downward biased estimates by
including transfers from their partners.
2.5.1.1 A potential upward bias at respondent level
We now discuss the possibility that our respondents have provided transaction
costs estimates which are upward biased relative to the true level, by including the
8Rogerson (1992) defines hold-up as the ex post request for renegotiation of an incomplete contract
when assets are specific. Hold-up is the commonly discussed form of opportunism, but is only one
of the many opportunistic measures a transacting partner may adopt ? see Masten (1996) for other.
110
transfers they may have made to their opportunistic partners into the transaction
costs response. We will discuss below two possible scenarios:
First scenario
For clarity, in what follows we?ll continue to comment on the case of our re-
sponding firm acting as the seller in the agreement (the other situation is similar).
The transaction costs question, as presented in Box 2.3 at the end of the chapter,
asks respondents to think back to the time ?before your enterprise began negotiating
this specific agreement?. Those who closely follow this instruction should have in
mind what they have thought of the agreement and their partners at that initial
time, regardless whether those expectations proved to be accurate or not during
agreement?s implementation. The following equality should then be in place for such
respondents:
Expected per unit profit = X1 ?Y1 ?Z1,
where X1 denotes the expected agreement price (reflects seller?s expectations on what
the sale price will end up being after initial negotiations with its partner), Y1 rep-
resents the expected per unit production costs (reflects seller?s initial expectations
on the usual production costs associated with the transacted good), and Z1 denotes
the expected per unit transaction costs (reflects seller?s expectations on those costs
incurred because of any undesirable features the buyer may be assumed to have).
In this setup, (expected) transfers do not appear as a distinct element in the
equality. This rests on the assumption that, at the initial stage of the agreement,
when forming its expectation about the agreement price, the respondent takes into
111
account the potential troubles to reach that price. Those troubles translate into
transaction costs and transfers. While transaction costs are captured by Z1, transfers
are encapsulated into X1 (which does not represent the price the seller initially asks
for, but rather his expectation on what the agreement price will end up being, thus
including the foreseen transfers).
Having in mind the initial stage, respondents should then consider they could
have faced another customer, who had the same features as those expected from the
actual partner, only that the imaginary one would be frank, trustworthy, and fair.
These three features may differ from what was expected from the actual partner, but
are not likely to impact on Y1 (expected per unit production costs remain the same
for the imaginary and actual partner). However, since these features could lower Z1
(expected per unit transaction costs), a lower price may be adopted without affecting
the expected per unit profit. This is exactly what our question tries to measure
and, under this scenario, our question is likely to accurately capture the expected
transaction costs. As we will see below, some respondents may have thought of a
different situation.
Second scenario
The second part of the question, especially the use of words profiting (without
a direct mentioning of expected profit) and of syntagm ?the actual one used? when
referring to price, may have induced respondents to think of the situation at the end
of agreement (at which stage the actual price used in transaction and the realized
profit are known).
112
We have to consider this scenario since this is the latest information available
to our respondents, so they may consider it when answering our question. Moreover,
the tendency of our respondents was to choose specific agreements which have been
completed: for 94.69% of agreements sellers have delivered the goods and buyers have
paid in full by the time of the interview.
If this is the case, then transfers may have been included in our respondents?
answers given that at the end of the agreement the following equality is in place:
Realized per unit profit = (X2 ?T)?Y2 ?Z2,
where X2 denotes the price agreed upon at the beginning of transaction, T represents
potential unexpected transfers to the transacting partner, Y2 denotes the realized
per unit production costs, and Z2 represents the realized per unit transaction costs.
Hence, whereas X1, Y1, and Z1 denote the expectations at the beginning of the
agreement regarding the per unit agreed price, production costs, and transaction
costs, X2, Y2, and Z2 denote the realized values.
We continue to assume that per unit production costs (Y2) are not affected by
those three undesirable features the partner may possess. If the partner (the buyer,
in the investigated situation) behaves opportunistically in the agreement, exceeding
seller?s predictions, then not only Z2 > Z1, but also the term T appears, lowering the
agreed upon price. In this situation, the actual price used in the transaction is X2?T,
and the realized per unit profit is lower than the expected one. This is where the
113
fine distinction between transaction costs and transfers becomes important. When
respondents have in mind the change in the actual price which would keep the realized
per unit profit at the same level they may include both Z2 and T in their answers.
This would lead to an upward bias in the transaction costs responses provided by
such respondents.
The question presented in Box 2.3 is surely the most elaborated in the survey
instrument, and yet not as detailed as it should have been in order to avoid any
possible confusion9. Still, the reader has to bear in mind that this question is just
one of the approximately one hundred questions included in each survey instruments.
Since these exhaustive questionnaires were addressed to the managers of sales and
procurement departments, they had to be kept as short as possible.
Given that the second scenario is plausible, we need to take into account and
analyze the possibility that our transaction costs estimate are biased upward.
Empirical investigations of the second scenario
As mentioned above, biases may arise only by inclusion of transfers into respon-
dents? transaction costs estimates. But transfers are made only following partner?s
opportunistic behavior, and the most commonly discussed form of opportunism is
hold-up, which entails a request to change the terms of the initial agreement (a rene-
gotiation). The question below has been addressed to the sales and procurement
department managers:
9A comprehensive question soliciting transaction costs information, which, on the one hand,
explains that transfers are not part of transaction costs, and which, on the other hand, makes a
clear distinction between close notions such as initially expected and realized profits, and expected,
initial, and actual price, can be constructed, but it would take the form of a questionnaire by itself.
114
Did your partner ask for the specific agreement to be
renegotiated during implementation? Yes/No
Ninety-three managers (18.31% of the sample) have answered ?Yes? to the ques-
tion above, and within these responses the transaction costs estimate was zero for
forty-five agreements10. Since four other observations have a missing value reported
for transaction costs, the remaining forty-four observations represent agreements in
which the partner has asked for renegotiations and the firm has incurred positive
transaction costs. These are the only observations susceptible of being biased (we are
confident that the other observations in our sample contain an unbiased transaction
costs measure).
The average transaction costs percentage for the group of forty-four agreements
mentioned above is 9.74%, while the average percentage for those agreements in which
the partner has not asked for renegotiations and yet positive transaction costs were in-
curred (one hundred sixty-five observations) is 7.93%. The comparison between these
two groups is the most relevant given our interest in the effect of the renegotiation
request on positive transaction costs11. The t-test on the equality of the two means,
which uses standard errors adjusted for intra-group correlations (since agreements
corresponding to the same firm may be similar), shows that the null hypothesis of
equal means cannot be be rejected for the usual confidence levels (p-value = 0.254).
10This may happen for at least two reasons. First, firms may sometimes find costless detterents
of hold-up (a well-designed contract may be such a tool, which implies higher costs when drafted,
but which may reduce costs thereafter). Second, it may be that some of the renegotiation requests
do not have an opportunistic nature, but are made in order to improve the transaction.
11We do not compare the group of forty-four agreements with the rest of the whole sample as the
latter would include many zero responses and would lead to spurious conclusions.
115
The relative low difference between the means in the two groups and the statistic the
test returns are both indicators that our transaction costs measure is not ridden by
the investigated upward bias. In fact, a higher transaction costs mean for the group of
forty-four agreements should have been expected since a request for renegotiation is
likely to lead to additional costs due to possible negotiations, unrevealed information,
and broken promises.
Another pair of questions in our survey helps us continue this investigation.
Sales and procurement managers were asked to evaluate the overall success of the
specific agreement on a scale from 0 to 10, where 0 represents a very unsatisfactory
transaction and 10 a very satisfactory one. Our conjecture is that those agreements
characterized as very successful did not probably involve opportunistic partners or, in
case they did, our respondents have found effective tools against this behavior, hence
a transfer was not needed. A dummy variable is constructed to equal one for very
successful transactions, evaluated at 9 or 10 on the satisfaction scale, zero otherwise
(58% of our two hundred nine observations representing agreements with positive
transaction costs were deemed very successful).
This newly constructed dummy variable will be used together with that indicat-
ing whether respondents? partners have asked for agreement renegotiation during its
implementation. Table 2.3 at the end of this dissertation chapter displays the trans-
action costs mean for those agreements characterized by positive transaction costs,
splitting them by the two dummy variables.
Making use of the cell labels in Table 2.3, the difference between A andB is likely
116
to represent only additional transaction costs incurred because of the renegotiation
request, while the difference between C and D is likely to include both additional
transaction costs and potential transfers. Then, the difference of differences (C ?
D) ? (A ? B) = 0.38 would provide an estimate of those transfers that may be
biasing upward our transaction costs estimate. The difference-in-difference regression
of the transaction costs variable on the two dummies and their interaction returns
a p-value = 0.905 for this estimated coefficient, making it non-significantly different
from zero. Moreover, the 0.38 value may in fact be an over-estimate of the level
of transfers. An implicit assumption made in the above computation is that the
additional transaction costs arising from the renegotiation request are the same for
all agreements (both the very successful and the rest). In fact, the transaction costs
component of the difference C ? D may be larger than that of A ? B (answering
9 or 10 on the satisfaction scale may imply that partners have found efficient ways
keep the transaction costs at a low level, but also to limit their evolution). Therefore,
apart from the potential transfer, the difference of differences (C?D)?(A?B) may
also include a transaction costs component, which would make 0.38 an over-estimate
of transfers? level.
The combined evidence above shows that respondents have not included trans-
fers in their estimates of transaction costs, hence our measure of the concept is not
biased upward at respondent level. We will now proceed and investigate a potential
downward bias which may affect our transaction costs measure.
117
2.5.1.2 A potential downward bias at respondent level
If a person answering our transaction costs question thinks he could have actu-
ally benefited from the desirable features of the imaginary partner, then the answer
provided may be below the true transaction costs value. Several arguments should
persuade the reader that this is not in fact a plausible scenario.
First of all, it is unlikely that respondents would easily admit on behaving
opportunistically if possibility would arise (we believe that such information would
be hard to extract by a single question from respondents such as ours).
Second, the survey question soliciting the transaction costs information instructs
the respondent to provide an estimate between zero and 100%, hence negative re-
sponses were ruled out. We do not believe that this constraint leads to a data censor-
ing problem (that respondents who would otherwise have provided a negative answer
have, for example, returned a zero).
Related to the one above, the third argument is that placing this restriction on
the acceptable answers, combined with the precise wording of the question, encourages
respondents to think of the potential reduction in costs, and not of the possible
increase in benefits from trading with a frank, trustworthy and fair partner. This is
because the features the imaginary partner possesses do not necessarily expose him to
respondent?s opportunistic behavior. Namely, the imaginary partner being willing to
share all relevant business information does not mean he would share all information
so as to become vulnerable. Then, the imaginary partner making it a priority to keep
all promises does not immediately imply he would accept our respondent not keeping
118
his promises. In connection to this point, it is assumed that promises in a contract
are contingent (a partner breaking his part of the deal should not expect the other
partner to keep his promises any longer). Lastly, the imaginary partner negotiating
equitably to solve any problems does not mean he would accept any terms, nor that
he can be easily deceived by its partner (represented by our respondent).
An empirical investigation is also possible. Sales and procurement managers in
our survey were asked whether their firm has demanded renegotiation of the specific
agreement ? same as before, transfers are used in case of opportunistic behavior, and
we view a request for renegotiations as a necessary condition for such behavior. In
our sample, 20.16% of interviewed firms have asked for agreement renegotiation and
the mean transaction costs as percentage of price for these agreements is 3.42, while
for the rest of the sample (79.84%) the mean is 3.50. Not surprisingly, these means
are not significantly different from one another (the probability level using standard
errors calculated under the assumption that agreements entered into by the same
firm have correlated errors is 0.903). Hence, data seem to support our belief that the
answers we record for each respondent are not biased downward relative to the true
transaction costs level.
2.5.2 Potential biases at aggregate level
The next step we take in analyzing the quality of our transaction costs measure
is to investigate whether it suffers from biases at aggregate level.
Despite the fact that, as shown above, our respondents have provided accurate
119
estimates of the incurred transaction costs, the overall picture of the phenomenon
may be distorted. This may be caused by the fact that the specific agreements we
observe have not entered our sample in a random way, hence it is possible that our
data is ridden by a selectivity bias.
For obvious practical considerations, our sample contains those agreements re-
spondents have chosen to discuss, which are not randomly drawn from the whole
population of agreements the responding firms have entered into. Rather, the survey
instruments have asked procurement and sales managers to choose one specific agree-
ment and relate to it when answering the questions in survey?s main section. The
exact instructions communicated to our respondents, displayed in Boxes 2.1 and 2.2
at the end of the chapter, suggest that the choice of specific agreement should not
depend on its success or failure. However, it is possible that our respondents had in
mind particularly better or worse agreements and, if so, even though they have each
provided accurate estimates of transaction costs, our aggregate view on the concept
may be under- or over-estimated.
After choosing the specific agreement, one of the first questions addressed to
sales and procurement managers is to compare the size of the chosen transaction
to other transactions the enterprise typically undertakes. Responses indicate that
specific agreements tend to have particular features (for instance, 40.55% are larger
than the typical ones). What matters for us the most is whether the specific agree-
ments posess particular features in what concerns transaction costs, and the way our
questionnaires are structured permits investigation of this potential problem.
120
The survey section focusing on the specific agreement is preceded by a set of
questions which extract information on all firm?s operations. Some questions in these
two sections are similar. For example, in connection to their activity and transactions
in general, the sales and procurement managers were each asked:
After finalizing contracts with your partners, what percentage of them
ask to renegotiate those contracts during the period of implementation
(regarding any issues, major or minor)? % of all contracts
and
After finalizing contracts with your partners, with what percentage of them does
your enterprise ask for contract renegotiation during the implementation period
(regarding any issues, major or minor)? % of all contracts
These two questions are similar to those presented in Subsections 2.5.1.1 and
2.5.1.2, which ask whether the firm or the partner have requested renegotiations of
the specific agreement. We will use these two pairs of questions in two ways. First,
they will help us assess whether the responding firm or its partner have asked for
agreement renegotiation during its implementation. Then, given that both pairs of
questions have been addressed to both interviewed managers (once the firm has acted
as a buyer and once as a seller), they will help us assess whether the buyer or the seller
in the agreement has asked for its renegotiation. Apart from these, three other pairs
of questions target both the specific agreement and all firm?s agreements. One such
pair asks about buyers? prepayments, one about the complaints of the responding
121
companies to their partners, and one captures information about past transactions
between the two partners.
Using similar information about the specific agreement and all firm?s agreements
will help determine whether the former have particular features in what concerns
transaction costs. This is accomplished by constructing a predicted transaction costs
mean over all firm?s agreements and comparing it with the transaction costs mean
over specific agreements. Table 2.4 shown at the end of the chapter contains the
relevant information, and an example of the analysis it contains is presented below.
Let us consider partner?s request for renegotiations. We observe that in 18.34%
of the specific agreements the partner has asked for renegotiation, while this happened
in 11.42% of all agreements the investigated companies have entered into, hence the
chosen specific agreements tend to have some different features from the rest. As
pointed above, what matters for us is whether these different features affect our
transaction costs measure. We proceed and compute several transaction costs means
(as percentage of sale price): for those specific agreements in which partners have
asked for renegotiation the mean is 4.81, for those in which they have not done
so it is 3.18, and for the whole sample of specific agreements it is 3.48. Then, we
can compute the predicted transaction costs mean for all firms? agreements. The
prediction uses the proportion of all agreements firms have entered into and in which
partners have asked for renegotiations (11.42%) and applies it to the mean transaction
costs for specific agreements with (4.81) and without (3.18) such a partner request.
Computations (0.1142?4.81+(1?0.1142)?3.18) lead to a predicted transaction costs
122
mean for all firm?s agreements of 3.37, which is very close to the mean over specific
agreements (3.48).
Based on the information presented in Table 2.4 it seems that the specific agree-
ments constituting our sample have slightly different features from the rest of firms?
agreements. In particular, except for the last two rows of Table 2.4, which have a
different meaning, percentages in columns (b) and (d) tend to indicate that the cho-
sen specific agreements are somewhat more problematic than the usual agreements
firms enter into. As a result, the mean transaction costs for the sample of specific
agreements ? column (h) ? tends to be slightly higher than that predicted for all
agreements firms have entered into ? column (i). Nonetheless, we don?t consider
these differences to be notable and conclude that, if present, the sampling bias is a
relatively small one.
Another source of distortion at aggregate level is some type of sample truncation
given that we do not observe agreements which never took place. That is, we do not
observe any characteristics, including the level of transaction costs, for those potential
exchanges for which the expected transaction costs exceed a certain threshold, and so
were deemed to be too risky to be pursued further12. Ideally, the survey should have
asked managers to choose an agreement their enterprise was or could have been part
of (but was not due to high levels of expected transaction costs). Then, the same
information on the real and potential/missed agreements should have been extracted.
12In the present context it does not matter whether the alternative to an exchange characterized
by too high transaction costs is an exchange with another partner, thus acceptance of lower profit
combined with avoidance of the high transaction costs, or integration (unification of control rights),
thus continued pursuit of the high profit, combined with reduction of transaction costs.
123
Unfortunately, the futility of such an attempt was obvious when designing the survey
instruments13 and led to our respondents being asked to pick one agreement their
company has actually been part of.
Whether this issue leads to a distortion or not depends on researcher?s goal.
