Trust, Trustworthiness, and Business Success: Lab and Field Findings from Entrepreneurs 1 Mongoljin Batsaikhan 2

JEL code: O17, L26, C91 Keywords: social capital, trust game, small enterprises, entrepreneurship

Abstract: This paper contributes to the literature on trust and small businesses in developing countries by linking the level of trust in others that entrepreneurs display in a lab setting to sales data outside of the lab. The results show a robust positive correlation between being trusting and business success (with sales used as a measure of success): The successful entrepreneurs invested in trust more than the less successful ones did. In the lab, trusting in others is highly profitable, with successful entrepreneurs doing so to the degree required to maximize revenue. On the other hand, I found no association between the trustworthiness of entrepreneurs and business success. But an entrepreneur’s trustworthiness did predict the amount of loans he or she could secure outside of the lab. 1

I would like to thank Andrew Foster, Louis Putterman, Pedro Dal Bo, and Dan Westbrook for their detailed comments and Ross Levine and Nathaniel Baum-Snow for their support. This project is supported by the William R. Rhodes Center for International Economics and Finance at Brown University. I also thank the sales marketing department at New Tel LLC, the local office managers of Mobicom, the Mongolian National University of Science and Technology, and local research assistants, especially, Amarsanaa Dashdavaa, Dulamzaya Batjargal, and Munkherdene Gochoo. All errors are mine. 2 School of Foreign Services in Qatar, Georgetown University. Email: [email protected] 1

1.

Introduction Trust in others is an essential element of social interactions: When there is trust within

an exchange, it can solve social dilemmas and achieve socially-optimal equilibria (Ostrom and Walker, 2005). For example, in developing countries, where contracts are hard to enforce, committing to economic transactions with others on the basis of trust may confer a decisive advantage. Trust and trustworthiness among agents in such settings can work as social capital to solve the social dilemma problem. Indeed, a rich body of literature shows that the level of trust exhibited by various stakeholders explains variations in economic development (Keefer and Knack, 1997; Tabellini, 2008) and growth (Knack and Zak, 2001). 3 Trust can play an important role in business settings as well: Bloom et al (2012) argue that showing trust increases the productivity of firms because it motivates decentralization. In the management literature on how trust affects entrepreneurship, the theoretical predictions show that the social network of venture creation and growth determines the success of business, and that this network relies on trust between entrepreneurs (Aldrich, 2000; Anderson and Jack, 2002; Liao and Welsch, 2005; and Zahra et al., 2006). In fact, entrepreneurs may even over-trust when the future is highly uncertain, and this is not necessarily a negative factor in entrepreneurship (Goel and Karri, 2006). Empirical studies on trust and entrepreneurship are still few in number (Welter and Smallbone, 2006). But using survey data on trust, a recent paper finds that German entrepreneurs show a higher degree of trust than non-entrepreneurs (Caliendo, 2012). And Howorth and Moro (2006) show empirically that trust fosters trustworthiness, reducing the moral hazard and loan default problem in the Italian banking industry.

3

Trust can also be an important factor in trade and investment patterns (see Guiso et al [2009]) and participation in financial markets (see Guiso et al [2004]).

2

These two veins of literature—the first, on trust and development, and the second, on trust and entrepreneurship—suggest that trust plays an important role in both economic development and the private sector, especially among entrepreneurs. The private sector has been the driving force in the economic development of the OECD nations, but in developing countries, the majority of the private sector consists of small and medium-size firms that have failed to grow (Ayyagari et al, 2007). Trust in the context of these sorts of small entrepreneurial businesses in developing countries has not been studied extensively. To date, Karlan (2005) has shown that trustworthiness predicts loan defaults in the microfinancing sector in the Philippines, but that trust plays no similar role in these defaults. While Karlan’s result is consistent with that of Howorth and Moro (2006), it contradicts both the findings of Caliendo (2012) and the theoretical predictions on how trust affects entrepreneurship. With this study, I make a new contribution to the emergent literature on trust and entrepreneurship in the development context. In this paper, I link measures of individual trust and trustworthiness collected in a lab experiment to high-quality administrative data on realworld business outcomes for a sample of entrepreneurs in a developing nation: small business owners in Mongolia. Using this unique data set, I show that trust is an important asset for these Mongolian entrepreneurs and that it predicts their success in business. These results provide empirical evidence that is consistent with theoretical predictions on the topic, yet which is acquired—unlike in most previous studies—from a developing nation. The reasons that empirical studies on trust and entrepreneurship are still in short supply—despite the importance of this topic in the theoretical work—are multiple. General surveys, such as the World Values Survey (WVS), have certainly made it easier over time to measure trust on the macro level. But the WVS is not often used to analyze individuals or firms and the survey itself is not incentivized. And although approaches have been developed

