ABSTRACT Title of Dissertation: SUPPLY CHAIN RISKS, RESILIENCE AND FIRM PERFORMANCE: AN EMPIRICAL STUDY Camil Martinez, Doctor of Philosophy in Business and Administration, 2018 Dissertation Directed By: Dr. Martin Dresner, Logistics, Business and Public Policy Department This dissertation’s main focus is the study of supply chain resilience. The two studies investigate the impact of supply chain geographical locations risks and supply chain resilience on performance and of supply chain risks and disruptive events in resilience strategies. Essay 1 seeks to understand the impact of supply chain resilience strategies on firm’s performance. We utilize a cross sectional data sample from 2014 containing detailed manufacturing location risk data and resilience planning at the location level for 313 publicly traded firms. We look at three supply chain resilience cultural traits, business continuity planning, inventory and financial stability. We find that resilience has a positive effect on firm performance. Essay 2 looks at the impact of two types of supply chain risks (internal and external) and two types of disruptive events (internal and external) in the development of supply chain resilience strategies. We find that external disruptive events have a positive impact on supply chain resilience but internal disruptive events have a negative impact in the development of resilience. However, once a business continuity plan is in place, previous internal disruptive events are associated with more agility. My findings for both essays contribute to the supply chain resilience literature by empirically testing the impact of resilience on performance and the impact of disruptive events on resilience strategies. SUPPLY CHAIN RISKS, RESILIENCE AND FIRM PERFORMANCE: AN EMPIRICAL STUDY By Camil Martínez 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 Doctorate of Philosophy 2018 Advisory Committee: Professor Martin Dresner [Chair] Professor Curtis Grimm Professor Thomas Corsi Professor Adams Steven Professor Gang-Len Chang [Dean’s Representative] © Copyright by [Camil Martinez] [2018] Dedication This dissertation is dedicated to my father Freddie Martínez, my mother Tomasita Arroyo and my siblings Freddie, Ferdin, Karen and Karol. The love, the education and support that I have received from my family have made me the person that I am. My parents built a home where curiosity and education could be freely pursued and my siblings made the pursuit of those an adventure. ii Acknowledgements Special thanks to Dr. Martin Dresner for being my adviser, the chair of the Dissertation committee and a great example to follow professionally and personally. Special thanks also to Dr. Curtis Grimm, Dr. Thomas Corsi, Dr. Sandor Boyson, Dr. Adams Steven and Dr. Gang-Len Chang for serving on my committee. Thank you for the significant time and effort you have each given helping me improve and refine my dissertation. Thanks also to my dear friend Dr. Daniel Aguilar, who has always been supportive of my work and whose curiosity about what I do has helped me learn how to communicate supply chain concepts to people outside the field. I have enjoyed all the questions and the deep thinking help within our friendship. I would not have been able to complete this program without your help. Thanks to the PhD students in the supply chain management program who helped me think throughout the course of the program, helped me improve as a researcher and through suggestions, comments and questions have improved the outcome of this dissertation. More than that, I am grateful for the friendship you have offered me. You have made the PhD program a great and beautiful experience. Heidi Celebi, John Patrick Paraskevas, Adams Steven, Isaac Elking, Anupam Kumar, Kevin Sweeney, Heidi Celebi, Alan Pritchard, Kate Ren, Xiaodan Pan, Rohan D`Lima; you have all been extremely helpful, and I appreciate your support. iii I would also like to thank Dr. Bob Windle, Dr. Phil Evers and Dr. Stephanie Eckerd for their coaching through the entire PhD program. I would not have been able to complete this program without each of your support. iv Table of Contents Dedication ..................................................................................................................... ii Acknowledgements ...................................................................................................... iii Table of Contents .......................................................................................................... v Chapter 1: Overview ..................................................................................................... 1 Chapter 2: The Impact of Supply Chain Geographical Location Risks and Resilience on Firm Performance .................................................................................................... 2 ABSTRACT .............................................................................................................. 2 INTRODUCTION .................................................................................................... 3 THEORY AND HYPOTHESIS DEVELOPMENT ................................................. 5 METHODOLOGY ................................................................................................. 25 RESULTS ............................................................................................................... 38 DISCUSSION AND CONCLUSIONS .................................................................. 44 APPENDIX A: ROBUSTNESS CHECKS ............................................................ 48 Chapter 3: The impact of Internal and External Disruptive Events on Supply Chain Resilience Strategies ................................................................................................... 53 ABSTRACT ............................................................................................................ 53 INTRODUCTION .................................................................................................. 54 LITERATURE AND HYPOTHESIS DEVELOPMENT ...................................... 56 METHODOLOGY ................................................................................................. 73 RESULTS ............................................................................................................... 89 DISCUSSION ......................................................................................................... 95 LIMITATIONS AND FUTURE RESEARCH ....................................................... 99 CONCLUSIONS AND MANAGERIAL IMPLICATIONS ................................ 100 Chapter 4: Future Extensions .................................................................................... 103 References ................................................................................................................. 104 v Chapter 1: Overview The two essays of this dissertation will focus on supply chain risks and resilience. Supply chain resilience has become an important topic for both researchers and practitioners due to the turbulent environment in which multinational firms have to operate in. Natural disasters, political instability of different countries and economical disasters can impact a firm’s capacity to move its products to consumers. Supply chain resilience presents an opportunity to protect the supply chain against disruptions and maintain or even improve firm performance. Essay 1 of my dissertation analyzes a cross sectional sample of 313 firms across 40 3-digit NAICS code industries and how supply chain geographical risks and resilience impact performance. We framed our hypotheses using Structure-Conduct- Performance paradigm to explain the relationships between our variables and establish our hypotheses. We expect, based on the literature, that resilience will pose a competitive advantage for firms that operate in risky environments, effectively mitigating the impact of risk on performance. Essay 2 analyzes a cross sectional sample of 159 firms across 3 industries. Essay 2 focuses on the impact of two types of risks and disruptive events, external and internal, on the resilience strategies put in place by the firm. We look at two strategies: business continuity planning and average projected recovery time after a disruption. We pose that the higher the exposure to risks and disruptive events a firm faces, the more likely it will be to have resilience strategies in place in order to cope with the risky environment. 