If one is interested in the level of transaction costs actually incurred by Romanian
companies, the point presented above does not necessarily lead to our measure being
biased at aggregate level compared to the true value. If instead the researcher wishes
to investigate the size of transaction costs companies are exposed to and based on
which decisions are taken, then the overall picture we have on transaction costs is
actually an under-estimate of the true one (even though individual respondents have
each provided accurate estimates). What is unfortunate in this case is that we cannot
measure or counteract in any way this possible bias ? in particular, we cannot make
any claim on whether this downward bias outweighs the small upward sampling bias
presented previously.
2.6 Conclusions
The burgeoning interest in institutions inevitably leads to a need to measure
phenomena that are much more elusive than the standard financial, demographic,
and policy measures that underpin most of empirical economics. Those who have
been in the forefront of developing new measurement methods freely acknowledge the
amount of error that must be present (Kaufmann, 2002). Thus, it is not appropriate to
13As an entertaining exercise, the reader can try to rephrase the transaction costs question in
terms of an agreement which never took place.
124
demand from a new methodology that it produces data as accurate as that of standard
measures, but rather that it bears on an important as-yet-unmeasured theoretical
concept, that it is eminently practical, and that resultant data avoid the worst biases.
This dissertation chapter has presented a methodology for measuring transac-
tion costs that succeeds on all three fronts. Our data has been shown to be free
of investigated biases, both at respondent and aggregate level, hence we believe it
can safely be used in regressions along with other transaction- and firm-level data.
However, we present below two solutions for those readers who may still feel skep-
tical about the arguments presented when discussing the potential upward bias at
respondent level.
One approach could be to implement a slightly changed version of our question.
The researcher may choose to construct a question similar to those displayed in Boxes
2.3 and 2.4, but more extensively worded so that to remove altogether the second sce-
nario we have mentioned in Subsection 2.5.1.1. Alternatively, very precise additional
questions could be included, to help the researcher assess whether respondents think
of the initial stages of the agreement or the final one, and, if the latter, whether they
have included transfers in their answers.
Another approach could be to re-code the collected data. A dummy variable
equal to one when positive transaction costs are reported and zero otherwise may
be used instead of the continuous variable we record. This approach leads to a
loss of information but completely removes the possibility of an upward bias (when a
transfer to partner takes place, transaction costs are surely positive since that transfer
125
is likely to follow a break of a promise, some hidden information, and probably some
renegotiations).
To conclude, the methodology is speculative and refinements can undoubtedly
be made. But there seems to be no reason why improved versions of this methodology
cannot be used to obtain accurate measures of the transaction costs of trade that
facilitate the comparison of firm behavior across sectors, regions, and countries.
126
Box 2.1: Instructions for sales managers on how to choose the specific agreement
Please choose one specific sales agreement entered into by your enterprise with one of your customers. In choosing an
agreement, please consider the following:
? You must be thoroughly familiar with the agreement and its implementation.
? The agreement, as originally made, provided for sale in the past six months.
? Feel free to choose either a successful or an unsuccessful agreement; either one in which you and the customer
fulfilled your obligations to each other satisfactorily or one where problems in implementing the agreement were
present.
? The agreement may relate to efforts (either successful or unsuccessful) to sell a new product, or to the sale of
traditional products of your company.
Box 2.2: Instructions for procurement managers on how to choose the specific agreement
Please choose one specific agreement entered into by your enterprise for the procurement of good or service from a specific
supplier. In choosing an agreement please consider the following:
? You must be thoroughly familiar with the agreement and its implementation.
? The agreement concerns the purchase of a good or service.
? The good or service is important to your enterprise, but it is not an energy input.
? The agreement, as originally made, provided for delivery in the past six months.
? Feel free to choose either a successful or an unsuccessful agreement; either one in which you and the supplier fulfilled
your obligations to each other satisfactorily or one where problems in implementing the agreement were present.
12
7
Box 2.3: The transaction costs question addressed to the sales manager
The next question is the last question on this specific agreement, but we ask you to make a very difficult estimation. But
a very rough estimate of the information that we seek will greatly help our understanding of the market environment in
Romania. Please study the information in the question carefully, reflect on the agreement you have just described, and
then answer as best you can. We know that your answer will be a very approximate one.
Think back to the situation before your enterprise began negotiating this specific agreement. Suppose at that time you
could have identified a different customer, an imaginary one, who had the same characteristics as the one with whom
you made the agreement, except that the imaginary customer:
? would be willing to share all relevant business information with you;
? would make it a priority to keep all promises, written or oral;
? would negotiate equitably to solve any problems that arose during implementation of the agreement.
Your enterprise might have been able to offer to the imaginary customer a lower price than the actual one used, while
still profiting as much.
How large a price discount (in percentage terms) could this have been? %
In forming your estimate, please consider the following:
A 0% discount means that the customer with whom you actually traded was the equivalent of the imaginary customer.
A 100% price discount means that agreement effectively resulted in your enterprise giving away the product.
12
8
Box 2.4: The transaction costs question addressed to the procurement manager
The next question is the last question on this specific agreement, but we ask you to make a very difficult estimation. But
a very rough estimate of the information that we seek will greatly help our understanding of the market environment in
Romania. Please study the information in the question carefully, reflect on the agreement you have just described, and
then answer as best you can. We know that your answer will be a very approximate one.
Think back to the situation before your enterprise began negotiating this specific agreement. Suppose at that time you
could have identified a different supplier, an imaginary one, who had the same characteristics as the one with whom you
made the agreement, except that the imaginary supplier:
? would be willing to share all relevant business information with you;
? would make it a priority to keep all promises, written or oral;
? would negotiate equitably to solve any problems that arose during implementation of the agreement.
Your enterprise might have been able to offer to the imaginary supplier a higher price than the actual one used, while
still profiting as much.
How large a price increase (in percentage terms) could this have been? %
In forming your estimate, please consider the following:
A 0% price increase means that the supplier with whom you actually traded was the equivalent of the imaginary supplier.
A 100% price increase means that it would be worthwhile for your enterprise to pay double the price to avoid the
transacting problems presented to you by the supplier who was in the specific agreement.
12
9
Figure 2.1: The distributions of the transaction costs responses
130
Table 2.1: The size of transaction costs
Respondent Respondent Respondent
firms have acted firms have acted firms have acted
as sellers as buyers on either side
Percentage of agreements in which positive transaction 49.20 34.66 41.92
costs were reported
Mean transaction costs as percentage of sale price 3.87 3.11 3.49
(mean over all respondents)
Mean transaction costs as percentage of sale price 7.86 8.97 8.32
(mean over all respondents reporting positive transaction costs)
Mean transaction costs as a percentage of value added 12.90 10.37 11.63
(mean over all respondents)
Mean transaction costs as a percentage of value added 26.20 29.90 27.73
(mean over all respondents reporting positive transaction costs)
Number of observations in each group 250 251 501
13
1
Table 2.2: The correlates of transaction costs
Number of Mean transaction costs Probability
agreements for agreements level for
with or without with or without two-sided test
Characteristic of the agreement or firm: characteristic characteristic of difference
with without with without between means
One of the firms has more than 250 employees 156 350 2.679 3.829 0.058
One of the firms has been founded after 1989 240 264 4.072 2.955 0.071
One of the firms has some state origins 316 192 3.200 3.963 0.247
State owns shares in one of the firms 86 422 2.982 3.591 0.379
State owns the majority of shares in one of the firms 76 432 2.967 3.580 0.407
Employees own the majority of shares in one of the firms 226 282 3.768 3.262 0.420
One of the firms has recently invested in building or equipment 452 56 3.709 1.732 0.000
One of the firms has recently ceased some operations 102 406 2.683 3.691 0.054
Written contracts cover all important elements of the agreement 402 106 3.482 3.506 0.971
The seller has made a specific investment in the agreement 129 374 4.781 3.064 0.025
The buyer has made a specific investment in the agreement 59 447 3.672 3.452 0.843
The seller could not easily find another buyer in case of default 234 258 3.824 3.243 0.346
The buyer could not easily find another seller in case of default 207 289 3.578 3.500 0.901
It takes more than a quick inspection to ascertain good?s quality 256 250 4.097 2.871 0.037
Quality defects are observable only by knowledgeable third-party 341 164 3.734 2.997 0.202
Exogenous uncertainty is very important 118 388 4.655 3.152 0.040
The agreement is part of a longer-term one (multiple exchanges) 358 149 3.369 3.792 0.521
Delivery of the good takes place at least four times a month 96 243 2.176 4.050 0.002
The seller has asked for agreement renegotiation 104 403 3.575 3.450 0.841
The buyer has asked for agreement renegotiation 91 411 4.653 3.231 0.129
The agreement has been renegotiated 111 395 4.299 3.259 0.189
The agreement is the first one between the two partners 150 358 3.907 3.309 0.371
The two partners are located in the same region 204 302 3.375 3.587 0.724
Probability levels use standard errors calculated under the assumption that observations on the same firm have correlated errors.
13
2
Table 2.3: Transaction costs information based on renegotiation requests and success
of agreements
Has the partner requested
renegotiations of the agreement?
Yes No
Was the transaction Yes 8.75 (20) A 7.38 (101) B
a very successful one? No 10.56 (24) C 8.81 (64) D
Each cell of the table contains the transaction costs mean over the agreements satisfying
the two criteria, the frequency of responses in each category (in parentheses), and a label.
133
Table 2.4: Investigation of biases at aggregate level
Percentage Percentage Transaction costs as a percentage of price
of specific of all firms? Mean over Predicted
agreements agreements specific Mean over mean over
with or with or agreements specific all firms?
without without with or agreements agreements
Characteristic of agreement: characteristic characteristic without using (d),
characteristic (e), (f),
with without with without with without and (g)
(a) (b) (c) (d) (e) (f) (g) (h) (i)
Partner has asked for agreement renegotiation 18.34 81.66 11.42 88.58 4.81 3.18 3.48 3.37
Firm has asked for agreement renegotiation 20.16 79.84 9.76 90.24 3.42 3.50 3.48 3.49
Buyer has asked for agreement renegotiation 17.98 82.02 10.02 89.98 4.65 3.23 3.48 3.37
Seller has asked for agreement renegotiation 20.51 79.49 11.16 88.84 3.58 3.45 3.48 3.46
Firm has complained to the partner 31.10 68.90 13.18 86.82 3.92 3.29 3.49 3.37
Buyer has made a partial prepayment 38.78 61.22 20.75 79.25 4.48 2.87 3.49 3.20
Partners have traded for at least two years 66.80 33.20 66.41 33.59 3.41 3.65 3.49 3.49
The values on column (h) vary slightly by row due to varying samples caused by missing observations on characteristic in column (a).
13
4
Chapter 3
An Analysis of the Factors Determining the Genesis and Size of
Transaction Costs
Abstract
This chapter of the dissertation investigates the determinants of transaction costs.
To accomplish this task, various factors proposed by the transaction cost economics
theory are gathered. This first step sheds light on the scarcity of existing empirical
work using transaction costs information at agreement level. A survey of Romanian
companies is employed. It contains the question which was proposed in the second
chapter, and which collects the transaction costs information expressed as percentage
of sale price. The data thus gathered is limited, in the sense that more than half
of agreements are characterized by zero transaction costs, while the positive answers
provided reach as high as sixty percent of the sale price. Since ordinary least squares
on the whole sample is not advisable, we first employ the model commonly used for
this type of data ? Tobit. The sample selection model is found to be more appropriate.
However, combined evidence indicate that sample selection bias would not affect the
estimates obtained by distinct regressions: an ordinary least squares regression using
the positive transaction costs responses, and a probit regression using a transaction
costs dummy variable. These distinct regressions provide the main results of this
135
chapter. They indicate that the existing theory is successful at predicting both the
existence and the size of transaction costs, with an emphasis on the former. Addi-
tional results include a discussion of the two-sided nature of the relationship-specific
investment decision, and of the potential endogeneity of several factors.
3.1 Introduction
A new way to measure transaction costs incurred during basic exchange agree-
ments has been presented in the second chapter of this dissertation. In short, in
connection with a particular agreement, respondents in a firm study were asked to
estimate how much it would have been worth to reverse history and deal with a non-
opportunistic partner (instead of trading with the actual partner in the agreement).
The answers to the proposed question, representing in particular those transactional
costs stemming from haggling, bargaining and mis-alignment, were expressed as per-
centage of sale price, and range from zero to sixty percent.
The second chapter of this dissertation includes a discussion on the size of
transaction costs faced by Romanian companies in dealing with their customers and
suppliers. Data show that positive transaction costs are incurred in 41.92% of agree-
ments, and that there seems to be some difference between buyers and sellers in
an agreement (sellers reported positive transaction costs in more agreements). The
average transaction costs, as percentage of the sale price, is 3.49 for the whole sample.
The goal of the present chapter is to use the same data and assess the impact of
various factors on the size and the probability of existence of transaction costs. Since
136
details on the survey have been provided in the previous two chapters, we only mention
here that in the middle of 2001, two hundred and fifty-four Romanian companies
were surveyed in an attempt to understand the role of formal legal system in firms?
activity. In each company four decision makers were interviewed: the company?s
general manager, the head of the legal department, and the managers of the sales
and procurement departments. The survey instruments addressed to the last two
respondents are similar, the only difference residing in questions? point of view; while
the sales manager was asked about company?s activity as a seller, the procurement
manager was inquired about company?s behavior as a buyer. Each of these two
survey instruments contains a large set of questions soliciting information on a specific
agreement the company has been involved into (once as a seller, and once as a buyer).
To assure informativeness, the question soliciting the transaction costs information is
the last one in the specific agreement section.
The transaction costs information we collect provides the dependent variables
of our regressions, and we use respondents? answers in three ways ? all recorded re-
sponses, the information coded as a dummy variable (as suggested in the second chap-
ter of this dissertation), and, in some regressions, only the positive values recorded.
The organization of this dissertation chapter is as follows. The next two sec-
tions discuss those factors thought to influence transaction costs, and also provide
details on the construction of our dependent variables. Interestingly, the transaction
cost economics (TCE) literature hints indirectly on the factors affecting transaction
costs, by mainly discussing the determinants of vertical integration (VI). However, in
137
his introduction to Case Studies in Contracting and Organization, Scott E. Masten
provides a clear and concise review of the TCE literature, and presents a set of factors
along with predictions on their impact on transaction costs ? these factors are asset
specificity, market uncertainty, frequency of exchange, ease of measurement, reputa-
tion, transaction complexity, and attitude toward risk. Apart from these, the impact
on transaction costs of other factors is investigated, namely whether both partners
are Romanian, the age of the firms, and their organization of legal matters.
Section 3.4 provides the econometric details of our analysis and reports the re-
sults. Given the limited nature of the transaction costs information we record, we
first employ the Tobit model. However, this is shown to impose restrictions which are
too restrictive in our case, and the sample selection model is found to be more ap-
propriate. Combined results indicate that sample selection bias would not affect our
estimates if the sample selection model would be replaced by two distinct regressions.
Hence, we proceed and estimate a probit regression using as dependent variable a
transaction costs dummy variable, and an ordinary least squares (OLS) regression
having as dependent variable the positive answers our survey question records. These
distinct regressions lead to the main results of this chapter. The most interesting
finding is that TCE theory seems to be very successful in predicting the existence
of transaction costs, and only moderately successful in explaining transaction costs?
size when such costs are incurred. The econometric section closes with some addi-
tional results, which take into account asset specificity?s two-sided nature and the
endogeneity of some of our variables. Section 3.5 concludes the chapter.
138
3.2 Determinants of transaction costs proposed by theory
The first step in a research such as ours is to gather the hypotheses theory
proposes (which are then tested by running the appropriate regressions). Interestingly
enough, this first step proved more challenging than one would expect, confirming
second chapter?s comments on the scarcity of empirical work on transaction costs at
agreement level.
Rather than directly providing the factors influencing transaction costs (and the
direction of the influence), most of the literature on transaction costs hints indirectly
on which those factors are. The central idea of TCE is that firms adopt different gover-
nance forms for their agreements in order to minimize transaction costs. Williamson
(1991b, p. 79) states that ?[t]he main hypothesis out of which transaction costs
economics works is this: align transactions, which differ in their attributes, with gov-
ernance structures, which differ in their costs and competencies, in a discriminating
(mainly, transaction costs economizing) way.?
The most often discussed governance structure is VI, when trading partners act
under unified ownership and control ? according to Shelanski and Klein (1995), the
make-or-buy decision has been the paradigm problem of TCE, with much of the ear-
liest empirical work addressing this topic. VI is seen as an alternative to contracting
or spot market transactions when hazards associated with exchanges between inde-
pendent partners are particularly severe. A significant number of papers propose and
test factors assumed to impact on the decision to integrate1. These factors are asset
1For examples of such papers the reader is guided toward surveys such as Joskow (1988), Shelanski
139
specificity, market or transaction uncertainty, transaction complexity and frequency.
Given that VI is considered to attenuate transaction costs associated with mar-
kets exchanges, implicitly, those factors increasing the chances of integration are those
increasing the size of transaction costs in agreements between independent partners.
To avoid injustice, we have to mention two authors who discuss the determi-
nants of transaction costs directly. The first is Joskow (1985, p. 36), who lists some
characteristics of transactions that are considered to affect ?the nature and magni-
tude of [...] transaction costs in important ways?. The factors are those presented
above. The other author, to the work of which we extensively relate below, is Masten
(1996). In his introduction to Case Studies in Contracting and Organization, Scott
E. Masten makes a clear and concise review of the TCE literature, and presents a
set of factors along with predictions on their impact on transactions costs. Masten?s
factors include those previously presented, to which several others are added.