3

to measure trust at the micro level using incentivized questionnaires in the lab, these approaches come with a particular set of methodological issues. The first of these problems stems from the fact that the data from the lab experiments commonly used to measure levels of trust in individuals and firms are not often connected to real-world economic variables outside of the lab. Moreover, data from actual business settings—which economists could compare with results from the lab—can be very difficult to come by, because typical business owners, first, do not participate in lab experiments and second, may be reluctant to disclose interesting data from their operations. Additional issues arise from the fact that lab experiments typically use student subjects rather than actual businesspeople, and so far, have been conducted mainly in western settings, so that results may not be generalizable to developing countries. By accounting for each of these methodological barriers, my study design also makes a broader contribution to the empirical literature on trust. In contrast to student subjects, my sample population consists of actual business owners: wholesale and retail entrepreneurs working in the mobile phone industry in Mongolia—which is a developing nation rather than a western country. The administrative data come directly from the real-life businesses of these Mongolian entrepreneurs, consisting of sales records logged by the computers that process all business transactions involving mobile credits. 4 This automatized collection process makes these data of higher quality and much more reliable than most field survey and administrative data gathered in developing countries. With this dataset, I am able to measure 4

In Mongolia, wholesalers often do business with retailers either on the basis of their previous business interactions or via referrals by friends, family, and other reliable business partners, rather than through formal contracts. In a pilot survey that I conducted in 2009, some retailers indicated that before the current electronic transaction system was built, wholesalers extended them business credit without a contract. While the business credit system is no longer used in Mongolia’s mobile phone industry, the trust measure in the lab in a pilot study that I conducted in 2009 was positively associated with the length of time that a business had been operating (only at the 10% level). All of this implies that trust is quite important in the Mongolian business environment, making it a fascinating context in which to look at how being trusting affects economic outcomes.

4

the sales success of each subject in the sample and can then correlate the behaviors of those individuals in the lab with their performance in the real business world. Using this combined lab and administrative data, I am able to establish that (1) trusting in others is highly profitable for this sample of Mongolian entrepreneurs in the lab setting, (2) the more successful entrepreneurs invest in trust more than the less successful ones do, and (3) the successful entrepreneurs invest in the optimal amount of trust required to maximize profit in the lab. The subjects who invest more in trust, being more trusting of others, are also more successful in their real-life business activities. This suggests that trust plays an important role in business and that successful entrepreneurs effectively determine the right degree of trust to exhibit in order to make their businesses profitable. In other words, successful entrepreneurs possess both the “private capital” to assess the level of trust and trustworthiness in the population, and the ability to optimize the level of trust they exhibit to maximize their own revenue.

2. Measuring Trust Measuring trust is difficult, and approaches vary even within different fields of economics. The simplest way of measuring it in a sample is asking respondents whether most people generally can be trusted. The World Value Survey calculates the national average of this measure of trust for each country. Previous studies (for example, Keefer and Knack, 1997; Bloom et al, 2012) that linked trust with business, or any other economic outcomes, often used the WVS. However, trust in the WVS in these previous studies is only measured at the country level and the elicitation is not incentivized. In contrast, the direct measure from a trust game—an incentivized questionnaire originally designed by Berg et al (1995) and commonly used in experimental economics—is

5

able to quantify individual levels of trust. I use the strategy method of this trust game in my research design for this paper, which permits observation of the underlying distribution of trustworthiness in the sample. Economists have used the strategy method of the trust game in a number of experimental settings to analyze the behavior of subjects facing social dilemmas. In this form of the game, subjects have to make decisions conditional on all possible outcomes without learning the opponent’s decisions. 5 The game involves two players: A and B. Each of them is endowed with a given sum of money. Player A makes the first move by deciding how much of her endowment to send to Player B (she is also permitted to send a zero amount, if she wants). The experimenter will triple any positive amount sent by Player A and then pass it on to Player B. After receiving the tripled amount, Player B has the option of sending back some, or all, of the amount received. Player B also has the option of keeping all of the money. The non-cooperative Nash equilibrium for the trust game is that Player A should send a zero amount, after predicting that a self-interested Player B will return nothing. However, most subjects in laboratory experiments deviate from this theoretical prediction: Player A usually sends a positive amount and Player B usually returns a positive amount as well. If trustworthiness is expected in a society, Player A has an incentive to trust Player B and invest a positive amount into the original transfer as a sign of trust. In experimental economics, the amount that Player A sends to Player B is considered to be a measure of trust, whereas the amount that Player B returns to Player A is considered to be a measure of trustworthiness. 6 The social dilemma that subjects face in this game is twofold: for Player A, whether, and how much, to trust others; and for Player B, how trustworthy to be. In this way, this experimental 5