1 Chapter 2: The Impact of Supply Chain Geographical Location Risks and Resilience on Firm Performance ABSTRACT The location of a manufacturing facility can present various risks to the firm’s supply chain. The impact of these risks is an understudied area that we address in this essay. Our study uses Resilinc's supplier database to analyze 3,262 manufacturing locations for 313 publicly traded firms across 40 industries. This study examines the impact on firm performance (measured as gross margin) of supply chain geographical location environmental risks (measured as natural disaster risk and geopolitical risk) and firm resilience (measured as financial solidity, recovery time and inventory). Using a linear regression model adapted for industry clusters, we find that natural disaster risk and resilience have a positive impact on performance. A surprising finding shows that frequency of potential disruptions also has a positive impact on firm performance. 2 INTRODUCTION On September 20, 2017, hurricane María struck the island of Puerto Rico with its 155 miles per hour maximum winds and lots of water. The Puerto Rico government is still trying to estimate the total impact on the island’s economy almost four months after the storm. The current estimate of monetary is impact is approximately $100 billion. The hurricane not only affected the Puerto Rican economy by destroying houses, businesses, the power system in the entire island and structures like roads, bridges and government buildings. Global supply chains have also taken their toll from this big disruption. Examples are the pharmaceutical and medical devices industries. The medical devices industry has an important amount of production in Puerto Rico and about 8% of the medicines consumed in the US are manufactured in Puerto Rico (Aton, 2017). Due to the interruption on production that the factories suffered in the aftermath of the hurricane the Food and Drug Administration is currently tracking 30 critical pharmaceuticals and 50 critical medical devices manufactured solely in Puerto Rico or majorly in Puerto Rico, according to an article published by Scientific American on October 25, 2017 by Aton on EE News. The current situation in Puerto Rico is only the latest example of what supply chain disruptions can do to firms, industries and countries. Hendricks and Singhal (2003) lead the research on the impact of disruptions to firm performance. They studied public announcements of disruptions to production and deliveries and the impact that these events had on the stock value. They found significant negative impacts that can last up to two years depending on the magnitude of the disruption. 3 Some studies of supply chain risk focused on assessing the risks with a focus on risk identification and avoidance (Manuj and Mentzer, 2008). More recent studies focus on developing ways to mitigate the risks, accepting that disruptions are part of the supply chain global environment. Supply chain resilience has been proposed as an effective way to cope with the unavoidably risky environment in which firms have to operate (Sheffi, 2005). Is there something that firms operating in Puerto Rico could have done to avoid the impact from hurricane María? Hurricanes are not a new experience to the island, it is situated right in the middle of the hurricane zone. Hurricane María had a historical impact, word of mouth in Puerto Rico says that it has been the worst storm in a hundred years. Can the pharmaceutical and medical devices industries recover fast enough in order to avoid a crisis? Sheffi (2005) suggests that firms with an organizational culture of resilience would be able to find a solution and recover faster than firms with no culture of resilience, but not only during a disruption, resilient firms should be better performing firms due to the benefits that developing resilient processes bring to the firm. Our study aims to look at the impact of supply chain resilience on firm performance. Using a cross sectional data sample from 2014, we test the impact of three supply chain resilience cultural traits. We use the Structure-Conduct- Performance paradigm to theoretically frame our study and explain our variables and hypotheses. 4 THEORY AND HYPOTHESIS DEVELOPMENT The Structure-Conduct-Performance (SCP) paradigm developed by Mason (1939) and Bain (1951), establishes that there is a direct relationship between market structure, firm conduct and performance. Firms respond to the market environment in order to compete and obtain advantage from other firms. The SCP framework has been widely used in the strategy field, as illustrated by Williams and Smart (1993) literature review, to explain the reasons why firms develop certain strategies. In the current market of global competition, some authors have claimed that competition is no longer firm to firm but supply chain to supply chain, (Trkman et al, 2007; Li et al, 2005). Not surprisingly, the supply chain literature has been intertwined with the strategy literature, (Glassman and Honeycutt, 2002; Hult, Ketchen and Arrfelt, 2007). Blackhurst et al. (2015) utilized the SCP framework to explain the positive impact that supply chain integration has on firm performance. Another theory that has been used to explain the dynamics of supply chains is Barney’s Resource Based View (RBV), (Barney, 1986). Squire, Autry and Petersen (2014) and Blackhurst et al. (2011) applied RBV to explain why firms develop resilience in their supply chains to cope with the risks of disruption that the business environment brings. Similarly, we propose that resiliency is one of the ways that firms respond to a turbulent market environment in order to enhance performance. We combine SCP and RBV with supply chain risk management and resilience literature to frame our hypotheses. We pose that in the modern global environment, there are characteristics to the market beyond market power and entry barriers that affect how a firm can compete in order to obtain higher performance. The risks of 5 operating in a particular geographical location are an example of these environmental factors. The interpretation of “structure” that we are using, takes into consideration the implications that operating in a global market brings. Structure In the SCP framework, “structure” refers to market structure. Traditionally the market environment is measured in terms of market concentration and barriers to entry. Bain defined entry barriers as market conditions that allow incumbent firms to raise prices above the competitive level without attracting entry (Waldman and Jensen, 2007). Making decisions that reduce costs in order to obtain economies of scale is one strategic way to set prices and increase profits. Holweg, Reichart and Hong (2011) state that cost assessment for global sourcing is highly dependent on cost of production and transportation, but that this assessment is not complete. They posit that supply chain risks are often not included in the cost analysis, leading firms to make decisions to operate in high risk environments. Managing these environments creates additional costs. Christopher and Holweg (2011) studied how the global economy has changed over 40 years from 1970 to 2010. Using a volatility index, they show that the global economy has been more volatile since the economic crisis of 2007 than it had been historically, indicating that it is less predictable to conduct global business now than it was before the crisis. Since supply chains are the engines that address the movement of goods around the world preserving connectivity between nodes and functions, they propose that the design of modern supply chains should have embedded into it structural flexibility to deal with the economic turbulence. 6 Knemeyer et al (2009) established that supply chain vulnerability has increased due to globalization and the availability of less slack. By operating in high risk countries in order to seek lower operational costs, the environment in which firms operate has become riskier for many supply chains. In a similar manner, many firms have chosen global networks that operate across different countries and move goods around the world for both production processes and customer deliveries. As global supply chains grow, the probability of facing disruptions also grows, forcing firms to operate in a potentially more disruptive environment. These trends are common across industries, even though the impact of globalization will vary by industry. i. Supply Chain Risk The British Dictionary defines risk as the possibility of incurring misfortune or loss. Carvalho and Cruz Machado (2007) state that when it comes to supply chain failures: “the sources of disturbance might be infinite, but the number of failures is finite.” They place supply chain/operational risks as resulting into five possible failures: 1) raw materials shortage, 2) labor shortage, 3) machine capacity shortage, 4) scrap/rework, 5) finished goods not delivered. In our study we consider supply chain risk to be the possibility of a firm incurring loss due to any of the five failures listed above. Rao and Goldsby (2009) developed a typology of the risks faced by a firm. These include both internal and external risks. One of the risk categories they propose encompasses