As mentioned before, transaction costs are considered to arise from two features
of the human nature: bounded rationality and opportunism. Masten (1996, p. 6-7)
provides details: ?Opportunistic behavior can be divided into two principal types:
(1) deviations from joint-surplus maximizing behavior that produces a de facto re-
distribution of gains from trade within the terms of an existing agreement, and (2)
efforts designed to extract more favorable terms at the outset or to force a renegoti-
ation and thus a de jure modification of terms previously agreed to. The first type
of opportunism, which is a response to price signals contained in a contract, is what
and Klein (1995), Masten (1996), Sykuta (2005), and Macher and Richman (2006).
140
economists commonly refer to as moral hazard. [...] Ideally, the transgressor would
like his behavior to go undetected by both the courts and his trading partner. The
second form of opportunism consists of efforts to ?hold up? a trading partner and
involves actions taken to set new terms rather than in reaction to existing ones.?
The author then discusses the factors which influence the relative efficiency of various
organizational forms, hence the size of transaction costs, and the predictions asso-
ciated with each of them. We follow the same order of exposition, and also briefly
present the variables we will use in our regressions (details on variables? construction
are available in Appendix E at the end of the dissertation).
Asset specificity
Following Klein et al. (1978), investments in relationship-specific assets generate
appropriable quasi-rents, the distribution of which may create tensions between trad-
ing partners. Hold-up, the second form of opportunistic behavior presented above, is
more likely to occur under such conditions, therefore transaction costs are expected
to be higher in market exchanges where specific investments have been made.
One thing to note is that, in general, TCE does not make a distinction between
relationship-specific investments made by one or the other party in a transaction (as-
set specificity is viewed as a singular concept for the transaction as a whole). However,
the TCE analysis provided in the first chapter of this dissertation has shown that buy-
ers? and sellers? relationship-specific investments have significantly different impacts
on contract complexity. We will follow the same approach here: after discussing the
impact on transaction costs of variable Specific investment, which represents invest-
141
ments made by either party in transaction, we also analyze the distinct implications of
buyers? and sellers? relationship-specific investments (variables labeled Buyer specific
investment and Seller specific investment).
Market uncertainty
Scott E. Masten considers that greater uncertainty makes it harder to design
complete agreements, hence parties will either fail to adapt to changing conditions
or will find it profitable to engage in costly efforts to evade performance. Uncertain
environments are therefore likely to encourage the first type of opportunistic behavior,
ultimately leading to higher transaction costs2.
General managers of the interviewed firms were asked about the importance of
certain events in their area of business. We focus on unpredictable changes in weather
conditions and transportation links in order to capture truly exogenous events, and
construct the Exogenous market uncertainty dummy variable to equal one if such
sources of unpredictability were considered to be very important by general managers,
zero otherwise.
Frequency of exchange
There seems to be a general consensus that transactions involving more frequent
exchanges are likely to be characterized by lower transaction costs. One reason is that
opportunistic behavior of the first type ? deviations from joint-surplus maximizing
behavior ? is probably easier to detect and counteract in such agreements. In addi-
2This view is different from that expressed by Shelanski and Klein (1995, p. 339), to which we do
not concur: ?Empirical studies sometimes treat [uncertainty about future events] as an independent
variable, regressing the choice of organizational form on the variance of sales or another variable, but
without including any measure of asset specificity in the model. Absent fixed investments, however,
TCE does not predict that uncertainty would itself lead to hierarchical governance.?
142
tion, sellers in exchanges involving multiple deliveries may reduce their exposure to
trading partners? opportunism by threatening or actually imposing delays (or even
cancellations) of subsequent deliveries for defaulting buyers.
In our regressions we employ a dummy variable which equals one if the specific
agreement is part of a long-term deal involving an average of four or more sepa-
rate deliveries per month, and equals zero if the transaction covered by the specific
agreement is a one-time exchange or entails less than four deliveries per month.
Ease of measurement
Masten (1996) claims that problems of wasteful sorting and searching arise
when it is difficult for a party to obtain information about the characteristics of a
good. Ease of measurement is therefore considered to affect chances of opportunistic
behavior of first type: market exchanges featuring goods whose true characteristics
are observable are expected to be associated with lower transaction costs.
The interviewed procurement and sales managers were asked how easily can the
buyer in the specific agreement ascertain the quality of the traded good. The dummy
variable we construct, labeled Good easy to measure, equals one if the good?s true
characteristics are observable after only a quick inspection, and zero if they are hard
to observe (it would take an intensive inspection or extended use for the buyer to
assess the quality of the transacted good).
Reputation
Partners in a transaction are likely to take into consideration the impact of their
current actions on future transactions, therefore, should information about current
143
events disseminate rapidly among firms, others may be reluctant to trade with those
who have defaulted on their obligations or have otherwise behaved in a opportunistic
fashion. In the words of Masten (1996, p. 15), ?In such settings, the long-run loss
of confidence of one?s current or potential future trading partners tends to outweigh
the short-run gain from cheating a trading partner.? Thus, the testable hypothesis
is that lower transaction costs should be observed for agreements between partners
who act on markets characterized by rapid dissemination of relevant information.
Given the two-sided character of this market feature, we construct two distinct
dummy variables: Information dissemination among buyers and Information dissem-
ination among sellers.
Transaction complexity
The degree of complexity associated with the transaction is also assumed to
influence the design of contractual agreements through its impact on transaction
costs. Masten (1996, p. 14) elaborates on the role of transaction complexity: ?The
more complex the transaction, the harder it becomes to describe fully and accurately
the responsibilities of each party in a contract, and the more difficult it will be for
courts to assess whether those obligations have been fulfilled.? The first type of
opportunism (moral hazard or deviations from joint-surplus maximizing behavior) is
more likely to take place for more complex agreements, hence transaction costs are
probably higher for these exchanges.
As transactions may be complex in many dimensions and for many reasons,
our survey instruments do not include a unique question capturing information on
144
transaction complexity, thus in testing the above hypothesis we rely on a proxy.
Dummy variable Customized contracts used frequently is employed, and we use this
piece of information under the assumption that more complex transactions usually
require more attention, therefore custom contracts.
Attitude toward risk
Masten (1996) briefly discusses a factor which is generally absent from the
transaction costs literature: risk aversion, and uses the term in in the textbook sense.
However, in our framework a better name may be attitude toward transactional risk
or even acceptance of the risk of betrayal in an exchange.
Since we do not measure directly respondents? attitude toward risk, we have
to rely on a proxy in our regressions. We relate respondents? attitude toward risk
to their reported attitude toward trusting people. We are not alone in making this
inference. Bohnet and Zeckhauser (2004, p. 468) also mention that ?a large body
of work on trust, crossing many disciplines, assumes that the willingness to trust is
closely associated with the willingness to take risk?, and provide references to papers
in economics, philosophy, and sociology. A similar view is expressed by Eckel and
Wilson (2004, p. 447) and Karlan (2005, p. 1698).
However, some cautionary notes are needed. In a discussion of a survey question
measuring trust, which is similar to the one we use, Glaeser et al. (2000, p. 812) men-
tion that ?while these survey questions are interesting, they are also vague, abstract,
and hard to interpret.? Based on experiments with monetary rewards the authors
claim that attitudinal trust surveys at best weakly predict any individual?s level of
145
trust, and instead they may be better at predicting the overall level of trustworthi-
ness in society. Also, despite their conceptual similarities, some empirical results have
pointed out the differences between trust and risk aversion ? see Eckel and Wilson
(2004, p. 463-464).
3.3 Other variables we construct
After discussing the factors proposed by theory and the corresponding variables
we construct, we now proceed to briefly describe the other variables we employ in this
dissertation chapter. We begin with those that will the used as dependent variables
in our regressions, and continue with the additional variables we use in the right
hand side of these regressions. We hold back from the presentation of some other
variables, employed when the potential endogenous nature of the decision to invest in
relationship-specific assets is discussed. These variables have been presented in the
first chapter of this dissertation.
Dependent transaction costs variables
The second chapter of this dissertation has presented a new way tomeasure (part
of) the transaction costs firms incur in their sales agreements, by means of a survey
question. Sales and procurement managers in a sample of Romanian companies were
presented the proposed question. Figure 3.1 contains the distribution of answers we
collected (for presentation purposes, only the positive responses are displayed). The
figure shows that many respondents have returned 5% and 10% as their answers, but
also that a full range of other intermediate values have been chosen.
146
The variable Transaction costs: All responses contains the information gathered
by the proposed survey question (both the positive values displayed in Figure 3.1 at
the end of this chapter and the two hundred ninety-one zero answers recorded). Apart
from using the information exactly as collected by our question, we also proceed to
some data transformations. In some regressions we need to use only the positive
values recorded ? this variable is called Transaction costs: Positive responses, and
is the one Figure 3.1 actually displays. The econometric procedures we will employ
also require us to construct a dummy variable equal to one if a positive answer was
provided by our respondent, and zero if not transaction costs were reported ? we call
this variable Transaction costs: Dummy variable3.
Apart from those theory proposes, our regressions will include additional factors
assumed to have an impact on transaction costs Romanian companies face.
Both partners are Romanian
The particular features of the Romanian economic environment and legal system
may influence the level of transaction costs. To investigate this impact, our regressions
will contain a dummy variable called Both partners are Romanian, whose construction
is evident from the name.
The testable prediction is that agreements of Romanian companies with foreign
partners are characterized by higher transaction costs that those involving only Ro-
manian partners. There are at least two reasons for this. First, exchanges between
3The second chapter of this dissertation has argued that our data is not affected by investigated
biases. It has also shown that employing Transaction costs: Dummy variable removes any residual
doubts about an upward bias at respondent level.
147
Romanian and foreign companies are likely to require the use of translators and of
specialized legal assistance (since they are probably governed by a different body
of law). Second, such agreements are more exposed to opportunistic behavior since
reputation safeguards are probably less efficient.
Age of the firms
The age of the firms may also have an effect on transaction costs: it is plausible
that over time firms develop better trading practices, which lead to smoother trans-
actions, thus the prediction is that the age of the companies is inversely related to
the level of transaction costs they face. To investigate this, our regressions will incor-
porate a dummy variable called One of the firms founded before 1990. The particular
year used to differentiate ?new? from ?old? companies represents the beginning of
transition after the fall of the centralized system in Romania.
Organization of the legal matters
A significant number of papers in the comparative institutional literature con-
sider the formal legal system to play a major role in the way businesses organize their
activity. A somewhat different view is that, given the poor standards of the legal
system in transition economies, firms in such environments rely more on informal
measures when conducting and enforcing their agreements.
The first chapter of this dissertation has shown that Romanian companies seem
to use the formal legal system, by designing more complex contracts as theory would
predict, and when doing so, by taking into account the quality of legal services pro-
vided by courts. In this chapter we try to investigate whether firms? organization
148
of legal matters impacts on the level of transaction costs. We rely on two dummy
variables in this endeavor, which are presented below.
Each of our survey instruments contains a set of true/false or multiple choice
legal questions (details on the survey section incorporating these questions are pre-
sented in Appendix E). Given that most of the information we rely on in this chapter
comes from the sales and procurement managers, our focus is on the answers these
respondents have provided to the legal questions. Dummy variable Respondent has
some legal knowledge is constructed to equal one if the respondent has correctly an-
swered at least two of the five legal questions, zero otherwise.
As previously mentioned, one of the surveys was addressed to the head of the
legal department ? where the company did not have a legal department established,
the legal consultant most frequently interacting with the company was interviewed
instead. One piece of information this questionnaire gathers is the location of re-
spondent?s office relative to the general manager?s office (when the respondent was a
legal consultant from outside the enterprise, the interviewer was instructed to refer
to respondent?s office when working on site for the company).
Dummy variable Legal adviser?s office close to manager?s is constructed to equal
one if the two offices are in the same building and within twenty meters from each
other, zero otherwise. We believe this to be a reasonable proxy for the importance
the company places on the formal legal system, one of whose goals is to provide a
smooth trading environment.
149
3.4 Econometric details and results
This section goes over the econometric details of our analysis of transaction
costs, and our goal is to assess the impact on such costs of the various factors presented
above. We will first show that OLS estimation is not the best approach in our case.
Then, we will discuss and apply the estimating procedures that are more suitable for
the data at hand.
At least at the beginning of this section, we use Transaction costs: All responses
as the dependent variable in our regressions. Given the limited nature of this variable
(it is continuous over the set of positive values but has positive probability mass at
one point ? zero), OLS estimation using all observations is not advisable. Wooldridge
(2002, p. 518) shows some weaknesses of this approach4. Employing OLS for only
the positive transaction costs values is also in general not the right way to proceed.
Apart from the fact that this would mean throwing away observations, OLS estimates
of the coefficients of interest are likely to be inconsistent due to an omitted variable.
Achen (1986, p. 76-79) presents details on this point, to which we will also return
below. Therefore, the common approach for data such as ours is to employ censored
regression models5. We follow this trend and first use the Tobit model.
4Namely, there are three significant shortcomings: the fact that the dependent variable cannot be
linear in the independent variables (unless the range of the latter is fairly limited), that the model
implies constant partial effects, and that predicted values for the dependent variable can turn out
to be negative in many instances.
5We use the term censored regression model only because it seems to be deeply rooted in the
existing literature. As Wooldridge (2002) mentions, this is not quite appropriate for cases such as
ours since the information we record is not in fact censored (for example, through top coding). He
proposes the terms corner solution outcome and corner solution model, but these labels are not
the most suitable in our case either ? our framework does not assume economic agents solving an
optimization problem, so that for some agents the optimal choice would be the corner solution zero.
The generic term limited dependent variable is sometimes used in this dissertation chapter, with the
150
3.4.1 Tobit results
The first model we construct uses as independent variables only those proposed
by existing theory and as dependent variable the one containing all answers recorded
by the transaction costs question. The form of the model, known as ?the standard
censored Tobit model? or ?type I Tobit model? is:
Y ?1 = ?1 + ?1 ?X1 + e1,
Y1 = max(0,Y ?1 ),
where Y1 = Transaction costs: All responses, X1 = (Specific investment, Exoge-
nous market uncertainty, Four or more exchanges per month, Good easy to measure,
Information dissemination among buyers, Information dissemination among sellers,
Customized contracts used frequently, Distrustful sales/procurement manager), and
e1|X1 ? N(0,?21). One thing to note is that in our setup the latent variable Y ?1 does
not have a clear meaning and that emphasis is placed on Y1.
The Tobit results for this first model are reported in the first column of Table
3.2. Since they are preliminary results, we refrain from commenting them in detail,
and only state that existing theory seems to be strongly supported by data.
Apart from those variables theory proposes, others have been included in a
similar model to the one above. X2 is the new vector of explanatory variables, where
X2 = (X1, Both partners are Romanian, Respondent has some legal knowledge, One
cautionary note that we do not face data truncation.
151
of the firms founded before 1990, Legal adviser?s office close to manager?s).
The Tobit estimated coefficients for this model are presented in the second
column of Table 3.2. These results are very similar to the first ones in the sense
that theory?s predictions are once again strongly supported. Some of the variables
we propose also seem to have an impact on transaction costs firm incur in their
exchanges. Given the similarities between the two sets of results, and the significance
of the newly added variables, in what follows we will continue to use X2 as the set of
independent factors.
3.4.2 Sample selection results
As it will become apparent below, the Tobit model is not be the best way to
proceed, and the more appropriate model is the sample selection one ? the literature
on sample selection is vast; two survey articles on this topic are Vella (1998) and
Puhani (2000). In our terms, the usual form of the sample selection model is:
Y ?2 = ?2 + ?2 ?X2 + e2
Y ?3 = ?3 + ?3 ?X2 + e3
Y3 = 1 if Y ?3 > 0; Y3 = 0 otherwise
Y2 = Y ?2 ?Y3,
where Y2 = Transaction costs: Positive responses, Y3 = Transaction costs: Dummy
variable, Y ?2 is a latent endogenous variable with observed counterpart Y2, Y ?3 is a
152
latent variable with associated indicator function Y3 reflecting whether transaction
costs are observed. Lastly, e2 and e3 are zero mean error terms with E[e2|e3] negationslash= 06.
There are several things to note about the above model. First, we fit the
exact sample selection framework by construction, by employing the transaction costs
dummy variable. Second (and related to the first point), the latent variables do not
have a clear meaning in our model and instead, as before, the emphasis is placed on
the transaction costs information we observe. Lastly, the above model incorporates an
assumption which is not always made in sample selection models. Namely, we assume
that those factors influencing the occurrence of transaction costs are the same as those
influencing the size of transaction costs when they are incurred. For this reason, the
extended vector of independent variables, X2, was used in the first two equations of
the model. As we will see, this assumption leads to some estimation difficulties, and
will be relaxed later in the chapter.
An important feature of the sample selection model is that it includes two
distinct equations, one determining the existence of transaction costs and one deter-
mining their size when such costs are incurred. The former is usually called the main
or the substantial equation (since researchers are generally primarily interested in it),
while the latter is usually called the selection equation (since it determines whether
an observation is included in the sample). We will continue to use this common
terminology despite the fact that we are equally interested in both of these equations.
6Regarding the use of OLS for the subsample characterized by positive transaction costs re-
sponses, Vella (1998, p. 130) points that this generally leads to inconsistent estimates of ?2 due to
the correlation between X2 and e2 operating through the relationship between e2 and e3.