Brandts and Charness (2011) show that the trust game strategy method does not produce substantially different results than direct responses. And several years before them, Barr and Lindelow (2005) assessed the strategy method’s validity in a trust game in a developing country. They showed that only 23% of the Ethiopian subjects exhibited a sign of misunderstanding the game and that the benefit of the strategy method outweighed the potential problem with comprehension among such a small percentage of the sample. 6 See Eckel and Wilson (2009) for a detailed discussion of the definition of trust in the context of the trust game.

6

approach is similar to a real-world situation and also provides both trust and trustworthiness measures. Whether or not trust is profitable depends on trustworthiness. If Player B returns more than 33.3% of the tripled (by the experimenter) amount that Player A originally sent him, then it is said that being trusting is profitable. If trust is profitable—or at least if subjects expect trust to be profitable in their society (i.e., they expect more trustworthiness from other subjects)—then being more trusting of others can be the optimal strategy for Player A. Even when being trusting is profitable, the optimal amount of trust that maximizes earnings will depend on both the underlying distribution of trustworthiness and how the subjects perceive that distribution. Player A tries to maximize her profit on the basis of her belief about how much Player B will return. Player A has to guess the distribution of the proportion that Player B will return out of the tripled amount he receives from Player A. With that in mind, Player A must then choose the level of trust to exhibit that will maximize her overall profit. Previous studies indicate that successful individuals tend to be accurate in capturing the relevant, but unobserved distribution parameters. 7 Despite its usefulness, measuring trust directly in the lab raises two thorny issues pertaining to external validity. First, in experimental economics, any adequate understanding of results from the lab must be grounded in the study of subjects' behaviors in a relevant setting outside of the laboratory. 8 Moreover, the ability to generalize the results depends on the subject pool in which trust is measured.

7

In a two-person, zero-sum game, Palacios-Huerta (2003) finds that professional soccer players make decisions that are close to the Nash equilibrium when they accurately predict the likelihood of the actions that other players will take. The field data in Hsu et al (2007) and Palacios-Huerta and Volij (2008) back up this finding. Levitt et al (2010) suggest that professional players are able to transfer their ability to play mixed strategy into the lab setting when what they face in the lab is similar to what they face in the field. 8

For a discussion of the relative advantages of lab, field, and lab-in-the-field experiments, see Harrison and List (2004).

7

2.1. Can Social Preferences as Measured in the Lab Be Generalized to Behavior Outside of the Lab?

For many years, experimental economists measured social preferences in the lab under the assumption that subjects will bring their normal, real-life preferences into the lab setting. However, in recent years, a growing number of voices have challenged this assumption. Levitt and List (2007) were the first to take a more skeptical view, raising concerns over experimenter demand effect, framing and context effects, non-anonymity, stake size effects, student subjects, and selection. They were followed by List and Cherry (2008), Bardsley et al (2009), Krawczyk (2011), Jackson (2012), and Cleave et al (2013). 9 Although traditional economics experiments are not designed to test external validity, recent studies have linked data from the lab to real life with the intention of providing evidence to counter this skeptical viewpoint. Camerer (2015) gives a comprehensive and detailed review of recent studies on this topic that take a more positive stance. Camerer shows that results in the lab are often comparable to data collected outside of the lab, when the lab experiments are designed to test the lab-to-field generalizability. The only exception to the overall positive stance in Camerer’s review is Stoop et al (2012), where the results actually tilt more toward the skeptics’ side of the debate when subjects are working adults from business settings (rather than the usual student subjects). Dutch adults in the sample studied by Stoop et al (2012) exhibit opposite behavior in the lab and in the field. This result could simply be due to the local culture, rather than a proclivity within the general adult population. But unfortunately, there are not many studies to date using subjects from the business world, with which the findings of Stoop et al could be compared.