external factors or “framework factors”, as defined by Ritchie and Marshall (1993). These factors describe the market environment in which the firm 7 operates. The factors include environmental risks, industry risks and organizational risks, (Rao and Goldsby, 2009). Environmental risks include natural disaster risks, political risks, social risks and macroeconomic risks, (Rao and Goldsby, 2009). We understand that environmental risks can set up the structure of the market per SCP. This study focuses on geopolitical risk and natural disaster risk, and the impact that these risks have on firm performance. The first link we look at is a direct link between structure and performance. Holweg et al. (2011) state that political risk is a hidden cost for firms, since they make the decision to go into higher risk countries based on predicted operating costs, without considering political risk. They pose that the additional hidden costs may exceed the operating cost savings from operating in high risk countries, but in their case study there is no irrefutable statistical evidence to support this argument. The tradeoff at hand is between the advantages that operating in a low cost country can bring and the hidden costs that political issues in that country can cause. Operational costs such as low labor costs, low inventory handling costs, low infrastructure costs and low cost of transportation are the main reasons why firms decide to invest in having a manufacturing presence at countries that are less politically stable, (Holweg et al., 2011). Ports and airports are critical points for any supply chain. If the country in which a firm is operating is impacted by a political crisis (a coup, an invasion, a dramatic change of government structure, etc) government owned agencies and processes will be affected as well along with the availability of employees to be to go to work, causing a disruption in the capacity to 8 produce and move goods through the supply chain. The latter are the risks that Holweg et al. (2011) argue that are not being considered and therefore become hidden costs. Although the literature (Christopher and Holweg, 2011; Holweg et al., 2011) portrays a negative impact from operating in risky places, suggesting that these risks should be avoided, we believe that the cost benefits are still higher than the impact of the hidden costs. Holweg et al. (2011) utilized three case studies to test the impact of risk on performance. We expect that looking at a greater sample, results will vary depending on the firm and the places that the firm is operating in. We think that the global environment of business and the turbulence that comes with it have become the normal way of doing business and to deliver a low cost product has become an entry barrier. In order to penetrate a market, it is necessary to have a product with a competitive price. To compete in a global market, economies of scale are necessary in order to deliver an attractive price. High risk countries can bring this benefit. We expect higher geopolitical risk to be associated with higher performance. Natural disaster risk is slightly different from geopolitical risk. When it comes to hurricanes, for example, there is a hurricane season every year that goes roughly from June to November. Firms operating in places located in the hurricane zone, can expect to be impacted during that time of the year more than at any other time of the year. Earthquakes are completely different and much less predictable. However, natural disasters are disasters that do follow patterns. Even though it would be almost impossible to predict one precisely, it can be estimated statiscally that certain places will be impacted in a certain time range. In “the next five or ten years”, depending on the historical statistical data. Insurance companies use this type of data to estimate the 9 probabilities of being affected and calculate insurance premiums (Sheffi 2005). Countries exposed to natural disasters are similar to countries with geopolitical risk in terms of low operational costs. They can provide the same advantages of low cost labor and low cost operations. They present an important difference from geopolitically unstable countries that it would be easier for firms to defend from the environment because statistical data is more reliable and the probability of an event does not depend on human action. These countries are also attractive for manufacturing due to the cost advantage they represent. For this reason we expect natural disaster risk to have a positive impact on performance. Our first hypothesis is divided in two parts to consider both risk types individually because we are interested in understanding if the difference between the two risk types are enough to have a different impact on performance. Hypothesis one is as follows: H1: Higher location environmental risk will be positively associated with firm performance. H1a: Higher geopolitical risk will be positively associated with firm performance. H1b: Higher natural disaster risk will be positively associated with firm performance. ii. Supply Chain Complexity Complexity in the supply chain is not only provided by the global environment, but also by the complexity of the products and the number of nodes in the network. Bode and Wagner (2015) explored three types of complexity in the 10 supply chain: horizontal (# of direct suppliers), vertical (# of tiers) and spatial (# of countries). They found that complexity increases disruption frequency, therefore impacting the performance of the supply chain network. Blackhurst, Dunn and Craighead (2011) studied complexity in terms of the size of the network given by the number of nodes and the connectivity between nodes. They find that complexity reduces resilience, making the firm more vulnerable to its environment. Although none of these two studies refers to product complexity, we understand the more components a product has, the more horizontal complexity the firm will have because it will require more suppliers. A product with a complex design could imply more stages of production which will also impact the number of tiers for its supply chain. In this study we will look at a combination of product complexity and network size complexity. We are not able to define complexity in the same ways Bode and Wagner (2015) and Craighead (2011) did due to data limitations, but product complexity combines both studies as it poses a risk for the supply chain to be bigger and more exposed to be disrupted. Following their findings, we expect product complexity to have a direct negative impact on performance. Complexity makes for a more difficult to manage environment, hence our second hypothesis is: H2: Supply chain product and network complexity will be negatively associated with firm performance iii. Disruptive Events The last environmental factor we consider is the number of disruptive events that a firm experiences. Although the media has contributed immensely to raising awareness as to why it is important to mitigate supply chain risks, the media typically 11 fails to delve into the factors that would be helpful for researchers to better understand the risk management process. In terms of supply chain risks and disruptions, low probability, high impact disruptions tend to get most of the attention, while “everyday risks” that might be low to medium impact events do not get much notice. Hendricks and Singhal (2003) have contributed to the literature by empirically establishing relationships between risks, disruptions and firm performance. Their research allows us to understand why the study of supply chain risk and supply chain risk management is not only interesting but also relevant and important. They have established that supply chain disruptions have a negative effect on stock performance (Hendricks and Singhal, 2003), that disruptions have a long-term effect on stock performance and equity risk (Hedricks and Singhal, 2005a), and that supply chain disruptions negatively impact operating performance, (Hendricks and Singhal, 2005b). They have also studied the impact of mitigation strategies, such as operational slack and diversification, on the disruption events (Hendricks et al., 2009). With their findings, the authors have established the importance of supply chain risk awareness and mitigation, and its main goal to protect the performance of the firm. Sheffi (2005) presents a collection of case studies based on different disruptions that a diverse selection of firms have faced. The aftermath of these disruptions ranges from loss of sales in that year to (at the extreme) bankruptcy. Sheffi finds that there are identifiable warning signs before high impact disruptions 12 happen that a firm may have missed. Suggesting that it is possible to learn from lesser impact events and be better prepared or even avoid a bigger impact event. Other studies