153
A significant characteristic of the Tobit methodology is that it combines the two
equations into one (for details see chapters 16 in both Kennedy, 2003, and Wooldridge,
2002). In our terms, this implies that the equation determining existence of transac-
tion costs is the same as the equation determining their size, assumption which we
believe to be unrealistic. Even though, as mentioned above, we find no reason why
the factors in the two equations should not be the same, we also find no reason why
the effects of those factors on the probability of existence and size of transaction costs
should be the same.
Our framework is similar to that of Melenberg and Van Soest (1996), who discuss
family expenditures on vacations. While similar factors are expected to exert influence
on the decision to go on vacation and on the amount spent on taken vacations, the
same factors? influence on the two decisions may be different. For example, while
the number of small children may reduce the probability of taking a vacation, it may
increase vacation expenditures when a vacation is taken. The authors? approach is to
consider two-equation models, one for the decision to take a vacation and one for the
decision on the amount to spend given that a vacation is taken.
3.4.2.1 Maximum likelihood estimation
Heckman (1974) has proposed a maximum likelihood (ML) estimator for the
sample selection model, closely related to the Tobit one, but which removes the strict
assumption mentioned earlier. The first and second columns of Table 3.3 shown at
the end of this dissertation chapter report the estimated coefficients obtained by ML.
154
The difference in size and statistical significance between the two sets of coef-
ficients shows that the Tobit model was not the most appropriate (the restriction it
imposed was too strong). The results indicate that theory?s predictions are strongly
confirmed for the sample selection equation ? see the second column, but not equally
strong for the main equation ? even though the signs of the coefficients are gener-
ally as expected, the statistical significance is weak, see the first column. Therefore,
the theory seems to be very successful in predicting transaction costs? existence but
only moderately successful in predicting transaction costs? size when such costs are
incurred. In what concerns the factors we have added, the results are mixed, and we
will return to them later.
Another significant piece of information is reported at the bottom of Table 3.3.
? is the correlation between the error terms of the two equations and, because this
parameter captures the dependence between e2 and e3, a test of whether ? is equal to
zero is in fact a test of sample selection bias. This statistic is of particular importance
in our case since it has significant implications on the estimating strategy to follow.
The estimate of ? is low, ?0.075. However, its standard error is needed in testing
the difference from zero, and the computation of this depends on particularities of
the estimation method. When standard errors are adjusted for clustering, Stata
returns a value of 0.036, while when clustering is not accounted for, a value of 0.311
is returned. More importantly in our case is that when clustering is used, the Wald
test is employed, which returns p-value = 0.038, thus indicating that sample selection
is present. However, when clustering is not used, the likelihood ratio (LR) test is
155
employed, and it returns p-value = 0.814, thus indicating that sample selection is
not actually present. The fact that these two tests return contradicting results has
serious consequences on the estimation approach to follow.
Thankfully, Nawata and McAleer (2001) shed light precisely on this issue: their
Monte Carlo experiments of sample selection models estimated by ML show that the
t-test (to which the Wald test is asymptotically equivalent) performs poorly under
conditions very similar to ours. Precisely, for samples of four hundred observations
and using the same regressors in the selection and main equations, the t-test rejects
the correct null hypothesis (H0: ? = 0) in 40.7% of cases. The authors suggest using
the LR test instead, which fails in only 5.2% of cases. Nawata and McAleer?s findings
are similar to those of Ekl?of and Karlsson (1999) and Sartori (2003). They show that
in a similar framework ? binary-outcome selection models, also called probit models
with selection ? the ML estimation leads to estimates of ? which may be misleading
under similar conditions on sample size and choice of regressors.
The above paragraph indicates that we should place less confidence on the Wald
test and more on the LR test. Since the latter implies that sample selection bias is not
present, we could take another estimation approach, and use two distinct regressions.
Namely, we can estimate the main equation by OLS (using the positive transaction
costs responses) and the selection equation by probit (using the transaction costs
dummy variable). The additional evidence we provide below also indicates that this
approach is safe to take.
156
3.4.2.2 Two-step estimation
According to Vella (1998, p. 133) the ML procedures are relatively uncommon
in empirical work and the two-step estimators are frequently employed for sample
selection models. The approach proposed by Heckman (1976, 1979) is to fit the selec-
tion equation by ML probit on all observations, construct an estimate of a particular
term ? called the inverse Mills ratio7, use it as an additional regressor in the main
equation, and then estimate the coefficients of this new equation by OLS over the
selected observations. In our terms, this new equation is:
Y2 = ?4 + ?4 ?X2 + ?? ?? + e4,
where ?? = ?(?c)/?(?c), with ?c = ??3 + ??3 ?X2.
Running OLS of this equation over the selected subsample returns unbiased
estimates of the coefficients of interest because the inclusion of ? leads to e4 being a
zero mean error term uncorrelated with the regressors. It has been shown that the
t-test on the null hypothesis ? = 0 is in fact a test of sample selectivity bias ? for
details see Vella (1998) and Wooldridge (2002).
The results of this two-step estimation are reported in the second part of Table
3.3. While the fourth column looks very much like the second one, the estimated
coefficients and standard errors of the second-stage regression (see the third column)
are surprisingly large. Vella (1998, p. 135) relates this to identification: ?The un-
7For any c, ?(c) = ?(c)/?(c), where ?(?) and ?(?) are the probability density and cumulative
distribution functions of the standard normal distribution.
157
derlying economic model often imposes the same variables to appear in both steps of
estimation. Thus many applications [...] identify ? through the nonlinearity if the
inverse Mills ratio. As the inverse Mills ratio is often linear, however, the degree of
identification is often ?weak? and this results in inflated second step standard errors
and unreliable estimates of ?.? Little and Rubin (1987) provide a similar argument.
The solution therefore rests in having additional variables included in the selection
equation relative to the main one.
3.4.2.3 A new specification
Based on the statistical significance of the ML estimators, we choose to remove
One of the firms founded before 1990 and Legal adviser?s office close to manager?s
from the main equation. Also, given their minimal impact on the probability of
observing transaction costs, we remove Both partners are Romanian and Respondent
has some legal knowledge from the selection equation.
For this new specification of our model, we employ the ML and two-step pro-
cedures again, and report these results in Table 3.4 at the end of the chapter. The
estimated coefficients and standard errors thus obtained are very similar to each other
and also very much like the ML results in Table 3.3, which validates our exclusion
decisions.
The results now indicate that sample selection should not be an issue. The ML
estimation returns ? = 0.023, which is lower in absolute value than it was before, and
158
is not different from zero8. Following the two-step estimation, ? (the coefficient of the
inverse Mills ratio) equals 0.124, with a standard error of 3.092 and p-value = 0.969.
To conclude on the sample selection matter, its avoidance is not a goal in itself
since Heckman?s estimation procedures properly correct for it. However, since the new
specification shows no evidence of this bias, we choose to estimate the two equations
separately, by OLS and probit. These are the main results of this dissertation chapter,
and are reported below.
3.4.3 Distinct ordinarily least squares and probit results
Table 3.5 at the end of the chapter displays the estimates obtained by OLS on
the positive transaction costs values, and by probit on the transaction costs dummy
variable. Apart from the estimated coefficients obtained by probit, the results also
include the mean of marginal effects on the probability of observing transaction costs9.
Unsurprisingly, the results of the probit regression are very much alike those
presented in the second and fourth columns of Table 3.4. In what concerns the OLS
results, the estimates in the first column confirm the fact that sample selection bias
is not in fact present since they are very similar to the sample selection ones obtained
by ML and two-step procedure.
8Table 3.4 displays the standard error adjusted for clustering, which equals 0.860. The Wald
test of independence of the two equations yields p-value = 0.978. When the clustering option is not
used, the standard error is 0.655, and the LR test returns p-value = 0.965.
9One thing to note is that all our right hand side variables are binary. Hence, computing the
marginal effect of an independent variable by evaluating the effect at the mean values of the other
variables may not be the most appropriate approach. This is because the means of the binary
variables can hardly be seen as a meaningful ?observation?. For each independent variable, an
alternative approach is to evaluate the marginal effect using each observation for the rest of variables,
and then take the mean over these marginal effects. The Stata code for these computations has been
written by Jonah B. Gelbach, and is available at http://glue.umd.edu/?gelbach/ado/.
159
The overall image we are offered by these results is once again that theory is suc-
cessful in predicting the existence and size of transaction costs. The first two columns
both indicate that six of the eight variables theory proposes have the predicted effect
on the transaction costs variables. However, same as before, theory seems to be more
successful in what concerns the existence of transaction costs. The second column
shows that five of the six coefficients having the expected sign are also significantly
different from zero at the usual levels, while this is the case for only two of the six
coefficients in the first column.
The OLS and probit results also indicate which particular factors have the
expected impact on the existence and size of transaction costs. Table 3.5 shows that
when specific investment is undertaken (by either party in the transaction), when
exogenous market uncertainty is very significant, when exchanges are less frequent,
and when good?s true characteristics are hard to observe, not only the probability of
incurring transaction costs is higher, but also the size of these costs increases.
The results in Table 3.5 also show that of the four coefficients representing the
information dissemination two have the predicted impact while two have an effect
opposite to what was expected. Also, of the two estimates which confirm our ex-
pectations only one is significantly different from zero. Hence, the aggregate impact
of these factors can be characterized as mixed. The same can be concluded for the
proxies we proposed for transaction complexity and risk aversion.
Regarding the factors we have added, the OLS results show that agreements
between Romanian and foreign companies are characterized by higher levels of trans-
160
action costs. One thing to note is that both Table 3.3 and 3.5 indicate that partners?
location influences the size of transaction costs but not their probability of appear-
ance, thus corroborating with the earlier findings which differentiate the two concepts
and lead to their separate analysis. The probit results indicate that newer firms are
more likely to face transaction costs ? the estimated coefficients of One of the firms
founded before 1990 in Tables 3.3 and 3.5 lead to a similar comment to the one we
made above.
Last but not least, an interesting finding is the one related to companies? orga-
nization of legal matters, proxied in our analysis by the location of the two offices and
the legal knowledge of two key employees. Both OLS and probit regressions indicate
a significant relationship between transaction costs and companies? attitude toward
the formal legal system: companies placing more weight on the law seem to be more
likely to face transaction costs, and also report higher levels of such costs. Since one
of legal system?s main goals is fostering a smooth business environment, these results
could be characterized as surprising. We will try to demonstrate below they are not.
We believe that firms placing significant weight on legal matters are likely to
rely on the use of the formal legal system both when trying to defend against part-
ners? opportunism and when trying to benefit from partners? exposure. But, on the
one hand, the mere use of the system is costly ? in particular it increases transaction
costs. This view is similar to that of Wallis and North (1986) who consider salaries
of lawyers and judges to be 100% transactional in nature. On the other hand, users
161
of the legal system benefit through the impact on transfers10. As the second chapter
of this dissertation has shown, our survey question accurately collects information
on transaction costs, but not on transfers. Then, our results provide evidence of the
(rather limited) negative impact of the legal system on its users through increased
transaction costs, and not of the (dominant) beneficial contribution of the legal sys-
tem, through its impact on transfers.
To summarize, the argument is that using the law improves user?s overall situa-
tion, but it also implies costs. Firms probably decide to use the legal system following
a cost-benefit analysis. However, our particular result is driven by the fact that we
only observe the costs (transaction costs) and not the benefits (transfers).
3.4.4 Additional results
3.4.4.1 The two-sided nature of the relationship-specific investment
As it has been discussed in the first chapter of this dissertation, the impact
on contract complexity of buyers? decision to invest in relationship-specific assets is
significantly different from that of sellers?. While the former lead to a decrease in
the complexity of contracts governing sales agreements, the latter make contracts
more complete (the arguments rests on partners? default unbalanced exposure to
10Contract renegotiations provide a good example. In one scenario, a firm faced with a false
claim of dissatisfaction submitted by an opportunistic partner may yield to the request and make a
transfer to that partner, or can contest the claim, leading to renegotiations. Conversely, in another
scenario, an opportunistic firm will increase its chances of receiving a transfer by turning to legal
technicalities which trigger contract renegotiations. In both cases, renegotiations (which increase
transaction costs) may improve the overall situation, by reducing transfers (in the first scenario) or
by increasing them (in the second scenario).
162
the opportunism of the other due to particular transition conditions such as lack of
external financing and high inflation).
Since opportunistic behavior plays a key role in the transaction costs discussion,
a similar significant difference between the impact of buyers? and sellers? specific
investments is expected on transaction costs. Thus, we separate buyers? from sellers?
relationship-specific investments, and two variables now replace the unique specific
investment variable previously used.
Given the different specification of our model, we again employ the sample
selection model and estimate it by ML and by the two-step procedure. Results from
Table 3.6 displayed at the end of this dissertation chapter show that sample selection
bias is not present (neither ?, the correlation between the two error terms, nor ?,
the coefficient of the inverse Mills ratio, are significantly different from zero). Hence,
as before, we can safely proceed and estimate the main equation by OLS using the
positive transaction costs responses, and the selection equation by probit using the
transaction costs dummy variable.
Table 3.7 contains the main results when we make the distinction between buy-
ers? and sellers? specific investments. The overall picture we obtain is similar to our
previous findings: the results of the sample selection equation are somewhat closer
to existing theory than are those of the main equation. Regarding the particular
aspect of interest at this stage, the first column indicates that in the main equation
both estimated coefficients for the relationship-specific investment variables are pos-
itive. However, the second column confirms our expectations about opposing signs,
163
even though only one of the two estimates on that column passes the usual statisti-
cal significance levels. In addition, the Wald test of equality of the two coefficients
yields ?2(1) = 4.04 and p-value = 0.045. The effects on the probability of existence
of transaction costs match those from the first chapter of this dissertation: buyers?
relationship-specific investments tend to decrease the chances of facing transaction
costs, while sellers? increase them.
3.4.4.2 The potentially endogenous nature of some variables
This dissertation chapter has shown that the factors proposed by existing theory
have the expected influence on the probability of existence and the size of transaction
costs. However, since many of these determinants themselves represent firm deci-
sions, one could question the reliability of our estimates. That is, endogeneity bias
may affect our results, especially the estimated coefficients of relationship-specific
investment, frequency of transaction, complexity of transaction, and companies? or-
ganization of legal matters.
In our setup, the endogeneity problem may arise for two reasons: reverse causal-
ity and omitted variables. In simplified terms, our approach is to consider a vector X
of explanatory variables which affect a measure of transaction costs, denoted Y. If,
under a specific scenario, Y can act as a determinant for one or more components of
X, then this reverse causality will lead to biases in the estimates of those components.
In addition, if there exists an unmeasured factor, denoted Z, which simultaneously
influences Y and a component of X, then this omission will lead to our estimate of
164
the X component being even asymptotically biased. We will show below that the two
channels through which endogeneity may affect our estimates are in fact limited.
The first argument rests on the dominant ex-post nature of the transaction
costs information we collect. As discussed in the second chapter of this dissertation,
contrary to our intentions, the large majority of managers answering our transaction
costs question may have thought of the situation at the end of the agreement, thus
the collected information is likely to represent transaction costs after the agreement
has ended. This removes the possibility of reverse causality since the explanatory
variables susceptible of being endogenous represent decisions taken before or during
the agreement implementation. Our second argument refers to the vector of right
hand side variables (denoted X above). We believe that we have successfully captured
all factors theory proposes, either directly or indirectly (through proxies), thus greatly
reducing the possibility that a significant factor Z has been omitted (this rests on our
choices of proxies being valid).
The two arguments above hold for all the possibly endogenous variables we
have listed. However, for the complexity of transaction there are additional argu-
ments, arising from the indirect measurement of the concept. Namely, we rely on a
proxy (the frequency the company uses customized contracts for all its operations),
and the information is provided by the legal manager. As in the first chapter of
this dissertation, technology and sector of activity are considered to be exogenously
determined for firms. These two factors influence the complexity of transactions for
the company as a whole, which in turn has an impact on the complexity of the inves-
165
tigated agreement. Our choice of proxy focuses on this channel of influence, which is
free of endogeneity11.
We acknowledge the fact that the reader may have residual doubts about the ar-
guments presented above. Ideally, we would run exogeneity tests for each of the four
possibly endogenous variables. Unfortunately, our survey of Romanian companies
does not offer good instruments to be used in such tests for all suspicious variables.
Nonetheless, there is one situation where endogeneity can be tested, namely for buy-
ers? and sellers? decisions to invest in relationship-specific assets.
We employ the same instrumental variables as in the first chapter of this disser-
tation: the interaction effects between dummy variables for the number of potential
trading partners and for companies? sectors of activity. Regarding the former, for
buyers? specific investments we use a dummy representing the number of potential
sellers, while for sellers? decision we use a dummy variable accounting for the num-
ber of potential buyers. The information on sector of activity is captured by four
dummy variables labeled Heavy industry, Light industry, Construction, and Other
sectors. The appropriateness of using these variables as instruments for partners?
specific investments has also been discussed in the first chapter.
We test the exogeneity of buyers? and sellers? relationship-specific investments
separately for the main and selection equation. For the former, we employ a test in-
troduced by Hausman (1978, 1983) ? the heteroskedasticity-robust joint F-test yields
p-value = 0.887. For the latter we rely on a test proposed by Rivers and Vuong (1988)
11Transaction-specific issues probably also affect the complexity of the investigated agreement,
but our proxy does not capture this (endogenous) information.
166
? the joint Wald test returns p-value = 0.901. These results lead to the conclusion
that we cannot reject the exogeneity of the two specific investment variables12.