9

For a recent summary of this view, see Galizzi and Navarro-Martinez (2015)

8

The work I present in this paper contributes to this debate in several ways. Not only is the research design intended to test the generalizability of the experimental measures by linking the performance of subjects in and outside the lab, but it uses subjects who are actual business owners rather than students. In this way, it offers a point of comparison to Stoop et al (2012). Moreover, it adds another dimension to the debate by using a subject pool from a non-western country. 2.2. Generalizing Results from the Lab When Subjects Are College Students

While lab experiments in economics have traditionally used student subjects, recent studies show that older subjects are more pro-social than students. Using a trust game, Kocher and Sutter (2007), Bellmare and Kröger (2007), Falk et al (2013), Belot et al (2015), and Hoffman and Morgan (2015) all show that older, non-student subjects from the general population exhibit a higher degree of trust. It is worth noting that there are two studies in which business people have actually played the trust game. The first one is Fehr and List (2004), who compare Chief Executive Officers (CEOs) with students. The authors find that the CEOs are considerably more trusting and achieve higher levels of efficiency than the students. Hoffman and Morgan (2015) show that students in the lab are less trusting than internet business people in a field set-up—even though the business people are working in a highly competitive industry. 2.3 Generalizing Studies in which the Subjects Are from Highly Developed Countries

There is an increasing concern among social scientists about the validity of generalizing results from studies that recruit subjects from highly developed countries where culture, the social context, institutional setups, and social norms are fundamentally different from those in developing countries (Henrich et al, 2010). This concern is particularly pertinent when it comes to an area of inquiry such as trust. Unlike in highly developed nations, legal institutions are frequently insecure in developing countries. Business owners in 9

these countries often have tenuous property rights and must engage in business relationships on the basis of trust rather than formal legal contracts. As a result, trust can play a substantial role in business in general, and in entrepreneurship specifically, in these countries. But it quite possibly does not play the same kind of role in highly developed countries. Yet despite the question mark about generalizability and the acknowledged importance of the topic of trust in development economics, trust has rarely been measured at the micro level in the development literature. Indeed, there are only a handful of existing experimental studies on this topic that use subjects from a developing country. Cardenas et al (2009) use Berg’s trust game with subjects from six Latin American cities, and find that these subjects exhibit similar results to subjects in the United States. However, Barr (2003) looks at Zimbabwean subjects and finds that they show a slightly lower level of trust than the US subjects. Carpenter et al (2004) do an experiment with Vietnamese and Thai subjects, and find that they are more trusting than the subjects in both Barr (2003) and Cardenas et al (2009). Because the results of these studies are so varied, they leave the question of generalizability as yet unanswered.

3. Data and Experiments 3.1. Subjects and Data

I recruited 120 subjects from among wholesalers and retailers in the mobile phone industry in Ulaanbaatar, the capital of Mongolia, to participate in two experimental games. Wholesalers have contracts with local mobile phone companies to distribute prepaid phone cards to retailers in the city. Most of these wholesalers specialize in prepaid cards. Retailers, in contrast, tend to sell other products as well—typically groceries, cleaning supplies, and alcoholic beverages.

10

To recruit retailers and wholesalers for the experiments, I went to Mobicom, the largest mobile carrier in Mongolia, circulating an advertisement through local Mobicom offices. I divided those who applied into six groups and invited them to the laboratory, group by group. Each of these six groups consisted of 22 to 27 subjects. 10 In addition, I drew administrative sales data 11 for May 2010 from Mobicom’s server computer. This computer digitally tracks all Mobicom prepaid cards sold at both the wholesaler and retailer level. 12 By observing which wholesaler was connected to which retailer, I was able to compute the sales of each wholesaler as well. 3.2. Experiments and Procedure

With each of the six groups, I conducted two experimental games. The first was the trust game, as described in Section 2. The second was a risk-aversion measure game—similar to the risk game proposed in Holt and Laury (2002)—to obtain a measure of each individual’s attitude toward risk. I used this measure to illuminate the effect of risk attitudes on outcome variables in regression analyses. At the beginning of each game, I explained the rules and payment information to the subjects, and gave them an opportunity to ask any questions they needed to clarify the rules. I ensured that the experiments were done anonymously, so that the subjects could make their choices without worrying about others identifying those choices. 13 After the experiments were completed, all subjects took an exit survey. The survey requested information on age, gender, the total amount of bank loans held