have focused on identifying ways to predict the impact of disruptions or to calculate the probability of disruptions happening. Such studies use methods such as field experiments (Hora and Klassen, 2013), experiments (Tazelaar and Snijders, 2013) and simulation (Neiger, Rotaru and Churilov, 2009). Regardless of the methodology, the study findings show that disruptions have a negative impact on firm performance. It is important to make a distinction between disruptions and disruptive events. A disruption is an event that temporally interrupts the normal flow of goods at one or more stages of the supply chain. Hendricks and Singhal (2003) use actual public announcements of business disruptions in their studies. These were identified by production delays or shipping delays. In our study we will look at disruptive events. For example, a flood in Thailand in 2014 closed over a thousand factories. A firm with manufacturing locations in Thailand might or might not be impacted by this event. Even more, a factory in the impacted region, could have been prepared and not have significant damage while another factory that was less lucky or less prepared was impacted more. A factory that did not get flooded, could still be impacted because of the difficulty to move goods that a flood poses. Disruptive events are events that a firm faces but may not actually interrupt the normal flow of goods of the firm’s supply chain. This study looks at events like this one and assumes that if a firm had manufacturing locations in the area impacted by the event, then this firm had to face dealing with this event even if the event did not become a network disruption for 13 the firm directly impacting performance. We expect that the probability of having a disruption that affects performance increases the more disruptive events a firm faces. These events make the supply chain environment more hostile and they posit threats for the supply chain that can add up to a significant impact to performance. We posit that: H3: The higher the frequency of disruptive events a firm has to face the lower the performance of the firm will be. Conduct Traditionally in the SCP framework, conduct refers to the responses that firms have to the competitive environment. In our case, we are looking at cultural traits that would make a firm stronger at reacting to risk, thus granting it competitive advantage. Supply Chain Risk and Supply Chain Risk Management are frequently studied together. The literature in Supply Chain Risk Management (SCRM) has identified four stages of SCRM: 1) identification (Neiger et al., 2009; Trkman and McCormack, 2009), 2) assessment (Ellis et al., 2010; Hendricks and Singhal, 2005), 3) mitigation (Jian et al., 2009; Knemeyer et al., 2009) and 4) responsiveness (Kleindorfer and Saad, 2005). Supply Chain Risk Management is defined by Manuj and Mentzer (2008) as: “The identification and evaluation of risks and consequent losses in the global supply chain and implementation of appropriate strategies through a coordinated approach among supply chain members with the objective of reducing one or more of the following – losses, probability, speed of event, speed of losses, the time for detection of the events, frequency, or exposure – 14 for supply chain outcomes that, in turn, lead to close matching of actual cost savings and profitability with those desired.” It is evident from this definition that SCRM involves anticipating and managing a potential or actual disruption, from before it happens until its effects are over. The SCRM literature has proposed many alternatives in which firms can cope with the risks in which they operate. Some examples of risk mitigation strategies include: supply chain agility (Braunscheidel and Suresh, 2009), flexibility (Seebacker and Winkler, 2013), redundancy (Carvalho et al., 2012; Sheffi, 2005), decision making process development (Manuj and Mentzer, 2008; Speier, Whipple, Closs and Douglass Voss, 2011), and proactive actions to identify and assess risks (Kleindorfer and Saad, 2005; Trkman and McCormack, 2009; Knemeyer et al, 2009). Supply chain resilience encompasses all these aspects. It is often considered to be a response to the risky environment, and can be seen as complementary to supply chain risk management (Mandal, 2012). Sheffi (2005) posits that an effective risk management strategy can cultivate a culture of resilience. A firm that has learned to cope with the challenges of the environment, and has obtained competitive advantage from doing so, will be more capable to cope with new unforeseen challenges, such as unpredictable catastrophic events. Sheffi (2005), in his book entitled, “The Resilient Enterprise: Overcoming Vulnerability for Competitive Advantage”, looks at different types of disruptions to the supply chain, and compares how companies impacted by the same disruption behaved before, during and after the disruption. He studies, for example, how a fire in a Philips plant in New Mexico in 2000 disrupted the supply chains of both Nokia and 15 Ericson. The significant difference in performance for these two firms after the disruption was due mostly to the way they managed the disruption from the time it took both firms to discover the disruption, to how they assessed the impact, managed their suppliers and customers and recovered from the inevitable loss, to the actions that were established afterwards to avoid a similar disruption in the future. All of these factors point to a “way of doing things”, the organizational culture of the firm, that can facilitate the speed and efficient recovery from a disruption. In his conclusion, Sheffi points to an organizational culture of resilience as an asset for competitive advantage. This implies that a firm is able to utilize its processes, ways of communication, knowledge, employees and strategies to figure out how to deal with a situation that was not foreseen and had not happened before. The firm has a way to acquire new collective experience and knowledge and use it to perform better in the future. Sheffi claims that resilience can be an organizational cultural trait that can give firms competitive advantage. Sheffi’s focus on culture is consistent with the Resource Based View (RBV) theory developed by Barney (1986). Barney’s RBV theory claims that the way in which a firm utilizes its resources can grant the firm sustainable competitive advantage. In order for a resource to bring competitive advantage, it has to be valuable, rare, inimitable and non-substitutable (VRIN). Barney (1986) specifically addresses the possibility for organizational culture to be a source of sustainable competitive advantage. He explains how some cultural traits bring economic value to a firm, and that these traits, if also rare and inimitable, can be the source of sustainable competitive advantage. Some examples of cultural traits with economic 16 value presented by Barney are: creativity and innovativeness, employee productivity and value of worth of employees, customer service and satisfaction. Our study aims to understand the link, if any, between supply chain resilience and firm performance. We look at resilience factors that constitute cultural traits and the impact of these factors on firm performance. Organizational culture is a multilevel, complex concept. According to Schein, (1986) there are three levels of cultural phenomena in organizations: 1) behaviors and physical manifestations, 2) values and 3) basic assumptions. The basic assumptions are at the deepest level of the culture, and are the traits that are understood as “correct ways” to cope with the environment. This is the most difficult level to measure because it is the most taken-for-granted behavior of a culture. In a firm with supply chain resilience as part of its basic assumptions, it would be expected for manufacturing locations that face potential disruptions to have a business continuity plan in case a disruptive event happens. These basic assumptions have measurable expressions that we can observe and study. Gordon (1991) explains how the environment in which a firm is operating, specifically the industry, determines the organizational culture that a firm develops to compete in its industry. He states that: “the competitive framework in which a company operates is an important dimension on which core assumptions in the company culture are developed.” In his conceptual model, he establishes that the industry environment (constituted by customer requirements, competitive environment and societal expectations) has an impact in the formation of assumptions and values for the organizational culture of the firm, and that these values are 17 translated into “forms” (constituted by strategies, structures and processes). These forms subsequently have an impact on firm performance. Following this reasoning, we argue that the current global market environment can trigger