To conclude, we have argued above that endogeneity bias should not be a major
concern in the current analysis. The reason is that in our setup two main features
leading to such biases, reverse causality and omitted variables, play a minor role. In
the one situation in which we can test for endogeneity, the results support our view,
thus strengthening our confidence in the results of this chapter.
3.5 Conclusions
The second chapter of the dissertation has indicated that, despite the concept?s
importance, there is still no commonly accepted way to assess the size of transaction
costs at agreement level. The lack of a consistent measure led to the vast majority of
the TCE empirical work focusing on the impact of various factors on the governance
structure of agreements rather than directly on the size of transaction costs.
The present chapter has first gathered all factors assumed by theory to have
an impact on transaction costs. Then, these factors? effect on the probability of
existence and size of transaction costs has been investigated. In this attempt, we rely
on the transaction costs measure we have collected in a study of Romanian companies
which included the proposed survey question discussed in the second chapter. Results
12The same conclusion is reached when, similarly to first chapter?s approach, we add to the set
X two variables capturing buyers? and sellers? dependence on their partners, thus countering the
possibility that our instruments proxy some post-specific investment market thinness. The reason
we have not included these variables in the set of explanatory variables from the beginning of the
analysis is that the investigated theory does not mention them.
167
indicate that theory is very successful in predicting the existence of transaction costs,
and only moderately successful in predicting the size of these costs when they are
incurred. We believe that the slight difference in strength between the two sets of
results deserves more deliberation, and present below some thoughts on why this
difference may arise.
First, this difference may arise naturally from the act of research; a theory
can be expected to be more successful regarding something that is easier to predict,
such as the existence of a phenomenon, but less successful regarding finer details of
that phenomenon13. Second, a technical aspect may explain the observed difference
in results. As one would expect, detail and accuracy level of the data entering the
analysis are likely to impact on the quality of the outcome. Despite the multitude
of transaction features we observe, since our data are generated by a survey most
of the variables we employ are dummies, and this may also explain the difference in
strength between the two sets of results14.
Lastly, our results may be an indicator that the existing TCE theory needs to
be improved, especially in what concerns its predictions on the size of transaction
costs. However, attempts to refine a theory are futile without proper data (this is
why we consider the second chapter of this dissertation to take a much needed step
in the transaction costs area of research).
13A weather-related analogy may persuade the reader: a meteorologist may find it easier to predict
whether rain will fall, but will probably find it harder to predict the precise amount of water falling
should the rain occur.
14If we continue the analogy above, the meteorologist can easily predict rain if he observes dark
clouds (hence little information suffices) but he will need more detailed information to predict the
amount of the rainfall (simply put, he will need to know how dark the clouds are).
168
Apart from the factors proposed by theory the present dissertation chapter has
investigated the impact of other transaction costs determinants. Most importantly,
firms? reliance on the formal legal system was shown to have a significant effect on
these costs (we have provided arguments for the apparently surprising results). The
age of the firms and whether they both belong to the same legal system and economic
environment appear to have the expected impact on transaction costs. Additional re-
sults include an analysis of the two-sided nature of the relationship-specific investment
and a discussion on the endogeneity of various factors.
Figure 3.1: Distribution of answers to the transaction costs question
Note: For presentation purposes, this figure only shows the positive values recorded (variable Trans-
action costs: Positive responses). Two hundred ninety-one zero values have also been recorded.
169
Table 3.1: Descriptive statistics
Number Mean Standard Min Max
Variables: of obs. deviation
Transaction costs: All responses 501 3.488 6.518 0 60
Transaction costs: Positive responses 210 8.321 7.825 1 60
Transaction costs: Dummy variable 501 0.419 0.494 0 1
Specific investment 502 0.333 0.472 0 1
Buyer specific investment 506 0.117 0.321 0 1
Seller specific investment 503 0.256 0.437 0 1
Exogenous market uncertainty 506 0.233 0.423 0 1
Four or more exchanges per month 508 0.189 0.392 0 1
Good easy to measure 506 0.494 0.500 0 1
Information dissemination among buyers 488 0.836 0.371 0 1
Information dissemination among sellers 497 0.803 0.398 0 1
Customized contracts used frequently 500 0.408 0.492 0 1
Distrustful sales/procurement manager 507 0.268 0.443 0 1
Both partners are Romanian 506 0.836 0.371 0 1
Respondent has some legal knowledge 508 0.606 0.489 0 1
One of the firms founded before 1990 504 0.524 0.500 0 1
Legal adviser?s office close to manager?s 502 0.534 0.499 0 1
Dependence on partner 489 0.579 0.494 0 1
Buyer dependence on partner 496 0.417 0.494 0 1
Seller dependence on partner 492 0.476 0.500 0 1
Few potential buyers 478 0.203 0.403 0 1
Few potential sellers 483 0.511 0.500 0 1
Heavy industry 508 0.280 0.449 0 1
Light industry 508 0.262 0.440 0 1
Construction 508 0.201 0.401 0 1
Other sectors 508 0.258 0.438 0 1
Few potential buyers in heavy industry 478 0.065 0.247 0 1
Few potential buyers in light industry 478 0.050 0.219 0 1
Few potential buyers in construction 478 0.033 0.180 0 1
Few potential buyers in other sectors 478 0.054 0.227 0 1
Many potential buyers in heavy industry 478 0.218 0.413 0 1
Many potential buyers in light industry 478 0.209 0.407 0 1
Many potential buyers in construction 478 0.167 0.374 0 1
Many potential buyers in other sectors 478 0.203 0.403 0 1
Few potential sellers in heavy industry 483 0.176 0.381 0 1
Few potential sellers in light industry 483 0.135 0.342 0 1
Few potential sellers in construction 483 0.068 0.253 0 1
Few potential sellers in other sectors 483 0.133 0.339 0 1
Many potential sellers in heavy industry 483 0.106 0.308 0 1
Many potential sellers in light industry 483 0.128 0.335 0 1
Many potential sellers in construction 483 0.128 0.335 0 1
Many potential sellers in other sectors 483 0.126 0.333 0 1
170
Table 3.2: Censored regression estimates of two different specifications
Tobit
Transaction costs:
Independent variables: All responses
Y1 Y1
Specific investment 3.090 2.971
(1.382)** (1.361)**
Exogenous market uncertainty 4.708 5.096
(1.512)*** (1.491)***
Four or more exchanges per month -4.119 -4.044
(1.790)** (1.760)**
Good easy to measure -2.980 -3.271
(1.328)** (1.308)**
Information dissemination among buyers -0.971 -1.135
(1.956) (1.921)
Information dissemination among sellers -2.170 -2.002
(1.844) (1.811)
Customized contracts used frequently -1.192 -0.972
(1.346) (1.324)
Distrustful sales/procurement manager 2.291 2.845
(1.449) (1.426)**
Both partners are Romanian -2.831
(1.734)
Respondent has some legal knowledge 1.734
(1.344)
One of the firms founded before 1990 -3.536
(1.324)***
Legal adviser?s office close to manager?s 4.197
(1.323)***
Constant -0.307 0.600
(2.044) (2.728)
Observations 460 448
Censored observations 263 253
Log-likelihood -923.941 -900.066
Standard errors displayed in parentheses.
* significant at 10%; ** significant at 5%; *** significant at 1%
171
Table 3.3: Sample selection estimates of the complete specification
Heckman ML Heckman Two-step
Transaction costs: Transaction costs:
Positive Dummy Positive Dummy
Independent variables: responses variable responses variable
Y2 Y3 Y2 Y3
Specific investment 1.223 0.258 -2.582 0.256
(1.167) (0.133)* (7.763) (0.132)*
Exogenous market uncertainty 0.418 0.535 -7.467 0.535
(1.189) (0.166)*** (15.207) (0.146)***
Four or more exchanges per month -2.586 -0.274 1.605 -0.273
(1.145) (0.165)* (8.848) (0.163)*
Good easy to measure -1.007 -0.290 3.336 -0.290
(1.167) (0.129)** (8.716) (0.125)**
Information dissemination among buyers 1.292 -0.218 4.556 -0.219
(1.615) (0.200) (7.460) (0.186)
Information dissemination among sellers -4.311 0.071 -5.392 0.073
(1.679)*** (0.181) (4.483) (0.176)
Customized contracts used frequently 0.297 -0.117 2.057 -0.116
(1.255) (0.144) (4.459) (0.127)
Distrustful sales/procurement manager -0.214 0.321 -4.951 0.322
(1.050) (0.155)** (9.502) (0.137)**
Both partners are Romanian -4.867 0.024 -5.111 0.024
(2.120)** (0.170) (3.909) (0.169)
Respondent has some legal knowledge 1.780 0.048 1.045 0.049
(1.044) (0.117) (3.261) (0.129)
One of the firms founded before 1990 1.094 -0.433 7.584 -0.432
(1.077) (0.144)*** (12.522) (0.127)***
Legal adviser?s office close to manager?s 0.477 0.408 -5.711 0.407
(1.142) (0.143)*** (12.006) (0.126)***
Constant 13.668 -0.159 35.142 -0.159
(3.046)*** (0.269) (40.246) (0.262)
? (correlation between the error terms) -0.075a
? (inverse Mills ratio) -24.142
(43.667)
Observations/Censored observations 448/253 448/253
Clusters 240
Log-likelihood -952.614
Standard errors displayed in parentheses, adjusted when clustering was used (a cluster is a firm).
* significant at 10%; ** significant at 5%; *** significant at 1%
a A discussion of ??s standard error is provided in Subsection 3.4.2.1.
172
Table 3.4: Sample selection estimates of the reduced specification
Heckman ML Heckman Two-step
Transaction costs: Transaction costs:
Positive Dummy Positive Dummy
Independent variables: responses variable responses variable
Y2 Y3 Y2 Y3
Specific investment 1.342 0.244 1.334 0.244
(1.642) (0.133)* (1.231) (0.131)*
Exogenous market uncertainty 0.595 0.546 0.577 0.546
(2.440) (0.165)*** (1.594) (0.145)***
Four or more exchanges per month -2.731 -0.289 -2.720 -0.289
(2.021) (0.164)* (1.737) (0.162)*
Good easy to measure -1.118 -0.298 -1.108 -0.298
(2.148) (0.129)** (1.279) (0.124)**
Information dissemination among buyers 1.159 -0.231 1.163 -0.230
(1.424) (0.199) (1.594) (0.185)
Information dissemination among sellers -4.107 0.069 -4.108 0.068
(1.705)** (0.181) (1.522)*** (0.175)
Customized contracts used frequently 0.232 -0.116 0.235 -0.117
(1.230) (0.146) (1.140) (0.126)
Distrustful sales/procurement manager -0.027 0.310 -0.034 0.310
(1.154) (0.154)** (1.269) (0.137)**
Both partners are Romanian -4.649 -4.645
(2.267)** (1.484)***
Respondent has some legal knowledge 1.900 1.901
(1.049)* (1.139)*
One of the firms founded before 1990 -0.422 -0.421
(0.149)*** (0.124)***
Legal adviser?s office close to manager?s 0.400 0.401
(0.156)*** (0.125)***
Constant 13.470 -0.091 13.511 -0.092
(5.653)** (0.235) (3.276)*** (0.205)
? (correlation between the error terms) 0.023
(0.860)
? (inverse Mills ratio) 0.124
(3.146)
Observations/Censored observations 450/255 450/255
Clusters 240
Log-likelihood -954.345
Standard errors displayed in parentheses, adjusted when clustering was used (a cluster is a firm).
* significant at 10%; ** significant at 5%; *** significant at 1%
173
Table 3.5: OLS and probit estimates
OLS Probit
Transaction Transaction costs:
costs: Dummy variable
Positive Estimated Mean of
Independent variables: responses coefficients marginal effects
Y2 Y3 Prob(Y3 = 1)
Specific investment 1.301 0.244 0.089
(1.186) (0.131)* (0.048)*
Exogenous market uncertainty 0.532 0.546 0.201
(1.197) (0.166)*** (0.059)***
Four or more exchanges per month -2.681 -0.289 -0.102
(1.152)** (0.165)* (0.057)*
Good easy to measure -0.990 -0.298 -0.108
(1.181) (0.129)** (0.046)**
Information dissemination among buyers 1.172 -0.230 -0.084
(1.579) (0.199) (0.072)
Information dissemination among sellers -4.128 0.068 0.025
(1.746)** (0.180) (0.064)
Customized contracts used frequently 0.250 -0.117 -0.042
(1.280) (0.144) (0.051)
Distrustful sales/procurement manager -0.066 0.310 0.112
(1.015) (0.154)** (0.056)**
Both partners are Romanian -4.641
(2.157)**
Respondent has some legal knowledge 1.804
(1.083)*
One of the firms founded before 1990 -0.421 -0.153
(0.141)*** (0.050)***
Legal adviser?s office close to manager?s 0.401 0.145
(0.143)*** (0.051)***
Constant 13.648 -0.092
(2.790)*** (0.231)
Observations 197 450
Clusters 136 240
R-squared 0.119 0.080
Log-likelihood -283.394
Clustered standard errors displayed in parentheses, where each cluster is a firm.
* significant at 10%; ** significant at 5%; *** significant at 1%
174
Table 3.6: Distinction between the two specific investments (I)
Heckman ML Heckman Two-step
Transaction costs: Transaction costs:
Positive Dummy Positive Dummy
Independent variables: responses variable responses variable
Y2 Y3 Y2 Y3
Buyer specific investment 1.189 -0.086 1.192 -0.086
(2.003) (0.184) (1.801) (0.195)
Seller specific investment 1.150 0.394 1.144 0.393
(1.329) (0.142)*** (1.416) (0.142)***
Exogenous market uncertainty 0.480 0.539 0.472 0.539
(1.327) (0.167)*** (1.604) (0.145)***
Four or more exchanges per month -2.689 -0.284 -2.683 -0.284
(1.289)** (0.164)* (1.735) (0.162)*
Good easy to measure -0.986 -0.289 -0.982 -0.289
(1.387) (0.129)** (1.271) (0.125)**
Information dissemination among buyers 1.154 -0.237 1.157 -0.237
(1.494) (0.202) (1.600) (0.185)
Information dissemination among sellers -4.055 0.059 -4.055 0.059
(1.713)** (0.183) (1.522)*** (0.176)
Customized contracts used frequently 0.161 -0.112 0.162 -0.112
(1.209) (0.145) (1.139) (0.127)
Distrustful sales/procurement manager -0.064 0.319 -0.067 0.319
(0.967) (0.154)** (1.280) (0.137)**
Both partners are Romanian -4.511 -4.510
(2.106)** (1.481)***
Respondent has some legal knowledge 1.850 1.850
(1.061)* (1.142)
One of the firms founded before 1990 -0.414 -0.415
(0.145)*** (0.124)***
Legal adviser?s office close to manager?s 0.390 0.389
(0.147)*** (0.126)***
Constant 13.675 -0.096 13.695 -0.094
(3.053)*** (0.232) (3.332)*** (0.206)
? (correlation between the error terms) -0.026
(0.248)
? (inverse Mills ratio) -0.220
(3.216)
Observations/Censored observations 450/255 450/255
Clusters 240
Log-likelihood -952.057
Standard errors displayed in parentheses, adjusted when clustering was used (a cluster is a firm).
* significant at 10%; ** significant at 5%; *** significant at 1%
175
Table 3.7: Distinction between the two specific investments (II)
OLS Probit
Transaction Transaction costs:
costs: Dummy variable
Positive Estimated Mean of
Independent variables: responses coefficients marginal effects
Y2 Y3 Prob(Y3 = 1)
Buyer specific investment 1.180 -0.086 -0.030
(2.101) (0.184) (0.065)
Seller specific investment 1.184 0.393 0.143
(1.231) (0.141)*** (0.051)***
Exogenous market uncertainty 0.536 0.538 0.197
(1.199) (0.167)*** (0.060)***
Four or more exchanges per month -2.720 -0.284 -0.100
(1.166)** (0.166)* (0.057)*
Good easy to measure -0.928 -0.289 -0.103
(1.176) (0.129)** (0.046)**
Information dissemination among buyers 1.139 -0.237 -0.085
(1.596) (0.202) (0.073)
Information dissemination among sellers -4.071 0.059 0.021
(1.766)** (0.183) (0.065)
Customized contracts used frequently 0.161 -0.112 -0.040
(1.255) (0.145) (0.051)
Distrustful sales/procurement manager -0.054 0.319 0.115
(1.016) (0.155)** (0.056)**
Both partners are Romanian -4.530
(2.133)**
Respondent has some legal knowledge 1.753
(1.080)*
One of the firms founded before 1990 -0.415 -0.150
(0.143)*** (0.051)***
Legal adviser?s office close to manager?s 0.389 0.140
(0.145)*** (0.051)***
Constant 13.562 -0.094
(2.792)*** (0.232)
Observations 197 450
Clusters 136 240
R-squared 0.121 0.087
Log-likelihood -281.259
Clustered standard errors displayed in parentheses, where each cluster is a firm.
* significant at 10%; ** significant at 5%; *** significant at 1%
176
Appendices to Chapter 1
Appendix A: Further details on the survey
We use a survey of two hundred fifty-four Romanian companies conducted in
2001 in an attempt to understand the way Romanian enterprises run their business,
use the formal legal system, and the alternatives they rely on when interacting with
each other. Located in and closely around the twelve largest cities of the country, the
targeted companies were medium- and large-sized ones, which undertake activities in
various productive sectors of activity (services were excluded, given the difficulty to
phrase generic survey questions applicable to both goods and services).