10

Two subjects could not participate in the experiment due to over-enrollment, and so they earned only the show-up payment of 5,000 MNT. 11 Ideally, I would have used profit, rather than sales data, because businesses incur costs. However, cost data are hard to measure and often estimated structurally in the industrial organization literature. In this case, thanks to the new technology that transformed scratch cards into digital form, the cost of digital phone cards is minimal because they do not require transportation or storage space. Thus, the sales data on the cards are a good proxy for profit. 12 When a customer pays a certain amount to a retailer, the retailer deposits an equal amount from his or her business cell phone account into the customer’s cell phone account. This transaction is recorded by Mobicom’s server computers. For details of the original population data, see Batsaikhan (2014). 13 The instructions were provided in Mongolian. The English translation of both the instructions and the actual questions asked in the trust and risk measure games can be found in the supplementary material.

11

by the subject, the interest rate on any such loans, if the subject was a wholesaler, the length of time (in years) that the subject had been in business, and whether the subject owned property. 3.2.1

The Trust Game

For my lab experiment, I used the incentivized questionnaire, or well-established trust game from Berg et al. (1995). I took advantage of this trust game strategy method to obtain detailed data on the variation in the degree to which the Mongolian entrepreneurs invest in trust in the lab setting. I also used it to uncover the underlying distribution of trustworthiness in the sample. In the trust game, I asked each subject to indicate her choices as if she were playing the part, first, of Player A, and then, of Player B. After all of the subjects made all possible choices, I randomly assigned them the roles of either Player A or Player B, and then paired them in random order in order to calculate the trust game payment. I explained the steps in this process to the subjects during the initial instructional period to ensure that they did not, and would not, know with whom they would be paired. The trust game data include the measure of trust, the amount that each subject sends as Player A (denoted as the t1 variable), and all possible returns for any given t1 (denoted as the t2 variable). The variable t1 starts at zero and increases by increments of 500 Mongolian Tugriks (MNT) up to 6,000 MNT (approximately US$5 at the time of the experiment). Each

subject chose one of those increments as their t1. For each value of t1, I calculated the percentage of money returned out of the tripled amount (i.e., the amount equal to t1 times 3). The measure of trustworthiness is the average of all possible percentages for each t1. This is denoted as t2p, or formally stated: 1

𝑡2

𝑚 𝑡2𝑝 = 12 ∑12 𝑚=1 3∗500𝑚

12

(1)

where 𝑡2𝑚 stands for the choice of t2 for any given m (ranging 1 to 12). 3.2.2 The Risk-Aversion Measure Game

My motivation for running a risk attitude game was that an individual’s particular level of risk aversion might affect his or her decision about how much to send initially as Player A. While Eckel and Wilson (2004) show that trust is not correlated with various risk measures, Bohnet et al (2008) find that aversion to betrayal is another type of risk aversion. It seems probable that betrayal aversion could well be related to how trusting an individual is. To investigate this hypothesis, I analyzed to what extent risk aversion may have played a role in the subjects’ decisions in the trust game in the laboratory and in their real-world sales figures. In the risk-aversion measure game, I asked the subjects to answer five different questions, each of which offered a choice between a certain payment (i.e., a safe asset) and an uncertain payment (i.e., a risky asset). The first question asked whether the subject preferred to get 1,500 MNT with certainty, or alternatively, to take a 50% chance of getting 0 MNT and a 50% chance of getting 2,700 MNT. For each of the remaining four questions, the third amount in the formulation increased progressively: to 3,000 MNT, 3,500 MNT, 4,000 MNT, and finally, in the fifth question, to 4,500 MNT. 14 The risk-aversion measure, frisk, is the switching point (measured on a scale of 1 to 5): that is, the question after which the subject starts to choose risky assets consistently. For example, if a subject were to choose safe assets in the first two questions, but risky assets in the last three questions, the value for frisk would be 3. For the subject in this example, at this switching point, she begins to prefer risky assets, finding it more worthwhile to possibly get 14

After subjects answer all five questions, they roll a die to determine which question out of the five will be the basis for calculating their payment. If the die shows a one, it will be question 1, if it shows a two, it will be question 2, and so on. If the die shows a six, I ask the subject to roll the die until he or she gets a number less than six, because there are only five questions in the game. Once the question is determined, if the subject chose the risky asset in that question, then he or she rolls the die again. He or she receives the larger amount if the die shows an even number and zero if the die shows an odd number.