a culture of resilience that is translated into “forms” that are resilience strategies. For example keeping more inventory than necessary in certain locations so that order fulfillment can be continued when a location is impacted by a disruptive event. We build on Gordon (1991) model, expanding the industry environment to the global market environment. Our assumption is that the organizational cultural forms of resilience can be captured in business practices, and these can be used to assess their impact on firm performance. The assumptions and values of the culture remain unmeasurable, and can be considered as inimitable and rare. Kamalahmadi and Parast (2016), in their literature review of Supply Chain Resilience, defined resiliance as: “The adaptive capability of a supply chain to reduce the probability of facing sudden disturbances, resist the spread of disturbances by maintaining control over structures and functions, and recover and respond by immediate and effective reactive plans to transcend the disturbance and restore the supply chain to a robust state of operations.” Christopher and Peck (2004) identified principles of supply chain resilience. The main four principles are reengineering, collaboration, agility and a culture of SCRM. Combining the contributions of Christopher and Peck (2004) with the definition of Kalahmadi and Parast (2016), we can say that a culture of SCRM can be identified with the anticipation phase of Kahlamadi and Parast (2016) definition, re- 18 engineering is identified with the resistance phase and agility is identified with the recovery phase. In our theoretical frame, the structure of the environment presents risks that in order to develop and maintain competitive advantage, firms develop an organizational culture of resilience that allows them to utilize and reorganize resources to better serve the needs of the firm according to RBV and enhance performance. Consistent with the supply chain resilience literature, RBV and organizational culture theory, we propose our fourth hypothesis as the positive impact of resilience on firm performance. This hypothesis establishes a link between conduct and performance. H4: Supply chain resilience will be positively associated with firm performance Since resilience is an organizational cultural trait, we divide H4 into three parts, for three different indicators of resilience: financial resilience, redundancy in the form of inventory and anticipation and recovery in the form of estimated recovery time. Fiksel et al. (2015) state that financial strength is a resilience factor. A firm must be able to absorb fluctuations in cash flow in order to be resilient to disruptions. Kamalahmadi and Parast (2016) establish that it is necessary to “recover and respond by immediate and effective reactive plans to transcend the disturbance and restore the supply chain to a robust state of operations.” For a firm to be able to accomplish this, it must have the resources available to absorb the immediate cost that going back to a robust state of operations will require when the firm is impacted by a disruptive event. 19 Our assumption is that a firm that is financially solid, would know what kind of financial measure to take in specific circumstances (insurance against certain type of events, inventory investments, multiple production facilities, etc.). Sheffi (2005) gives the example of using insurance when exposed to natural disasters. He says that since there are historical statistical data available, insurance companies are able to develop reliable statistical predictions, and insurance can be one of the measures to build resilience. Such a firm, would know not only how and when to protect from potential disruptions but how to invest money in general to increase performance. The first indicator of resilience that we look at is the financial solidity of the firm that we are calling financial resilience. H4a: Financial resilience (financial solidity) will be positively associated with firm performance. Redundancy and slack have been identified with resilience by multiple studies (Hendricks and Singhal, 2009; Blackhurst et al., 2011; Fiksel et al., 2015; Kamalahmadi and Parast, 2016). Redundancy is one of the key indicators of re- engineering (Kalahmadi and Parast, 2016). It allows a firm to rearrange its resources to ensure a faster recovery from a disruption while order fulfillment is not interrupted. During the 1980’s and 1990’s, keeping a lean inventory strategy was the main trend in many industries. Firms invested in lean manufacturing implementations involving inventory analysis, process revisions and design and employee training to use new analytical tools. Some firms even combined a lean manufacturing culture with a six sigma analytical approach to lean processes in order to improve firm performance. However, enormous global disruptions, such as the Japan tsunami in 20 2011, put the lean culture to the test. The interruptions in business were especially telling for Toyota, in particular, since Toyota is the firm where lean manufacturing originated. Empirical evidence on inventory shows that total inventory did indeed decrease for US manufacturing firms from 1961 to 1994 (Rajagopalan and Malhotra, 2001). When accounting for the trends in the three types of inventory, raw materials, work in progress and finished goods, the study finds that finished goods inventory did not decrease as steadily as the others. The results for finished goods varied by industry and did not significantly change during the period for more than half of the industries in the study. These findings are not completely supportive of the lean inventory push. Chen et al (2005) examined inventory trends from 1981 to 2000. They found that inventory decreased at a rate of 2% per year; but again, finished goods inventory did not change. They found that firms with abnormally high inventory perform poorly, while firms with slightly lower than average inventory had good returns. They found no evidence for extremely lean inventory providing the best performance. Moreover, the lowest level inventory firms performed at about average levels. Therefore, empirical research does not provide strong evidence in favor of lean inventory policies contributing to better performance. It does seem to provide evidence for a “reasonable leanness”; that is, keeping inventory close to the industry mean seems to pay off. The supply chain risk management and resilience literature suggests that inventory levels should be kept at some level above the lowest necessary level in 21 order to reduce the potential impact of unforeseen events. Empirical findings show that inventory can effectively mitigate the impact of a disruption on performance. Hendricks et al (2009) found that inventory slack mitigates the negative effect of a disruption on stock value, using actual disruptions announcements and publicly available data. Schmitt and Singh (2012) found through a simulation study involving a multi-echelon supply chain that inventory placement can have unforeseen benefits in the recovery from a disruption. Liu, et al. (2016) present an analytical model that allows a firm to perform a virtual stockpile of inventory to increase resilience and avoid excess inventory. The model operates at a network level by targeting a virtual transshipment effect that proves to be more cost efficient than simply keeping safety stock at a node level. While inventory buffers or safety stock have often been considered to be part of a supply chain resilience construct (Mandal, 2012; Blackhurst et al., 2011; Ambulkar et al., 2015; Carvalho et al., 2012; Park et al., 2016), high levels of inventory have been found to be detrimental to firm performance, and are associated with inefficiencies. High inventory levels have been associated with lower product quality and more product recalls (Steven and Britto, 2016) and lower sales (Ton and Raman, 2010). As a result, there are contradictory findings on the benefits of holding inventory in the different streams of literature. The findings of Chen et al. (2005) provide a path to better understanding this contradiction. Their empirical findings indicate that staying close to the industry mean on days-of-inventory delivers, on average the best operating performance. Firms that operated at slightly below the mean and slightly above the mean inventory 22 levels performed better than very lean firms and firms with much greater than average inventories. Based on this finding and the findings on redundancy from the supply chain resilience literature, we would expect that above the industry mean would be where a resilient firm will keep their inventory level. We pose the following hypothesis: H4b: Higher inventory will be positively associated with firm performance Knemeyer et al. (2009) proposes proactive planning and the creation of contingency plans as a risk mitigation