Small companies were not included in the investigated sample since the goal of
the survey was to understand whether law played a role in the day-to-day operations
of the economy (it is commonly accepted that small companies frequently use infor-
mal arrangements to conduct their exchanges). Small companies were also excluded
because of the objective of obtaining a wide variety of information from different
company officials, specialized in specific aspects of the transactional process.
Four different persons in each company were questioned in face-to-face inter-
views, and different survey instruments were used. Given the multitude of questions
and the fact they were addressed to top management, the duration of the interview
had to be limited. This was accomplished by asking for quick estimates rather than
precise figures that had to be looked up in books. Also, closed-ended questions were
177
frequently used. This is why our analysis is based mostly on dummy instead of
continuous variables.
As already mentioned, an important feature of the questionnaires addressed to
the sales and procurement department managers is a twofold symmetry. We provide
below two examples which illustrate the symmetry between and the symmetry within.
The managers of the procurement departments were asked whether their firms
(acting as buyers in the specific agreements) have made relationship-specific invest-
ments during the agreement in discussion. The same respondents were asked their
opinion on whether the partners in the agreement (the sellers) have made such in-
vestments. Similarly, the managers of the sales departments were asked whether their
firms (acting as sellers) have made specific investments when selling to the partner
in the specific agreements, and whether their partners (the buyers) have made such
investments.
In a similar fashion, we have information from both sides of the agreement on
the possible requests for contract renegotiation. The precise questions addressed to
the managers of the procurement and sales departments are presented below.
The procurement managers were asked:
Did your enterprise ask for the agreement to be renegotiated during im-
plementation? YES/NO
Did the supplier ask for the agreement to be renegotiated during imple-
mentation? YES/NO
The sales managers were asked:
178
Did your enterprise ask for the specific agreement to be renegotiated during
implementation? YES/NO
Did the customer ask for the specific agreement to be renegotiated during
implementation? YES/NO
A final point concerns information from the general manager and legal adviser
surveys. As it is useful to match information on a specific agreement with non-
transaction-specific characteristics (data on firm size, firm age etc.), information from
the general manager and legal adviser surveys need to be used. Since this information
is duplicated in the sample of agreements, the interpretation of these variables is
altered. For example, instead of discussing about sellers? size, we comment on the
influence of one of the two partners? size, regardless it is the seller or the buyer. Apart
from interpretation, since same information is used twice in the sample, clustering has
to be used, in order to account for this duplication and obtain corrected coefficients
and statistics in regressions.
Appendix B: Further details on the variables
The variables in the equation of interest
Contract complexity
Contract complexity variables are indexes created by assigning scores to features
of the contract and the contracting process. Buyer and seller contract complexity vari-
ables share significant information in common, referring to the form of the contract:
179
? Respondents were asked whether written contracts were used at all in their in-
teraction with the partner in the specific agreement. If written contracts were
used, we increased the contract complexity score by 1 if the respondent indi-
cated that ?written contracts were used in the transaction at all, but important
elements of the transaction were subject to additional oral agreements? and
by 1.5 if ?written contracts were used in the transaction and, moreover, they
covered all the important elements of the transaction?15;
? We were interested whether the legal staff was involved in writing the contracts
covering the specific agreement. If so, information on the duration of discussions
with legal staff was extracted, and contract complexity score increased by 0.5
if discussions took between 2 and 8 hours, and by 1 if they took more than 8
hours;
? Contract length, in number of pages, was recorded. Contract complexity vari-
ables increased by 0.5 if the written contract is reported to include between 3
and 5 pages, by 1 it has between 6 and 10 pages, and by 1.5 if it has more than
10 pages;
? Respondents were asked whether the contract governing the specific agreement
was authenticated or not. In Romania, contract authentication makes con-
tracts more easily enforceable in court. Authentication is optional, is made by
15We do not check whether the contracts were indeed properly designed to cover all the important
elements of the transaction. Such an endeavor would be nearly impossible as it would basically
require access to the actual contracts. However, this should not be an issue. We are mainly interested
in our respondents? view on their contracts since respondent?s beliefs are what guide their decisions.
180
a notary public, and leads to extra expenses and greater precision of the con-
tract. Complexity scores were increased by 1 if the contract in discussion was
authenticated;
? Another feature we observe on contract form is whether it was specifically de-
signed for the specific transaction, case in which contract complexity increases
by 1, as opposed to being a form contract provided by one of the partners.
What differentiates the buyer and seller contract complexity variables is a main
clause each may contain. Buyer contract complexity score increases by 1 if a penalty
clause for late delivery was included in the written contract, while seller contract
complexity score increases by 1 if a penalty clause for late payment was included in
the written contract. The delivery penalty is added to the buyer index, given that we
use the index in regressions that follow existing work in relating this index to buyer
characteristics. A similar argument applies to the penalties for late payment and the
seller index. Minor flaws were encountered in respondents? answers: some indicated
no written contracts were used at all but also that a penalty clause was used in a
written contract. In these cases we only added 0.75 to the contract complexity score.
The overall contract complexity variable captures all collected information (both
on the form of the contract and on the two penalty clauses that may be included). The
histograms of Figure 1.1 presented at the end of the first chapter show the patterns
in the data.
Relationship-specific investment
To construct the buyer specific investment dummy variable we have asked our
181
respondents whether the buyer had to undertake a substantial amount of special
investment to be able to use the product when supplied by the seller in the specific
agreement rather than an alternative supplier. The given options were:
1. No;
2. Yes, a small amount of special investment;
3. Yes, a significant amount of special investment;
4. Yes, a large amount of special investment.
Buyer specific investment dummy is equal to one if the respondent indicated any
amount of special investment is needed (answers 2, 3 or 4 above), zero otherwise.
The following pair of questions was used for the seller relationship-specific in-
vestment variable:
Were any elements of this product custom-made for the specific needs of
the customer? YES/NO (If ?NO? then skip the next question).
At what cost could the product be modified to sell to other enterprises if
the seller had surplus amounts?
1. Virtually no cost;
2. Small cost;
3. Moderate cost;
4. High cost;
5. Prohibitive cost (could not be sold to other enterprises).
We have constructed the seller specific investment dummy to equal one if any relationship-
182
specific investment has been made (answer ?Yes? to the first question followed by
answer 2, 3, 4, or 5 to the second one).
Quality of the courts
Legal advisers of each company16 were asked to rate the quality of the com-
mercial section of the local Tribunal courts on eight different dimensions. The exact
question they were addressed was:
Below is a list of problems that might arise when filing suit in the commer-
cial section of the Tribunal. Please evaluate how serious an obstacle each
of these potential problems appears to you when you are considering filing
suit. Give answers on a scale from 0 to 10. A ?0? means the potential
problem is not an actual problem. A ?10? means that the problem is so
great that it alone effectively prevents your using the court.
1. Filing a claim is expensive
2. Court procedures are complex
3. Legal counsel is expensive or not available
4. Judges are not impartial
5. Judges are not knowledgeable about market transactions
6. The time between filing a claim and obtaining a judgment is long
7. Judgments of the court are not executed
8. We are afraid our business secrets will become public knowledge
Without placing any weights on the different eight sub-parts, the answers were
16Because the information comes from the legal adviser?s questionnaire, it reflects the interpreta-
tion of only one side of the transaction.
183
added up for each respondent, leading to an overall score ranging from 0 to 80.
Then, the quality of the courts dummy equals one if the court is considered to offer
acceptable services (the above computed score is lower than 40), 0 otherwise.
Information dissemination
Information dissemination among sellers is a dummy variable, equal to one if
suppliers of the same good would learn about the buyer in the specific agreement not
paying its obligations, zero otherwise. The following question was included in the
procurement manager questionnaire:
If your enterprise did not pay its obligations under this agreement, do you
think that other suppliers would learn about this? YES/NO
A similar question was included in the sales manager questionnaire.
If this customer did not pay its obligations under this agreement to you,
do you think that other enterprises like yours would learn about this?
YES/NO
Information dissemination among buyers is a dummy variable, which equals
one if other buyers of the good would learn about the seller in the specific agreement
defaulting its obligations, zero otherwise. Similar questions to the ones above were
used to extract this information.
First agreement
Both the sales and the procurement managers were asked whether the chosen
specific agreement is the first one the company has entered into with that partner
184
(buyer or seller, depending on respondent?s position in the interviewed company).
Based on their answers, the first agreement dummy variable was constructed.
Exogenous market uncertainty
This piece of information is provided by the general manager of each com-
pany participating in our study, therefore this variable is used twice in the sample of
agreements, once representing buyers? and once sellers? opinion on the importance of
unpredictable changes on the area of activity.
To focus on truly exogenous events, we have restricted attention to two main
sources of unpredictability: weather-induced variation in demand or supply and prob-
lems in transportation links that cause changes in the level of demand or supply.
Exogenous market uncertainty is a dummy variable which equals one if any of these
sources of uncertainty is considered to be very important for enterprises in that par-
ticular area of business, zero otherwise (we have tried to capture general experience
of firms in a that domain, rather than very specific experience of the interviewed
company).
Dependence on partner
Buyer dependence on partner is a dummy variable which equals one when, in
case the supplier fails to deliver the contracted goods, the buyer either cannot find
a replacing seller, or finding another seller would take him a long period of time (a
month or more). Conversely, buyer dependence on partner equals zero if the buyer is
able to find another supplier in a short period of time.
Seller dependence on partner is a dummy variable constructed in a similar fash-
185
ion; equals one if the seller in the specific agreement depends on his partner since, in
case the buyer refuses delivery of the order, finding an alternative buyer would take
a month or more, or such an alternative could not be found, zero otherwise.
The instrumental variables
The instrumental variables are constructed as the interaction of two sets of
dummies, representing the sector of activity and the market thickness.
Sector of activity
Several dummy variables were constructed to reflect the sector of activity of
transacted good. These dummies are:
? Construction ? Dummy variable, equals one if the good traded in the specific
agreement belongs to the construction sector, zero otherwise;
? Heavy industry ? Dummy variable, equals one if the good traded in the spe-
cific agreement belongs to the metallurgic, metallic constructions, tools and
equipment, or office supplies sectors, zero otherwise;
? Light industry ? Dummy variable, equals one if the good traded in the specific
agreement belongs to sectors such as the textile industry, clothing industry, or
leather and shoe industry, zero otherwise;
? Other sectors ? Dummy variable, equals one if the good traded in the specific
agreement belongs to other productive sectors than the ones presented above,
zero otherwise. The list of other sectors includes 26 additional industries, such as
186
agriculture, forestry, minerals processing, coal processing, food and beverages,
fiber industry, and transportations.
Market thickness
Few potential buyers is a dummy variable, which equals one if there are relatively
few potential buyers (10 or less) in Romania for the transacted good, zero otherwise.
Few potential sellers is a similarly constructed dummy variable, which equals one
if there are relatively few potential sellers (10 or less) of the transacted good, zero
otherwise.
Market thickness by sector
The market thickness dummy variables have been interacted with the sector
of activity dummies, leading to an extended set of variables which constitute the
instruments for seller and buyer specific investments in our regressions.
For seller relationship-specific investment the instrumental variables are: few
potential buyers in construction, few potential buyers in heavy industry, few potential
buyers in light industry, few potential buyers in other sectors, many potential buyers
in construction, many potential buyers in heavy industry, many potential buyers in
light industry, and many potential buyers in other sectors. The meaning of these
variables is clearly reflected by their names.
Eight other dummy variables, representing the interaction between the sector
of activity dummies and the one indicating the number of potential sellers, are used
as instruments for the buyer relationship-specific investment in our regressions.
187
The variables used for robustness checks
State has control is a dummy variable which equals one if Romanian state
owns 50% or more of the shares of any one of the parties, zero otherwise.
State owns shares is a dummy variable which equals one if Romanian state
owns shares of any one of the parties, zero otherwise. Some state origins is a
dummy variable which equals one if the origins of one enterprise lie in the state, even
quite indirectly, zero otherwise. These connections to the state may be evident (firms
may be the result of the privatization process) or less direct (the company founders
may have undertaken similar activities in a state enterprise).
In same region is a dummy variable which equals one if the parties engaged
in the specific agreement are located in the same county, zero otherwise (partners are
both located in Romania, but in different counties).
Firm founded before 1990 is a dummy variable which equals one if any of
the trading partners has been founded before 1990 (before transition began), zero
otherwise.
Large firm is a dummy variable which equals one if one of the companies
employs more than 250 people, zero otherwise.
Significant buyer prepayment is a dummy variable which equals one if the
original agreement stipulated that the buyer had to pay at least 30% of the total bill
before provision of the good or service, zero otherwise.
188
Appendix C: A different use of the dependence variables
We conclude the appendix to the first chapter with the presentation of some
additional results. They are obtained by using differently than before the variables
capturing information on firms? dependence on their current trading partners.
We have argued, especially in Subsection 1.8.1, that the dependence on partner
variables should belong (only) in the second stage equation to explain part of the
contract complexity partners choose for their transactions. One reason for this ap-
proach was given by the similarity between the concepts captured by these and the
relationship-specific investment variables. Another argument is that the inclusion of
the dependence variables in the second stage equation removes the possibility that
our instrumental variables proxy some post-specific investment market thinness.
However, for those readers who do not agree with our view, we present below
the results when the dependence on partner variables are used in a manner different
than before. We first choose to remove these variables from the set of explanatory
variables, hence Tables 1.15, 1.16, and 1.17 show the results when dependence on
partner does not appear in either stage. Then, we decide to use these variables both
as instruments (for relationship-specific investment) and independent variables (for
contract complexity). Tables 1.18 and 1.19 present the findings when the dependence
on partner variables are used in both stages. Finally, Tables 1.20 and 1.21 display
the results when the dependence on partner variables are used only as instruments
for the decision to invest in relationship-specific assets.
As the tables below indicate, the different use of the dependence variables does
189
not lead to great changes in the magnitude and statistical significance of the coef-
ficients of main interest. Namely, when its endogeneity is taken into account, the
impact of the relationship-specific investment on contract complexity maintains the
opposing signs for the buyer?s and seller?s side. The estimated coefficients are also
of roughly equal magnitude as before (and to each other on the two sides of the
agreement), while their high statistical significance is preserved. The OLS results we
obtain are also very similar to those we have previously shown.
Table 1.15: OLS estimates when dependence on partner is not used
Dependent variable:
Seller Buyer Overall
Independent variables: contract contract contract
complexity complexity complexity
Seller specific investment 0.742 0.805
(0.158)*** (0.193)***
Buyer specific investment 0.366 0.261
(0.237) (0.268)
Quality of the courts 0.306 0.382 0.342
(0.155)** (0.163)** (0.186)*
Information dissemination among sellers 0.157 0.168
(0.176) (0.241)
Information dissemination among buyers 0.274 0.174
(0.186) (0.255)
First agreement 0.109 0.188 0.099
(0.169) (0.176) (0.204)
Exogenous market uncertainty -0.136 -0.172 -0.180
(0.156) (0.169) (0.184)
Constant 2.953 2.806 3.436
(0.177)*** (0.188)*** (0.242)***
Observations 400 398 388
Clusters 232 232 231
R-squared 0.063 0.035 0.064
Clustered standard errors displayed in parentheses, where each cluster is a firm.