13

3,500 MNT or more with a probability of 50% than to get 1,500 MNT for certain. In this scale of measures, a value of 1 indicates the most risk-tolerant subject, and a value of 5 the most risk-averse. While my rendering of the game was a simplified version of Holt and Laury (2002), 35 out of the 120 subjects in my sample did not make consistent choices and switched more than once within five questions. Charness et al (2013) show that such inconsistency is a common problem among non-student subjects playing this game in the lab—as the wholesalers and retailers in my sample, indeed, were. But I had a lower percentage of inconsistent answers than investigators in previous studies with non-student subjects did. The summary statistics for the two games are shown in Table 1.a. [Insert Table 1.a. here] 4. Empirical Results In this section, I first show the results from the trust game. Next, I examine the relationship between trust and the non-experimental variables—especially the sales data indicating the subjects' real-world business success. I then compare individual decisions about how much to invest in trust with the distribution of the entire subject pool, and analyze how optimal those individual decisions are, given that distribution. Lastly, I analyze how risk aversion affects these results. 4.1. Analysis of the Trust Game

On average, the Mongolian entrepreneurs in my sample who were playing the role of Player A sent 3,092 MNT—or 51.53% of their endowment—to Player B. The entrepreneurs playing the role of Player B then returned to Player A, on average, 51.5% of the amount that they received (i.e., the amount sent to them by Player A, which I had tripled). These results show trust to be highly profitable. 14

Figure 1 presents the proportion and the actual amount of t2 (trustworthiness) for every t1 (show of trust). The percentage of t2 is greater than 33.33%, which is the boundary above which trust is profitable. Although the percentage of t2 falls as t1 increases, the earnings increase as t1 increases. [Insert Figure 1 here] While the trust game in my experiment did not explicitly elicit the expected trustworthiness of each subject, use of the strategy method allowed me to calculate the earnings for each subject under the assumption that the level of trustworthiness to be expected for each subject is reflected in how she answered those questions in the game related to trustworthiness. In other words, I could calculate the earnings for each subject using her answers as Player B. For each t1, let m range from 1 to 12 and express the trust measure as t1=500m (m={1,2, …, 12}). Then the equation for the earnings for subject i is: 𝐸𝑖𝑖 = 6000 − 500𝑚 + 𝑡2𝑖𝑖

(2)

where 𝑡2𝑖𝑖 is how much subject i returns when t1=500m.

For every m, I calculated the average of these earnings across subjects in order to

calculate the mean earning when Player A sends 500m. The result, shown in Figure 2, indicates that these earnings reach the maximum when the subject sends the entire endowment. This means that subjects should send a higher t1 to maximize their earnings; however, the show of trust observed among the subjects in my sample is lower than this optimal show of trust, given the underlying distribution of trustworthiness. The result holds when the average t2 is used instead of each subject’s own t2, for each t1. [Insert Figure 2 here]

15

4.2. Trust and its Link to the Sales Data and Other Non-Experimental Variables

Among the Mongolian entrepreneurs in my sample, being trusting turns out to be profitable. Moreover, the larger the amount invested in trust, the greater the return. If the behaviors shown in the experimental setting are representative of those outside of the lab, then those subjects who invested more in trust in the game should be more successful in real life, since trust capital pays off. Indeed, Figure 3—a simple scatter plot of trust and sales— shows a positive correlation between the trust measure from the experiment and the sales data from the field. [Insert Figure 3 here] Table 2 exhibits the results of the OLS regressions, which I conducted to study the robustness of this correlation, both with and without controlling for trustworthiness, demographic variables, and other business-related indicators (e.g., bank loans, whether or not the subject is a wholesaler dummy). I calculated the economic significance of the relationship between trust and sales by taking the coefficient from the last specification of Table 2, which is the most conservative estimate among the six specifications. Increasing the trust measure from 0 to 6,000 increases sales by at least 9,828,000 MNT. This is approximately 1.12 standard deviations of sales. [Insert Table 2 here] Trustworthiness in the lab, in contrast, is actually negatively correlated with real-life sales success, after controlling for trust—even though this finding is significant only at the 10% level. This indicates that when levels of trust are equal, being more trustworthy is not positively associated with success in business. With regard to the other non-experimental variables on which I had data, I find that male subjects tend to send more money (t1) when playing the role of Player A, but neither 16

age nor age-squared predicts t1. 15 Table 1.b shows that trustworthiness is correlated with the amount of bank loans the subject holds. I also find trustworthiness to be correlated with age and the number of years of business experience that a subject has. [Insert Table 1.b here] 4.3. Optimizing Trust to Maximize Success