strategy. Kamalahmadi and Parast (2016) identify proactive planning as the first stage of resilience and Blackhurst, Dunn and Craighead (2011) find contingency planning to be a resilience enhancer. Therefore, we expect that firms that identify potential disruptions and calculate recovery time scenarios, will be better performing firms because they have embedded resilience as a part of their organizational culture. Hence: H4c: Recovery resilience will be positively associated with firm performance Performance In the Supply Chain Resilience literature, resilience has been used as both an independent and dependent variable. Studies have examined the impact of resilience measures on costs (Liu, Song and Tong, 2016), disruption impact and depth (Ambulkar, Blackhurst and Grawe, 2015; Kim, Chen and Linderman, 2015; Carvalho, Barroso, Machado, Azevedo, Cruz-Machado, 2012) Other studies have focused on how to build resilience, studying antecedents, strategies, enhancers and 23 reducers of resilience (Christopher and Peck, 2004; Soni, Jain and Kumar, 2014; Blackhurst, Dunn and Craighead, 2011). While looking at the impact of disruptions, Hendricks and Singhal (2003) empirically test the impact of disruptions on firm performance using stock value, inventory costs, operational costs and long term stock value as a performance measures. They also test the mitigation impact of operational slack that sheds light on the impact of resilience. There is no study in the literature that empirically tests the impact of risk and resilience using a performance variable that encompasses both costs and revenue. In this study we are interested in the net effects of our variables on firm performance; for this purpose we chose gross margin as the performance measure. Gross margin measures the percentage of each dollar of revenue that the firm keeps after considering the costs of goods. It is a good measure for the understanding of the impact of risks and resilience on firm performance. To summarize, Figure 1 shows Gordon’s model published in the Academy of Management Review on the top (Gordon, 1991) and a conceptual model of our hypotheses on the bottom. Figure 1: Gordon’s Model and Hypotheses Model 24 METHODOLOGY a. Data In this study we combine data from 3 data sources. i. Resilinc Data Our first data source is from the firm, Resilinc. Resilinc provides supply chain risk management services to its clients. The Resilinc software tool allows firms to 25 track their supply chains, as well as their supplier’s supply chains, identify vulnerabilities in the supply chains, design resilient strategies, take mitigating actions to reduce vulnerabilities, and receive notification of disruptions using social networks, among other services. A Resilinc customer can use the software to map its internal supply chain and the supply chain of its suppliers, linking bills of materials to component suppliers and the manufacturing locations where these components are processed. Resilinc keeps track of news through social media notification, and provides announcements of potential disruptive events affecting a geographical area. Since the locations of a firm are mapped using the Resilinc software tool, the program forwards notifications with impact estimations to firm executives according to a notification hierarchy established by the firm that is also part of the tool. A customer is able to identify vulnerabilities across supply chain tiers, and identify if the firm will be impacted by a disruptive event and the potential extent of the impact. The customer can also assess the potential risk of its suppliers, its products or the geographical regions in which it holds operations. Resilinc provided us the population of disruption alerts that were sent during the year, 2014. These events are categorized into 4 types: 1) hurricanes, 2) fires, 3) earthquakes and 4) other. When a potential disruption is identified, Resilinc sends a notification customized for each firm with relevant details such as, the number of sites impacted and the potential revenue impact. The customer has the advantage of immediate notification, and managers can begin to make plans and decisions 26 providing the firm visibility into the disruption and the opportunity to recover from the disruption. For the purposes of identification, the Resilinc customers are referred to as the “focal firms”. These are the firms that are making investments in developing resilience for their supply chains. For this study, we were also provided with data from focal firms’ suppliers for the year, 2014. Our dataset, therefore, is a cross section for the year, 2014 consisting of data related to the supply chains of Reslinc customers. The database contains information on both tier 1 and tier 2 suppliers to the focal firms. The Reslinc service includes identifying “critical” production sites for the focal firms. This criticality is most often given by the fact that in those sites critical activities that affect high revenue products are performed. Therefore, an interruption to these sites could have high revenue impact. For example, these sites are often places were single-source activities are taking place. Since these places are linked to high revenue impacts, the risk assessment exercise includes business continuity plan and recovery time calculations. The dataset often contains information on multiple manufacturing locations per supplier firm. This information includes: 1) site geographical location (country and coordinates), 2) site risk scores (based on country risk scores provided by the Economist Intelligence Unit), 3) recovery time (self-reported analysis of disaster recovery given in weeks), 4) critical parts that are handled at that location and 5) actual potential disruptive events in the year 2014 that affected the geographical area where the facility is located. 27 We gathered a sample of firms from the Resilinc database that contains all the publicly traded firms and their manufacturing locations. We then matched these firms to firms in the Compustat database in order to get financial information on the firms. A total of 313 firms matched with Compustat. These are linked to 3,262 manufacturing locations, 75 countries, and 40 industries following a three-digit NAICS code. Even though the risk scores for each manufacturing location were provided by Resilinc, it is important to note that the geopolitical risk scores originated at the Economist Intelligence Unit (EUI). The EUI provides many services of risk assessment using scores from 1 (lowest risk) to 10 (highest risk). In our study, we use three of these scores to assess the geographical risk associated with a location: Geopolitical Risk, Natural Disaster Risk and Macroeconomic Risk. These are revised and updated every three to five years by the EUI, depending on the score and the country assessed. ii. Compustat In order to evaluate the effects of risk and resilience on firm performance, we use Compustat to obtained data on publicly traded firms. We use the year 2014 to calculate the gross margin, days of inventory and the Altman Zscore (a measure of financial risk). We use only publicly traded firms from the Resilinc dataset to match with the Compustat data set. The end result is a dataset that contains 313 firm level observations. 28 For the long-term variables, we use 3, 5 and 10 year Compustat industry aggregated data. Using a 3-digit NAICS code, we calculate the mean and standard deviation of the industry for inventory days of supply and gross margin. iii. International Labour Organization (ILO) Typically firms choose higher risk environments in order to benefit from lower costs of labor. Therefore, to complete our data sample, we use a measure for labor costs. The International Labour Organization provides data on minimum wages for countries. The minimum wage data is provided as a monthly wage according to the most recent laws in the various countries. For this study we use the year of 2012 and convert all the values to US dollars to allow comparability of costs. The year 2012 was the most recent year available for data provided by the ILO before 2014. iv. Final Database We started with over 1,000 firms and 25,000 manufacturing locations provided by Resilinc. 325 of the firms were publicly traded, with financial information reported in Compustat. 12 firms were eliminated. 4 were duplicates were found and deleted and 8 were name mismatches. 