* significant at 10%; ** significant at 5%; *** significant at 1%
190
Table 1.16: Seller ML TE estimates when dependence on partner is not used
Main equation results First stage results
Seller Seller
Independent variables: contract Instrumental variables: specific
complexity investment
Seller specific investment 2.127 Few potential buyers in heavy industry 0.720
(0.439)*** (0.260)***
Quality of the courts 0.332 Few potential buyers in light industry -0.061
(0.160)** (0.557)
Information dissemination among sellers 0.081 Few potential buyers in construction 1.226
(0.182) (0.340)***
First agreement 0.074 Few potential buyers in other sectors 0.839
(0.169) (0.314)***
Market exogenous uncertainty -0.133 Many potential buyers in heavy industry 0.136
(0.163) (0.215)
Constant 2.710 Many potential buyers in light industry 0.100
(0.218)*** (0.241)
Many potential buyers in construction 0.602
(0.218)***
Constant -1.065
(0.163)***
Observations 380
Clusters 227
? (correlation between the error terms) -0.559
(0.111)***
Log-likelihood -873.159
Clustered standard errors displayed in parentheses, where each cluster is a firm. * significant at 10%; ** significant at 5%; *** significant at 1%
19
1
Table 1.17: Buyer ML TE estimates when dependence on partner is not used
Main equation results First stage results
Buyer Buyer
Independent variables: contract Instrumental variables: specific
complexity investment
Buyer specific investment -2.067 Few potential sellers in heavy industry -0.301
(0.333)*** (0.233)
Quality of the courts 0.410 Few potential sellers in light industry -0.608
(0.158)*** (0.260)**
Information dissemination among buyers 0.290 Few potential sellers in construction -1.126
(0.187) (0.382)***
First agreement 0.134 Few potential sellers in other sectors -0.702
(0.166) (0.256)***
Market exogenous uncertainty -0.187 Many potential sellers in heavy industry -0.832
(0.166) (0.272)***
Constant 3.092 Many potential sellers in light industry -0.332
(0.184)*** (0.268)
Many potential sellers in construction -0.729
(0.248)***
Constant -0.653
(0.181)***
Observations 379
Clusters 226
? (correlation between the error terms) 0.766
(0.076)***
Log-likelihood -812.403
Clustered standard errors displayed in parentheses, where each cluster is a firm. * significant at 10%; ** significant at 5%; *** significant at 1%
19
2
Table 1.18: Seller ML TE estimates when dependence on partner is used in both equations
Main equation results First stage results
Seller Seller
Independent variables: contract Instrumental variables: specific
complexity investment
Seller specific investment 1.787 Few potential buyers in heavy industry 0.569
(0.541)*** (0.289)**
Quality of the courts 0.323 Few potential buyers in light industry -0.168
(0.158)** (0.547)
Information dissemination among sellers 0.080 Few potential buyers in construction 0.968
(0.181) (0.380)***
First agreement 0.106 Few potential buyers in other sectors 0.601
(0.168) (0.328)*
Market exogenous uncertainty -0.078 Many potential buyers in heavy industry 0.074
(0.166) (0.218)
Seller dependence on partner 0.328 Many potential buyers in light industry 0.035
(0.173)* (0.246)
Constant 2.613 Many potential buyers in construction 0.432
(0.228)*** (0.263)*
Seller dependence on partner 0.346
(0.165)**
Constant -1.124
(0.174)***
Observations 373
Clusters 227
? (correlation between the error terms) -0.465
(0.164)***
Log-likelihood -851.290
Clustered standard errors displayed in parentheses, where each cluster is a firm. * significant at 10%; ** significant at 5%; *** significant at 1%
19
3
Table 1.19: Buyer ML TE estimates when dependence on partner is used in both equations
Main equation results First stage results
Buyer Buyer
Independent variables: contract Instrumental variables: specific
complexity investment
Buyer specific investment -1.933 Few potential sellers in heavy industry -0.274
(0.360)*** (0.253)
Quality of the courts 0.375 Few potential sellers in light industry -0.579
(0.160)** (0.276)**
Information dissemination among buyers 0.250 Few potential sellers in construction -1.094
(0.191) (0.394)***
First agreement 0.178 Few potential sellers in other sectors -0.665
(0.167) (0.283)**
Market exogenous uncertainty -0.149 Many potential sellers in heavy industry -0.820
(0.169) (0.279)***
Buyer dependence on partner 0.591 Many potential sellers in light industry -0.342
(0.167)*** (0.279)
Constant 2.895 Many potential sellers in construction -0.682
(0.202)*** (0.260)***
Buyer dependence on partner 0.161
(0.179)
Constant -0.747
(0.186)***
Observations 374
Clusters 224
? (correlation between the error terms) 0.733
(0.092)***
Log-likelihood -793.898
Clustered standard errors displayed in parentheses, where each cluster is a firm. * significant at 10%; ** significant at 5%; *** significant at 1%
19
4
Table 1.20: Seller ML TE estimates when dependence on partner is used only as an instrument
Main equation results First stage results
Seller Seller
Independent variables: contract Instrumental variables: specific
complexity investment
Seller specific investment 2.219 Few potential buyers in heavy industry 0.512
(0.400)*** (0.265)*
Quality of the courts 0.323 Few potential buyers in light industry -0.157
(0.159)** (0.513)
Information dissemination among sellers 0.069 Few potential buyers in construction 0.970
(0.181) (0.347)***
First agreement 0.090 Few potential buyers in other sectors 0.617
(0.169) (0.301)**
Market exogenous uncertainty -0.100 Many potential buyers in heavy industry 0.092
(0.164) (0.207)
Constant 2.673 Many potential buyers in light industry 0.059
(0.217)*** (0.239)
Many potential buyers in construction 0.485
(0.227)**
Seller dependence on partner 0.440
(0.144)***
Constant -1.185
(0.166)***
Observations 373
Clusters 227
? (correlation between the error terms) -0.589
(0.098)***
Log-likelihood -852.887
Clustered standard errors displayed in parentheses, where each cluster is a firm. * significant at 10%; ** significant at 5%; *** significant at 1%
19
5
Table 1.21: Buyer ML TE estimates when dependence on partner is used only as an instrument
Main equation results First stage results
Buyer Buyer
Independent variables: contract Instrumental variables: specific
complexity investment
Buyer specific investment -2.086 Few potential sellers in heavy industry -0.273
(0.332)*** (0.243)
Quality of the courts 0.392 Few potential sellers in light industry -0.568
(0.158)** (0.265)**
Information dissemination among buyers 0.277 Few potential sellers in construction -1.072
(0.189) (0.378)***
First agreement 0.140 Few potential sellers in other sectors -0.655
(0.168) (0.270)**
Market exogenous uncertainty -0.198 Many potential sellers in heavy industry -0.809
(0.166) (0.269)***
Constant 3.122 Many potential sellers in light industry -0.337
(0.189)*** (0.269)
Many potential sellers in construction -0.668
(0.253)***
Buyer dependence on partner -0.109
(0.153)
Constant -0.640
(0.182)***
Observations 374
Clusters 224
? (correlation between the error terms) 0.765
(0.074)***
Log-likelihood -799.623
Clustered standard errors displayed in parentheses, where each cluster is a firm. * significant at 10%; ** significant at 5%; *** significant at 1%
19
6
Appendix D: Other results using the overall contract complexity score
As mentioned at the end of Subsection 1.8.1, we display below the main results
when the overall contract complexity score is the dependent variable for all three
regressions (one for buyer?s side, one for seller?s, and one for both of them together).
Table 1.22: OLS estimates of the determinants of overall contract complexity
Dependent variable:
Independent variables: Overall contract complexity
Seller specific investment 0.689 0.651
(0.188)*** (0.191)***
Buyer specific investment 0.275 0.247
(0.236) (0.277)
Quality of the courts 0.334 0.343 0.314
(0.184)* (0.190)* (0.188)*
Information dissemination among sellers 0.333 0.210
(0.214) (0.241)
Information dissemination among buyers 0.348 0.106
(0.232) (0.271)
First agreement 0.092 0.222 0.133
(0.199) (0.194) (0.204)
Exogenous market uncertainty -0.109 -0.091 -0.097
(0.190) (0.191) (0.196)
Seller dependence on partner 0.580 0.400
(0.175)*** (0.197)**
Buyer dependence on partner 0.621 0.376
(0.177)*** (0.197)*
Constant 3.227 3.311 3.149
(0.228)*** (0.249)*** (0.270)***
Observations 390 390 371
Clusters 231 230 227
R-squared 0.085 0.055 0.092
Clustered standard errors displayed in parentheses, where each cluster is a firm.
* significant at 10%; ** significant at 5%; *** significant at 1%
197
Table 1.23: ML TE estimates of the impact of seller specific investment on overall contract complexity
Main equation results First stage results
Overall Seller
Independent variables: contract Instrumental variables: specific
complexity investment
Seller specific investment 1.808 Few potential buyers in heavy industry 0.715
(0.621)*** (0.280)***
Quality of the courts 0.387 Few potential buyers in light industry -0.009
(0.189)** (0.608)
Information dissemination among sellers 0.251 Few potential buyers in construction 1.126
(0.219) (0.375)***
First agreement 0.088 Few potential buyers in other sectors 0.746
(0.199) (0.335)**
Market exogenous uncertainty -0.093 Many potential buyers in heavy industry 0.106
(0.196) (0.226)
Seller dependence on partner 0.505 Many potential buyers in light industry 0.037
(0.175)*** (0.247)
Constant 3.051 Many potential buyers in construction 0.462
(0.281)*** (0.251)*
Constant -0.999
(0.169)***
Observations 373
Clusters 227
? (correlation between the error terms) -0.398
(0.167)**
Log-likelihood -916.546
Clustered standard errors displayed in parentheses, where each cluster is a firm. * significant at 10%; ** significant at 5%; *** significant at 1%
19
8
Table 1.24: ML TE estimates of the impact of buyer specific investment on overall contract complexity
Main equation results First stage results
Overall Buyer
Independent variables: contract Instrumental variables: specific
complexity investment
Buyer specific investment -2.318 Few potential sellers in heavy industry -0.197
(0.372)*** (0.238)
Quality of the courts 0.397 Few potential sellers in light industry -0.514
(0.186)** (0.271)*
Information dissemination among buyers 0.321 Few potential sellers in construction -0.989
(0.228) (0.378)***
First agreement 0.177 Few potential sellers in other sectors -0.576
(0.189) (0.249)**
Market exogenous uncertainty -0.111 Many potential sellers in heavy industry -0.770
(0.189) (0.262)***
Buyer dependence on partner 0.545 Many potential sellers in light industry -0.326
(0.175)*** (0.276)
Constant 3.679 Many potential sellers in construction -0.549
(0.239)*** (0.260)**
Constant -0.738
(0.182)***
Observations 374
Clusters 224
? (correlation between the error terms) 0.746
(0.085)***
Log-likelihood -850.652
Clustered standard errors displayed in parentheses, where each cluster is a firm. * significant at 10%; ** significant at 5%; *** significant at 1%
19
9
Table 1.25: Comparing the estimates obtained by different methods
Separate regressions One regression
OLS Maximum OLS Instrumental
likelihood variables
Seller specific investment 0.689 1.808 0.651 1.963
Buyer specific investment 0.275 -2.318 0.247 -2.022
200
Appendix to Chapter 3
Appendix E: Further details on the variables
In what follows, we provide details on the variables used in this dissertation
chapter, including the actual survey questions posed to our respondents.
Asset specificity
One question addressed to the sales manager was:
To the best of your knowledge, does the customer have to undertake a
substantial amount of special investment to be able to use the product
when it is supplied by you rather than an alternative supplier?
1. No
2. Yes, a small amount of special investment
3. Yes, a significant amount of special investment
4. Yes, a very large amount of special investment
The procurement manager was addressed a very similar question to the one
above. The information extracted by these questions is used to construct the dummy
variable Buyer specific investment ? it equals one if any amount of special investment
was needed, zero otherwise.
Two questions addressed to the sales manager were:
Were any elements of this product custom-made for the specific needs of
201
this customer? YES/NO (If ?NO? then skip the next question)
At what cost could the product be modified to sell to other enterprises if
your enterprise had surplus amounts?
1. Virtually no cost
2. Small cost
3. Moderate cost
4. High cost
5. Prohibitive cost (could not be sold to other enterprises)
Using this pair of questions, along with a very similar pair addressed to the
procurement manager, Seller specific investment dummy variable is constructed ? it
equals one if a small, moderate, high or prohibitive cost would be incurred to modify
the product, and zero if no elements were custom-made, or virtually no costs would
be incurred to modify the customized product.
Finally, the Specific investment dummy variable is constructed, and it equals
one if either the buyer or the seller has made a specific investment, zero otherwise.
Market uncertainty
In order to capture truly exogenous events, the general manager of each inter-
viewed firms was asked to assess the importance of uncertainty for his area of business
activity. We have restricted attention to unpredictable changes in weather conditions
and transportation links. The question addressed to company?s general manager was:
How important are unpredictable changes in the area of business activity
in which your enterprise is engaged? Your answer should reflect, as much
202
as possible, the general experience of enterprises in your area of business,
rather than the very specific experience of your enterprise.
Please indicate whether the following sources of unpredictability are:
1. Not important, 2. Not very important, 3. Somewhat important, 4.
Very important
a. Weather-induced variation in demand for the good or service sold by
enterprises like yours. 1 2 3 4
b. Weather-induced variation in supply of goods or services that enter-
prises like yours must buy. 1 2 3 4
c. Problems in transportation links that cause changes in the level of de-
mand for the good or service sold by enterprises like yours. 1 2 3 4
d. Problems in transportation links that cause changes in supply of goods
or services that enterprises like yours must buy. 1 2 3 4
The Exogenous market uncertainty dummy variable equals one if any of these
four sources of uncertainty is considered to be very important, zero otherwise.
Frequency of exchange
The following three questions were addressed to the sales department manager:
Was the specific agreement with this customer part of a longer-term frame-
work agreement in which multiple sales of the product were anticipated?
YES/NO (If ?NO? then skip the next two questions)
How long a time period, in months, did this longer-term framework agree-
ment cover? [number of months]
203
Approximately, how many separate deliveries of your product did this
longer-term framework agreement cover? [number of deliveries]
The procurement department manager was asked very similar questions. The
first step we took was to divide the number of deliveries by the number of months
the agreement covered, after which a dummy variable was constructed, equal to one
if the number thus obtained was greater than or equal to four, zero otherwise ? this
variable was labeled Four or more exchanges per month.
Ease of measurement
A question addressed to the sales department manager was:
For the product in the specific agreement, how easily can the customer
ascertain quality?
1. After a quick inspection
2. After an intensive inspection over a few days
3. Only after extended use by the customer
The procurement manager was asked an almost identical question, in which
only the point of view is changed as it is the respondent who is in fact ascertaining
quality. As mentioned before, the dummy variable Good easy to measure equals one
if the answer to the question above was ?after a quick inspection?, zero otherwise.
Reputation
The sales manager was asked the following two questions:
If your enterprise defaulted in its obligations under this agreement, do you
204
think that your other customers would learn about this? YES/NO
If this customer did not pay its obligations under this agreement, do you
think that other enterprises like yours would learn about this? YES/NO
The procurement manager was asked a pair of similar questions. Based on the
information collected we have constructed two dummy variables: Information dissem-
ination among buyers equals one if other customers would learn if the seller would
default in its obligations under the specific agreement, zero otherwise, and, similarly,
Information dissemination among sellers equals one if other potential sellers would
learn if the current buyer did not pay its obligations under the specific agreement,
zero otherwise.
Transaction complexity
One of the few questions addressed to legal advisers that we use here is:
How often does your enterprise use the following types of documents as the
basis for contracts for the selling of your enterprise?s goods or services?
1. Never, 2. Sometimes, 3. Often, 4. Always
a. Form contract of your enterprise 1 2 3 4
b. Form contract of the customer 1 2 3 4
c. Form contract from legal suppliers, publications, internet 1 2 3 4
d. Contract specifically designed for transaction 1 2 3 4
e. Invoice used as a contract 1 2 3 4
Dummy Customized contracts used frequently equals one if contracts specifically
designed for the particular transactions are used often or always, zero otherwise.
205
Attitude toward risk
In a distinct section focusing on respondents? opinions and attitudes, the in-
terviewed managers were asked the following question (from which we only present
the relevant part, resembling a question from National Opinion Research Center?s
General Social Survey):
We would like to obtain your views on certain issues. Below are pairs of
contrasting statements. Please convey your views by choosing a point on
a scale from 0 to 10. A ?0? means you completely agree with the statement
on the left and a ?10? means you completely agree with the statement on
the right.
Statement Your mark (0 to 10) Statement
Most people cannot be trusted Most people can be trusted
Dummy variable Distrustful sales/procurement manager equals one if respon-
dents tend to agree more with the view that most people cannot be trusted (by
choosing scores lower than or equal to four), zero otherwise.
Age of the firms
The general managers of each company were asked about the founding date
of their enterprises and dummy variable One of the firms founded before 1990 was
constructed. One thing to note is that, since information comes from the general
manager, it is not transaction-specific in the sense that we do not have similar in-
formation from the partner in the specific agreement. We possess two hundred and
fifty-four responses, from companies why act as buyers in some agreements and as
206
sellers in others, therefore the slightly different interpretation, also apparent from
variable?s name.
Organization of legal matters
Each of our respondents was presented a set of legal questions ? five commercial
law medium difficulty questions (chosen from actual multiple choice or true/false
questions asked at law-school examinations).
Given that the questions addressed to the sales managers are different from those
addressed to the procurement managers, and considering their length, we choose not
to present them here. Instead, we display below the introduction to this section,
which tried to assure responsiveness and informativeness by stressing that we are not
testing respondent?s knowledge but rather legal system?s ability to make pertinent
information available to its users, and by instructing survey operators to make sure
respondents provide their best guesses (without looking up answers). The exact
introduction to this section and instructions for the survey operator were:
In this part of the survey, we ask you some questions about laws and or-
dinances that are intended to facilitate transactions. Please do not look
up the formal laws and ordinances in answering these questions. We are
interested in learning whether certain aspects of these laws and ordinances
have become part of the business culture - whether they are known by pro-
fessionals without using reference works. This is not a test of you; rather,
it is a test of the ability of the legal system to make the law known to you.
Please provide your best guess of the correct answer.
207
Interviewer: It is absolutely crucial that the respondent answers the ques-
tions without looking up source materials. Convince the respondent that
we are interested in his or her personal opinion and that the answers are
anonymous. If that does not work, then ask the respondent to skip this
section if he or she absolutely refuses to answer it spontaneously.
Dummy variable Respondent has some legal knowledge is constructed to equal
one if the respondent has correctly answered at least two out of the five questions,
zero otherwise. In addition, the head of the legal department is asked:
Which of the following statements would best describe the location of your
office?
1. Within 20 meters of General Manager?s office
2. In the same building as General Manager?s office but not close to it
3. In the same location as General Manager?s office but in a different
building
4. At a different location than the General Manager
Since some companies did not have a legal department established, and instead
rely on services provided by an outside legal consultant, the interviewers were told
to refer to that person?s office when working on site for the company. Legal adviser?s
office close to manager?s is a dummy variable which equals one if the two offices are
located within 20 meters from each other, zero otherwise.
208
BIBLIOGRAPHY
Acemoglu, Daron, Philippe Aghion, Rachel Griffith and Fabrizio Zilibotti [2005] ?Ver-
tical Integration and Technology: Theory and Evidence?, working paper available
at http://econ-www.mit.edu/faculty/download pdf.php?id=1041
Achen, Christopher H. [1986] The Statistical Analysis of Quasi-Experiments, Univer-
sity of California Press.
Adler, Teryy R., Robert F. Scherer, Sidney L. Barton and Ralph Katerberg [1998]
?An Empirical Test of Transaction Cost Theory: Validating Contract Typology?,
Journal of Applied Management Studies, 7(2), 185?200.