In Section 4.1, I reported that the subjects participating in the trust game maximize their earnings when they exhibit the maximum trust. However, in that context, I calculated the optimal level of trust (the one that maximizes earnings) from the average trustworthiness of all 120 subjects in the experiment. Recognizing that each individual subject might have a unique optimal level, based on his or her belief about how much Player B will return, I use two different ways of arriving at the t2 to find the optimal t1 that maximizes earnings. In the first of the two approaches, I infer each subject’s perception of t2 from their own choice for t2. I calculate individual earnings for each t1, using each subject’s own t2 for every t1, and then identify the t1 that maximizes the expected earnings. In other words, I assume here that each subject’s own t2 represents their guess as to how much t2 others would return for a given t1. Mathematically, for every i, I find t1=500m that: max𝑚 (6000 − 500𝑚 + 𝑡2𝑖𝑖 )

(3)

This is the optimal trust level for each subject. It might be different from his or her actual choice of t1 in the experiment—that is, his or her actual trust level—depending on how accurately each subject guesses the underlying distribution of trust and trustworthiness. I then calculate the absolute difference between each subject’s optimal and actual trust levels and analyze whether that difference is correlated with the real-life business data for 15

Similar results for gender have been found in one study of urban slums in Vietnam and Thailand (Carpenter et al, 2004).

17

that subject. Figure 6 shows that entrepreneurs with higher real-life sales tend to deviate less in the game from their optimal choice of how trusting to be. This correlation indicates that when playing the game, entrepreneurs who are successful in their actual business dealings tend to get closer to the optimal level of trust that will maximize their earnings. [Insert Figure 6 here] I check the robustness of this correlation with a regression analysis. Table 3 presents the results from the OLS regressions (with the absolute difference as the dependent variable). Here we can see that the successful entrepreneurs in the experiment optimize their expected earnings better than their less successful counterparts, and that this result is robust after controlling for business and personal characteristics. [Insert Table 3 here] I take a second approach to arriving at the t2 by using the average of t2 for each t1; that is, the average percentage of trustworthiness for all 120 subjects. I then use this as each subject’s perception of t2p in order to derive the optimal trust level that will maximize his or her expected earnings. Table 4 shows the results of a regression analysis similar to the one I used to check the robustness of the first approach. The results here are very similar to those in the first analysis, both with and without control variables. [Insert Table 4 here] 4.4 Analysis of the Risk-Aversion Measure and Robustness

In his 2005 study, Karlan has adult subjects working in the business world play the trust game. He concludes that the degree of trust that they exhibit in the game simply reflects their level of risk aversion, because he does not find any correlation between the trust measure and whether a given subject had paid back his or her loans in real life. In light of this

18

conclusion, I analyze the level of risk aversion among the subjects in my sample by having them play a risk-aversion measure game (explained in Section 3.2.2). Because about a third of the subjects switched their answers to the five questions more than once while engaged in the game, I analyze the data in two ways. In the first of the two, I exclude the subjects who answered inconsistently in the risk-aversion game and then replicate the analyses presented in sections 4.1, 4.2, and 4.3. In the second approach, I retain the full sample, but concentrate the analysis on whether the inconsistency in some subjects’ answers affects the relationship between sales and trust shown in Table 2. 4.4.1 Subjects with Inconsistent Answers Excluded

I replicate the correlation between trust in the lab and sales in the field (along with other variables controlled for in Table 2) using the risk-aversion measure taken from the subjects who did not change their answers during the risk-aversion game more than once. While excluding the subjects with inconsistent answers reduces the sample size to 81, the level of risk aversion among the remaining subjects (those who answered consistently) does not affect the correlation between trust and sales. Specification (1) in Table 5 shows the results. The magnitude of the coefficient on trust even actually increases a bit. Moreover, the correlation between each subject’s degree of deviation from the optimal level of trust and their real-life sales (Table 3) is also robust when I control for the risk-aversion measure among this remaining group of subjects (specification (4) of Table 5). To check the robustness even further, I control for the risk-aversion measure as an indicator variable for each of the five questions in these two regressions. Specifications (2) and (5) in Table 5 report these results, respectively, for the correlations between trust and sales, and sales and deviation from the optimal level of trust. Again, the only difference is a minor increase in the magnitude of the relationship between sales and the dependent variables.