313 publicly traded firms were left. There are 3,262 sites attached to these firms. Therefore, the 2014 cross sectional sample contains 313 firm level observations. The data that were originally provided at the manufacturing location level are aggregated to create firm level variables. Geopolitical risk, natural disaster risk, minimum wage and recovery time are aggregated to the firm level. The total number of countries in which sites are located for a firm is a count variable at the firm level. The number of critical parts that are managed at a location and the events that 29 impacted each site are also aggregated as a count variable to the firm level. Days of inventory, gross margin and the financial resilience variable based on the Altman z score are calculated at the firm level using Compustat data. Finally the long term variables for days of inventory and gross margin are also calculated at the firm level. Hence, the 2014 cross sectional database has 313 firm level observations that include: 1) average geopolitical risk, 2) average natural disaster risk, 3) average minimum wage, 4) average recovery time, 5) number of countries, 6) number of parts, 7) number of events, 8) days of inventory, 9) financial resilience, 10) standard deviations from industry mean days of supply over 3, 5 and 10 years; as independent variables and gross margin and standard deviations from industry mean gross margin over 3,5 and 10 years as dependent variables. This database is divided into sub-sets for the purpose of analysis. Details about these groups are discussed with the models. v. Missing values Some firms failed to provide their recovery time and the number of critical parts managed in the site. In order to address these omissions, we use a standard procedure for estimating missing values and input the mean value for each of the above mentioned firms (Rencher, 2002). b. Variables i. Environmental Risks The environmental risks variables are chosen following the typology established by Rao and Goldsby (2009). Environmental risks are external to the firm and are related to the firm’s geographical environment. Out of the possibilities listed in their study, we include two environmental risk measurements: 30 1. Natural Disaster Risk The natural disaster risk score ranges from 1 to 10 and measures the probability of the geographical region where the manufacturing location is situated being hit by a natural disaster (hurricane, tsunami, earthquake, tornado, etc). 10 represents the highest probability of an event and 1 represents the lowest. This score is developed using Resilinc proprietary algorithms and may differ marginally from other publish scores since it takes into account different regions within a big country such as the United States. A natural disaster can temporarily block the capacity to move goods into and out of the country, as well as restrict the capacity to produce due to the country’s crisis. This measure is aggregated to the firm level by calculating an average between all the sites belonging to the firm. We refer to this variable in short form as “Natural”. 2. Geopolitical Risk The geopolitical risk scores range from 1 to 10 and measure the political stability of the country where the manufacturing sites are located. 10 represents a highly unstable country and 1 represents a very stable country. This score is developed and published by the Economist Intelligence Unit using proprietary algorithms to determine the probability of a country going through an invasion, a change of government, a coup, or experiencing some other kind of political instability. Political instability can impact the supply chain by blocking the normal flow of goods in the country. Government managed critical points, such as ports and borders, might not function properly under a political crisis, for example. This 31 measure is assessed at the firm level by calculating an average between all the sites belonging to the firm. We refer to this variable in short term as “Geo”. ii. Complexity We follow the definition of complexity that refers to the size of the network and the connectivity between the nodes as per Blackhurst, Dunn and Craighead (2011). For each manufacturing site in the Resilinc database, the number of critical parts managed for a manufacturing process are indicated. This variable provides a count of how many critical parts are processed at the location. The location count s are aggregated to the firm level by adding the total critical parts that are managed at all sites for the firm. Note that this is not the measure of a firm’s total number of critical parts, nor is it a measure of how many nodes there are in the network. It is a measure of how many times a critical part must be managed at a location. This variable provides an approximation of the complexity of a supply chain. In some cases, there may be multiple nodes for a particular critical part. iii. Disruptive Events In 2014 there over 90 disruptive events tracked by the Resilinc program. These events are grouped into 4 categories: Hurricanes, Fires, Earthquakes and Other. These events were provided in the data base with the geographical region that was impacted. The geographical coordinates of the regions allowed us to link manufacturing locations to events. Our variable Events, measures how many times a site is impacted by an event. This quantity is measured at the firm level by adding all the events of all the sites for a firm. This variable provides an estimate of how many times the firm had potential 32 disruptions in 2014. Note that this is not a measure of the impact that the events had on the firm, as it is possible that the site managed the event without interruption. With the frequency of disruptive events, we are measuring events that the firm faced, not the impact of the events on firm performance. This variable, therefore, is different from the disruptions variable used by Hendricks and Singhal (2003) that measured the impact of publicly announced disruptions. iv. Resilience Following the definition of resilience from Kamalahmadi and Parast (2016), we establish three measures of resilience for each firm: 1. Financial Resilience Financial resilience is a score from 1 to 10 based on the Altman Z-score. The z-score measures how likely it is for a firm to go bankrupt in the next two years. It can be used to assess the financial health of a firm. We introduce this score to the supply chain resilience literature as a measure of financial solidity. Following Fiksel et al, (2015), a financially solid firm would more likely be able to recover and bounce back from a disruption than a financially unstable firm. The score is normalized to a 1 to 10 measure. 10 represents the most financially solid firm and 1 represents the weakest firm. We refer to this variable in short form as “finres”. 2. Inventory We use days of inventory as a measure of inventory. Days of inventory standardizes the amount of inventory that the firm keeps regardless of firm size or industry. This variable is calculated using the following formula: (1/inventory 33 turns)*365, where inventory turns is calculated as: Cost of Goods Sold / ((Beginning Inventory + Ending Inventory) / 2). We refer to this variable in short form as “DOI”. 3. Recovery Resilience Recovery Resilience is adjusted to range from -10 to -1. Each site in the Resilinc database may report the time it would take to return to operations after the site has been shut down due to a disruption. Therefore, the variable does not measure actual recovery time, but projected recovery time following a disaster. An analysis of the impact of a disaster is conducted at a site, and an estimate of how long it would to return to regular operations is calculated. This time was reported originally in weeks and converted into a score from 1 to 10. The sign is inverted such that the higher the recovery resilience, the faster the site will recover. This adjustment is made to aid in interpreting the regression models. The fastest recovery is represented by -1 and slowest recovery is represented by -10. Then the site measures are assessed at the firm level by averaging the recovery resilience scores of all the sites belonging to a firm. We refer to this variable in short form as “RecRes”. v. Performance The dependent variable for this study is gross margin, a measure of firm performance. Gross margin allows us to understand how much per dollar of revenue a firm retains after the costs of production. Gross margin is calculated by subtracting the cost of goods sold from the total revenue and dividing it by total revenue. We use the short form “gmargin” for this variable. The dependent variable for the long term effects models is also based on gross margin. However, the variable measures the number of standard deviations above or 34 below the industry mean gross margin. To calculate this variable the mean gross margins over 3, 5 and 10 years of the 3 digit NAICS code industry were calculated, along with corresponding standard deviations. The firm’s average gross margin over those years was calculated as well. Then the following calculation is used: (firm average gross margin – industry mean gross margin)/industry gross margin standard deviation. We call these variables gmarginsd3, gmarginsd5 and gmarginsd10. vi. Controls Since the cost of labor is a major reason for operating in a country, we control for labor cost. We use the minimum wage reported for the country in which the site is located by the International Labor Organization for 2012. This amount is converted to dollars. This measure is assessed by an average at the firm level. We use the short name “mwage” for this variable. We have established through the literature that globalization has an impact on firm operations and that this can impact the firm’s performance. We control for spatial complexity per Bode and Wagner (2015) as the number of countries in which a firm operates. Finally, we control for industry effects through the model. The model adjusts the standard errors by clustering the firms by industry. We control for firm size by choosing the dependent variable to be a ratio, instead of Total Revenue or some other size-specific measure of firm performance. Table 1 shows a summary of variable descriptions and data sources. 