Allen, Douglas W. [2000] ?Transaction Costs?, in Encyclopedia of Law and Eco-
nomics, Volume 1: The History and Methodology of Law and Economics, Bouckaert
and De Geest (eds.), Edward Elgar Press, 893?926.
Anderson, Erin and David C. Schmittlein [1984] ?Integration of the Sales Force: An
Empirical Examination?, The RAND Journal of Economics, 15(3), 385?395.
Arrow, Kenneth J. [1969] ?The Organization of Economic Activity: Issues Pertinent
to the Choice of Market Versus Nonmarket Allocation?, in The Analysis and Evalu-
ation of Public Expenditure, Vol. 1, U.S. Joint Economic Committee, 91st Congress.
Banerjee, Abhijit V. and Esther Duflo [2000] ?Reputation Effects and the Limits of
Contracting: A Study of the Indian Software Industry?, The Quarterly Journal of
Economics, 115(3), 989?1017.
Benham, Alexandra and Lee Benham [2000] ?Measuring the Costs of Exchange?,
in Institutions, Contracts and Organizations: Perspectives from New Institutional
Economics, M?enard (ed.), Edward Elgar Press, 367?375.
Ben-Ner, Avner and Louis Putterman [2001] ?Trusting and Trustworthiness?, Boston
University Law Review, 81, 523?551.
Blinder, Alan S. [1994] ?On Sticky Prices: Academic Theories Meet the Real World?,
in Monetary Policy, Mankiw (ed.), University of Chicago Press, 117?150.
209
Bohnet, Iris and Richard Zeckhauser [2004] ?Trust, Risk and Betrayal?, Journal of
Economic Behavior & Organization, 55, 467?484.
Cheung, Steven N.S. [1998] ?The Transaction Costs Paradigm: 1998 Presidential
Address Western Economic Association?, Economic Inquiry, 36(4), 514?521.
Chiappori, Pierre A. and Bernand Salani?e [2002] ?Testing Contract Theory: A
Survey of Some Recent Work?, CESifo Working Paper No. 738 available at
http://www.cesifo.de/DocCIDL/738.pdf
Coase, Ronald H. [1992] ?The Institutional Structure of Production?, The American
Economic Review, 82(4), 713?719.
Cook, Karen S. and Robin M. Cooper [2003] ?Experimental Studies of Cooperation,
Trust, and Social Exchange? in Trust and Reciprocity, Ostrom and Walker (eds.),
Russell Sage, 209?244.
Crocker, Keith J. and Scott E. Masten [1991] ?Pretia Ex Machina? Prices and Process
in Long-term Contracts?, The Journal of Law & Economics, 34(1), 69?99.
Crocker, Keith J. and Kenneth J. Reynolds [1993] ?The Efficiency of Incomplete
Contracts: An Empirical Analysis of Air Force Engine Procurement?, The RAND
Journal of Economics, 24(1), 126?146.
Department of Trade and Industry, United Kingdom [2004] ?The Value Added Score-
board 2004: The Top 800 UK & 600 European Companies by Value Added?,
available at http://213.219.8.102/value added/pdfs/company data.pdf
Di Tella, Rafael and Robert MacCulloch [2006] ?Some Uses of Happiness Data in
Economics?, The Journal of Economic Perspectives, 20(1), 25?46.
Dietrich, Michael [2003] ?The Importance of Management and Transaction Costs for
Large UK Firms?, Applied Economics, 35(11), 1317?1329.
Eckel, Catherine C. and Rick K. Wilson [2004] ?Is Trust a Risky Decision??, Journal
of Economic Behavior & Organization, 55, 447?465.
Ekl?of, Jan A. and Sune Karlsson [1999] ?Testing and Correcting for Sample
Selection Bias in Discrete Choice Contingent Valuation Studies?, SSE/EFI
210
Working Paper Series in Economics and Finance, No. 171, available at
http://swopec.hhs.se/hastef/papers/hastef0171.pdf
Furubotn, Eirik G. and Rudolf Richter [1997] Institutions and Economic Theory: The
Contribution of the New Institutional Economics, Ann Arbor: The University of
Michigan Press.
Garrouste, Pierre and St?ephane Saussier [2005] ?Looking for a Theory of the Firm:
Future Challenges?, Journal of Economic Behavior & Organization, 58(2), 178?199.
Ghoshal, Sumantra and Peter Moran [1996] ?Bad for Practice: A Critique of the
Transaction Cost Theory?, The Academy of Management Review, 21(1), 13?47.
Gibbons, Robert [2005] ?Four Formal(izable) Theories of the Firm??, Journal of
Economic Behavior & Organization, 58(2), 200?245.
Glaeser, Edward L., David I. Laibson, Jose A. Scheinkman, and Christine L. Soutter
[2000] ?Measuring Trust?, The Quarterly Journal of Economics, 115(3), 811?846.
Gompers, Paul and Josh Lerner [1996] ?The Use of Covenants: An Empirical Analysis
of Venture Partnership Agreements?, The Journal of Law & Economics, 39(2), 463?
498.
Gow, Hamish R., Deborah H. Streeter and Johan F.M. Swinnen [2000] ?How Private
Contract Enforcement Mechanisms can Succeed where Private Institutions Fails:
The Case of Juhocukor a.s.?, Agricultural Economics, 23(3), 253?265.
Grossman, Sanford J. and Oliver D. Hart [1986] ?The Costs and Benefits of Own-
ership: A Theory of Vertical and Lateral Integration?, The Journal of Political
Economy, 94(4), 691?719.
Hausman, Jerry A. [1978] ?Specification Tests in Econometrics?, Econometrica, 46
(6), 1251?1271.
Hausman, Jerry A. [1983] ?Specification and Estimation of Simultaneous Equations
Models?, in Handbook of Econometrics, Griliches and Intriligator (eds.), North
Holland, 391?448.
211
Heckman, James [1974] ?Shadow Prices, Market Wages and Labor Supply?, Econo-
metrica, 42(4), 679?694.
Heckman, James [1976] ?The Common Structure of Statistical Models of Truncation,
Sample Selection and Limited Dependent Variables and a Simple Estimator for
Such Models?, Annals of Economic and Social Measurement, 5, 475?492.
Heckman, James [1979] ?Sample Selection Bias as a Specification Error?, Economet-
rica, 47(1), 53?161.
Heide, Jan B. and George John [1990] ?Alliances in Industrial Purchasing: The De-
terminants of Joint Action in Buyer-Supplier Relationships?, Journal of Marketing
Research, 27(1), 24?36.
Hendley, Kathryn and Peter Murrell [2003] ?Which Mechanisms Support the Fulfill-
ment of Sales Agreements? Asking Decision-Makers in Firms?, Economics Letters,
78(1), 49?54.
Holmstr?om, Bengt R. [1999] ?The Firm as a Subeconomy?, The Journal of Law,
Economics, & Organization, 15(1), 74?102.
Hubbard, Thomas N. [2001] ?Contractual Form and Market Thickness in Trucking?,
The RAND Journal of Economics, 32(2), 369?386.
John, George and Barton A. Weitz [1988] ?Forward Integration into Distribution: An
Empirical Test of Transaction Cost Analysis?, The Journal of Law, Economics, &
Organization, 4(2), 337?355.
Johnson, Simon, John McMillan and Christopher Woodruff [2000] ?Entrepreneurs
and the Ordering of Institutional Reform. Poland, Slovakia, Romania, Russian and
Ukraine Compared?, The Economics of Transition, 8(1), 1?36.
Johnson, Simon, John McMillan and Christopher Woodruff [2002] ?Courts and Rela-
tional Contracts?, The Journal of Law, Economics, & Organization, 18(1), 221?277.
Joshi, Ashwin W. and Rodney L. Stump [1999] ?The Contingent Effect of Specific
Asset Investments on Joint Action in Manufacturer-Supplier Relationships: An
Empirical Test of the Moderating Role of Reciprocal Asset Investments, Uncer-
tainty, and Trust?, Journal of the Academy of Marketing Science, 27(3), 291?305.
212
Joskow, Paul L. [1985] ?Vertical Integration and Long-term Contracts: The Case
of Coal-burning Electric Generating Plants?, The Journal of Law, Economics, &
Organization, 1(1), 33?80.
Joskow, Paul L. [1988] ?Asset Specificity and the Structure of Vertical Relationships:
Empirical Evidence?, The Journal of Law, Economics, & Organization, 4(1), 95?
117.
Kahneman, Daniel and Alan B. Krueger [2006] ?Developments in the Measurement
of Well-Being?, The Journal of Economic Perspectives, 20(1), 3?24.
Karlan, Dean S. [2005] ?Using Experimental Economics to Measure Social Capital
and Predict Financial Decisions?, The American Economic Review, 95(5), 1688?
1699.
Kaufmann, Daniel, Aart Kraay, and Pablo Zoido-Lobaton [2002] ?Governance Mat-
ters II: Updated Indicators for 2000-01?, Policy Research Working Paper no. 2772,
World Bank.
Kennedy, Peter [2003] A Guide to Econometrics, fifth edition, The MIT Press.
Kessler, Friedrich and Richard H. Stern [1959] ?Competition, Contract and Vertical
Integration?, The Yale Law Journal, 69(1), 1?129.
Klaes, Matthias [2000] ?The History of the Concept of Transaction Costs: Neglected
Aspects?, Journal of the History of Economic Thought, 22(2), 191?216.
Klein, Benjamin, Robert G. Crawford and Armen A. Alchian [1978] ?Vertical Integra-
tion, Appropriable Rents and the Competitive Contracting Process?, The Journal
of Law & Economics, 21(2), 297?326.
Koss, Patricia A. and Curtis B. Eaton [1997] ?Co-specific Investments, Hold-up and
Self-enforcing Contracts?, Journal of Economic Behavior & Organization, 32(3),
457?470.
Kuperan, K., Nik Mustapha, Raja Abdulah, Robert S. Pomeroy, E. Ge-
nio, and A. Salamanca [1998] ?Measuring Transaction Costs of Fish-
eries Co-Management?, paper presented at Crossing Boundaries, the Sev-
enth Annual Conference of the International Association for the Study
213
of Common Property, Vancouver, British Columbia, Canada, available at
http://dlc.dlib.indiana.edu/archive/00000075/00/kuperan.pdf
Lesmond, David A., Joseph P. Ogden, and Charles A. Trzcinka [1999] ?A New Esti-
mate of Transaction Costs?, The Review of Financial Studies, 12(5), 1113?1141.
Little, Roderick J. A. and Donald B. Rubin [1987] Statistical Analysis with Missing
Data, John Wiley & Sons.
Luhmann, Niklas [1979] ?Trust: A Mechanism for the Reduction of Social Complex-
ity? in Trust and Power, Luhmann (ed.), John Wiley & Sons, 4?103.
Lyons, Bruce R. [1994] ?Contracts and Specific Investment: An Empirical Test of
Transaction Cost Theory?, Journal of Economics & Management Strategy, 3(2),
257?278.
Macher, Jeffrey T. and Barak D. Richman [2006] ?Transaction Cost Economics: An
Assessment of Empirical Research in the Social Sciences?, Duke Law School Legal
Studies Paper No. 115 available at http://ssrn.com/abstract=924192
Maskin, Eric and Jean Tirole [1999] ?Two Remarks on the Property-Rights Litera-
ture?, The Review of Economic Studies, 66(1), 139?149.
Masten, Scott E. [1986] ?Institutional Choice and the Organization of Production:
the Make-or-Buy Decision?, Journal of Institutional and Theoretical Economics,
142, 493?509.
Masten, Scott E. [1996] ?Introduction? in Case Studies in Contracting and Organi-
zation, Masten (ed.), Oxford University Press, 3?25.
Masten, Scott E. and St?ephane Saussier [2002] ?Econometrics of Contracts: An As-
sessment of Developments in the Empirical Literature on Contracting?, in The Eco-
nomics of Contracts: Theories and Applications, Brousseau and Glachant (eds.),
Cambridge University Press, 273?292.
McMillan, John and Christopher Woodruff [1999] ?Dispute Prevention Without
Courts in Vietnam?, The Journal of Law, Economics, & Organization, 15(3), 637?
658.
214
McMillan, John and Christopher Woodruff [2002] ?The Central Role of Entrepreneurs
in Transition Economies?, The Journal of Economic Perspectives, 16(3), 153?170.
Melenberg, Bertrand and Arthur Van Soest [1996] ?Parametric and Semi-parametric
Modelling of Vacation Expenditures?, Journal of Applied Econometrics, 11, 59?76.
Monteverde, Kirk and David J. Teece [1982] ?Appropriable Rents and Quasi-Vertical
Integration?, Journal of Law & Economics, 25, 321?328.
Murrell, Peter [2001] Assessing the Value of Law in Transition Economies, Murrell
(ed.), University of Michigan Press.
Murrell, Peter [2003] ?Firms Facing New Institutions: Transactional Governance in
Romania?, Journal of Comparative Economics, 31(4), 695?714.
National Agency for Small and Medium Sized Enterprises and Cooperatives, Govern-
ment of Romania [2004] ?Situat?ie ?si Necesit?at?i: Opinii ale ?Intreprinderilor Mici ?si
Mij-locii? (Status and Necessities: Opinions of SMEs), October 2004 study avail-
able at http://www.animmc.ro/files/comunicate/Brosura 1 11oct2004 .pdf
Nawata, Kazumitsu and Michael McAleer [2001] ?Size Characteristics of Tests for
Sample Selection Bias: A Monte Carlo Comparison and Empirical Example?,
Econometric Reviews, 20(1), 105?112.
Nelson, Richard R. and Sidney G. Winter [1982] An Evolutionary Theory of Economic
Change, Cambridge: Harvard University Press.
Noldeke, Georg and Klaus M. Schmidt [1995] ?Option Contracts and Renegotiation:
A Solution to the Hold-up Problem?, The RAND Journal of Economics, 26(2),
163?179.
North, Douglass C and Robert P. Thomas [1973] The Rise of the Western World: A
New Economic History, Cambridge University Press.
Puhani, Patrick A. [2000] ?The Heckman Correction for Sample Selection and its
Critique?, Journal of Economic Surveys, 14(1), 53?68.
215
Rivers, Douglas and Quan H. Vuong [1988] ?Limited Information Estimators and
Exogeneity Tests for Simultaneous Probit Models?, Journal of Econometrics, 39(3),
347?366.
Rizov, Marian [2004] ?Firm Investment in Transition: Evidence from Romanian Man-
ufacturing?, The Economics of Transition, 12(4), 721?746.
Rogerson, William P. [1992] ?Contractual Solutions to the Hold-up Problem?, The
Review of Economic Studies, 59(201), 777?794.
Salani?e, Bernand [1997] The Economics of Contracts: A Primer, The MIT Press,
Cambridge.
Sartori, Anne E. [2003] ?An Estimator for Some Binary-Outcome Selection Models
Without Exclusion Restrictions?, Political Analysis, 11, 111?138.
Saussier, St?ephane [2000] ?Transaction Costs and Contractual Incompleteness: the
Case of ?Electricit?e de France?, Journal of Economic Behavior & Organization,
42(2), 189?206.
Shelanski, Howard A. and Peter G. Klein [1995] ?Empirical Research in Transaction
Cost Economics: A Review and Assessment?, The Journal of Law, Economics, &
Organization, 11(2), 335?361.
Sykuta, Michael E. [2005] ?New Institutional Economics: The Case of Contract-
ing and Organizations Research?, in New Institutional Economics: A Textbook,
Brusseau and Glachant (eds.), Cambridge University Press.
Vella, Francis [1998] ?Estimating Models with Sample Selection Bias: A Survey?,
Journal of Human Resources, 33(1), 127?169.
Wallis, John J. and Douglass C. North [1986] ?Measuring the Transactions Sector
in the American Economy: 1870 to 1970?, in Long Term Factors in American
Economic Growth, Studies in Income and Wealth, Engerman and Gallman (eds.),
University of Chicago Press, Chicago, 95?161.
Wang, Ning [2003] ?Measuring Transaction Costs: An Incomplete Survey?,
Ronald Coase Institute Working Paper Series, working paper available at
http://coase.org/workingpapers/wp-2.pdf
216
Whinston, Michael D. [2003] ?On the Transaction Costs Determinants of Vertical
Integration?, The Journal of Law, Economics, & Organization, 19(1), 1?23.
Williamson, Oliver E. [1979] ?Transaction-Cost Economics: The Governance of Con-
tractual Relations?, The Journal of Law & Economics, 22(2), 233?261.
Williamson, Oliver E. [1983] ?Credible Commitments: Using Hostages to Support
Exchange?, The American Economic Review, 73(4), 519?540.
Williamson, Oliver E. [1985] The Economic Institutions of Capitalism, New York:
Free Press.
Williamson, Oliver E. [1991a] ?Comparative Economic Organization: The Analysis
of Discreet Structural Alternatives?, Administrative Science Quarterly, 36(2), 269?
296.
Williamson, Oliver E. [1991b] ?Strategizing, Economizing, and Economic Organiza-
tion?, Strategic Management Journal, 12, Special Issue: Fundamental Research
Issues in Strategy and Economics, 75?94.
Williamson, Oliver E. [2000] ?The New Institutional Economics: Taking Stock, Look-
ing Ahead?, Journal of Economic Literature, 38(3), 595?613.
Wooldridge, Jeffrey M. [2002] Econometric Analysis of Cross Section and Panel Data,
The MIT Press.
217