35 Variable Name Source Description the degree to which political institutions are sufficiently Geopolitical Risk EIU stable to support the needs of businesses and investors Natural Disaster Risk Resilinc Probability of a natural disaster occurring in the region. How long it takes to go back to “normal” on an activity that Recovery Resilience Resilinc has been disrupted. Calculated: ( -10*recovery time/52) Min Wage ILO Min monthly wage for the country. (2012 USD) Countries Resilinc Number of countries the firm has operations in Critical Parts Resilinc Number of critical parts the firm is monitoring Altman Z score, measure of likelihood that the firm will go Financial Resilience Compustat bankrupt in the next 2 years. Calculated Events Resilinc Number of potential disruptions a firm had in 2014 DOI Compustat Days of Inventory. (1/inv turns)*365 Gross Margin Compustat (Total Revenue – COGS)/Total Revenue Number of standard deviations firm mean is below industry DOI SDS Compustat mean (Neg. No.) or above industry mean (Pos. No.) Number of standard deviations firm mean is below industry Gross Margin SDS Compustat mean (Neg. No.) or above industry mean (Pos. No.) Table 1: Variables and Data Sources Models In order to test our hypotheses we use a linear regression model with industry clusters. The general equation for all our models is: grossmargin = α1*(natural disaster) + α2*(geopolitical) + α3*(CritParts) + α4*(Events) + α5*(finresilience) + α6*(DOI) + α7*(resilience) + (controls) ………………………………………………(Equation 1) We estimate a robust clustered model using the Stata software package. This model controls for industry effects by adjusting the standard errors by industry clusters. i. Short Term Models (1,2,3) 36 The final database contains 313 firm level observations. Out of these 313, 156 contain data for the recovery time and 157 do not. The same 156 firms with recovery time data also have data on critical parts. In order to obtain robust results, we divide the database into three groups. The first group is named ALL and it contains all 313 firms. For the firms that are missing the recovery time and critical parts data, the mean values are assigned. Group 1 is constituted by the firms that do have recovery time and critical parts information, in order words, the firms with a resilience culture. Group 2 is constituted by the firms that do not have this information, in other words the firms without a culture of resilience. In this way, we can compare the results from all three groups. Model 1 is the regression ran with the group named All that contains all 313 firms. Model 2 corresponds to Group1 and Model 3 corresponds to Group 2. These three models use the cross-sectional data for 2014. ii. Long Term Models (4,5,6) Shein (1985), Barney (1986) and Gordon (1991) state that it is difficult to change culture because the deepest level of assumptions generate behaviors that are “automatic”, making them difficult to recognize and change. They all agree that if it is possible to change culture, it takes a long time. Working on this assumption, we would test our resilience variables over time. Model 4 corresponds to the group All, using the 10-year standard deviations from the industry mean gross margin variable as a dependent variable. Model 5 and Model 6 correspond to the 10-year models for Group 1 and Group 2 respectively. 37 RESULTS Table 2 shows the descriptive statistics for all variables for the three groups in the models: the total sample, the group that provided recovery estimates and that group that did not. Note that this table suggests that Group 1 and Group 2 might be different. Table 3 shows the correlation table for all variables in the main database. Table 2: Descriptive Statistics All Group 1 Group 2 Variable Obs Mean Std. Min Max Obs Mean Std. Min Max Obs Mean Std. Min Max Countries 313 4 4.820 1 42.000 156 6 5.639 1 42 157 3 3.253 1 22 Sites 313 24 165.863 1 2,922.000 156 42 233.758 1 2,922 157 6 9.637 1 78 CritParts 313 209 1039.636 0 14,153.000 156 419 1,444.626 0 14,153 Geo 313 3.005 1.077 1 6.5 156 3.248 0.958 1.667 6 157 2.764 1.136 1 6.5 Natural 313 3.709 1.490 1 9 156 4.140 1.384 1 9 157 3.281 1.472 1 9 finres 229 0.741 1.992 -9.563 10 122 0.807 2.513 -9.563 10 107 0.666 1.146 -4.197 5.167 DOI 301 92.146 70.765 0.652 567.778 154 84.261 54.355 0.652 398.897 147 100.406 84.021 1.692 567.778 mwage 306 927.149 332.343 42.557 1,565.885 156 849.413 302.673 177.625 1,565.885 150 1007.995 343.346 42.557 1,565.885 events 313 10 16.154 0 109 156 16 20.304 0 109 157 4 6.694 0 42 RecRes 156 -6.960 3.212 -10 -0.192308 156 -6.960 3.212 -10 -0.192 DOIstds10 262 -0.275 0.565 -1.143 2.996 142 -0.350 0.445 -1.143 1.591 120 -0.186 0.672 -1.066 2.996 gmargin 308 0.371 0.194 0.007 0.937 155 0.386 0.191 0.032 0.814 153 0.356 0.198 0.007 0.937 gmarginstds10 262 -0.026 0.873 -5.230 2.907 142 0.057 0.880 -1.870 2.907 120 -0.125 0.858 -5.230 1.522 Table 3: Correlations table Countries TotalParts Geo Natural finres DOI mwage events RecRes1 DOIstds gmargin Countries 1 CritParts 0.32 1 Geo 0.17 0.14 1 Natural 0.03 0.08 0.57 1 finres -0.05 0.07 -0.10 -0.06 1 DOI 0.02 0.16 0.09 0.16 0.25 1 mwage -0.14 -0.15 -0.87 -0.79 0.06 -0.07 1 events 0.72 0.19 0.25 0.30 -0.09 0.15 -0.27 1 RecRes1 -0.15 0.04 -0.06 -0.20 -0.0238 0.02 0.10 -0.12 1 DOIsds -0.03 0.09 -0.02 0.16 0.20 0.82 -0.04 0.12 0.02 1 gmargin -0.09 -0.15 0.01 0.29 0.30 0.49 -0.10 0.19 -0.02 0.44 1 gmarginsds10 -0.15 -0.20 -0.13 0.13 0.32 0.34 0.03 0.09 0.04 0.51 0.83 Note that there are high correlation between some of the independent variables. Events has a 0.72 correlation coefficient with Countries. This is to be expected since the higher the number of countries, the higher the events frequency. The Natural Disaster risk and the Geopolitical risk also have a high correlation with a coefficient of 0.57. We decided to keep both variables in all our models because these two risk scores provide different information that is relevant to our study. i. Difference of means test Before testing the hypotheses, we ran a difference of means test for all the variables in Group 1 and Group 2. First, we want to understand if these groups are statistically different. Table 4 summarizes the results for the means tests. As can be seen in the table, these two groups are statistically different in the environmental risk variables. Group 1 has a higher mean for both Natural Disaster risk and Geopolitical Risk than Group 2. In addition, the firms in Group 1 have operations in more countries and experience more disruptive events, on average. This result suggests that the firms in Group 1 operate in a riskier environment or have a higher propensity towards risk. These firms seem to be risk seekers. Interestingly, there is no statistical difference between the two groups when it comes to size (employees), revenue, gross margin in the year 2014 and financial resilience indicating that the firms are not differentiated by financial success or size. However, the firms in Group 1 seem to keep lower inventory levels than the firms in Group 2 at a 5% significance level. The firms in Group 1 practice more resilient 39 processes than the firms in Group 2. Recall we divided the sample using the recovery process and the critical parts tracking process as a differentiator. The only variable that is significant with a higher mean for Group 2, is inventory. This result could suggest that the firms in Group 2 cope with their risks by keeping higher levels of inventory. It could also be evidence of a combination of lean and resilience, as suggested by Harrington (2013). She revisits lean culture and suggests that it should be combined with a culture of resilience to maximize performance. Lastly, the measure for the distance of gross margin from the industry mean over ten years, is significantly higher at the 5% level for Group 1 suggesting that resilience might in fact be the source of a sustainable competitive advantage. Table 4: Difference of Means Tests Summary Group 1 Group 2 Mean Mean G1>G2 G1 ≠ G2 G1