19

4.4.2 Focus Placed on the Effects of Inconsistent Answers

When I analyze whether inconsistencies in the answers to the five questions in the risk-aversion measure game affect the correlation between sales and trust, there is a minor increase in the magnitude of the coefficients in Table 5, compared with the ones in Tables 3 and 4. It is possible that this increase is due to the subjects’ cognitive ability and as such, this would bias the main results. For example, Blau and Kahn (2005) and Heckman et al (2006) show that cognitive ability explains the variation in the labor market outcomes. Among small business owners in a developing country, De Mel et al (2008) suggest that greater cognitive ability is associated with firm growth. In my sample, I do not have a measure for cognitive ability. However, if subjects’ inconsistent answers are due to their level of cognitive ability, I can use the inconsistency as a proxy for cognitive ability and check whether the correlation between sales and trust changes. Although the number of the inconsistent subjects is lower in my study than in other studies (Charness, 2013), this must also be acknowledged as a factor potentially affecting the results. As a safeguard against this, as well as potential bias due to cognitive ability, I do use the dummy variable for inconsistent subjects—and replicate the results in Table 2 with the whole sample. Specifications (3) and (6) in Table 5 show that the correlation between sales and trust is robust to this inconsistency and that the magnitude of the coefficients is similar to the ones that appear in Tables 3 and 4. In other words, trust is a more robust measure than cognitive ability, as a means of predicting success in business. 16 [Insert Table 5 here]

5.

Discussion

16

Since the sales variable can take only positive values, I also used the Tobit regressions instead of OLS as a robustness check. I do not show these results in this paper (they are available upon request). But all results from the Tobit regressions are similar to the ones that I report here.

20

In this study, I utilize a trust game—an incentivized questionnaire, developed in experimental economics—with a sample of non-student, entrepreneur subjects in a developing country to measure each’s level of trust in the game and relate it to their success in business. I show that successful entrepreneurs, in fact, exhibit more trust than less successful ones. This is consistent with the predictions in the theoretical literature on trust and entrepreneurship (Aldrich, 2000; Anderson and Jack, 2002; Liao and Welsch, 2005; and Zahra et al., 2006). In terms of the empirical papers on the topic of trust and entrepreneurship, my findings are similar to those of both Caliendo (2012) and Howorth and Moro (2006). But contrary to those two studies, which use a survey measure of trust, I use the incentivized measure of trust as the core of my research design. In addition, whereas those studies use subjects from developed western nations (Caliendo’s from Germany, and Howorth and Moro’s from Italy), the subjects in my study are from a developing country, Mongolia. The incentivized trust game has not been run in many business settings in developing countries. It has been run using non-student subjects several times in developing countries, but without being linked to the data outside the lab. Barr (2003), Carpenter (2004), and Cardenas et al (2009) are the three primary examples, but their non-student subjects are not business people. Still, it is interesting and informative to compare my findings about entrepreneurs with theirs, as the common ground among all four studies—that the nonstudent subjects are from developing countries—is territory as yet little explored in the broader economics literature on trust. The average level of trust exhibited by the Mongolian subjects in my study lies somewhere between that of the Zimbabwean subjects in Barr (2003) and the Thai and Vietnamese subjects in Carpenter (2004). It is closest to that of the Latin American subjects in Cardenas et al (2009), who also exhibit an average level of trust in between that of the subjects in Barr’s and Carpenter’s papers.

21

The main difference between my study and these other experimental investigations done in developing countries is the uniqueness of the subjects in my sample. I recruited only business owners for my study. The only other paper to date that employs the trust game among business owners in a developing country is Karlan (2005). But there are key differences between my paper and Karlan’s. One of the most significant is that the Filipino subjects in Karlan (2005) could not read. In Karlan’s experiment, the instructions for the trust game were read to them. While this might not have had an impact on their comprehension of the experiment, it is an indication of a substantially different type of subject than I was able to recruit for my work. The Mongolian entrepreneurs who took part in my experiment are all literate. The nation currently has a literacy rate of more than 98%. The previous communist government made secondary education mandatory and almost everyone in the country today has completed it. Unlike in Karlan’s study, the subjects in my sample were given the instructions for the trust game in writing, as well as hearing them read aloud in their native language. The subjects in Karlan (2005) are substantially different in another way as well from the entrepreneurs who participated in my experiment. Karlan’s sample consisted entirely of female subjects who had received microfinance loans. Even though these loans are designed to help individuals expand small businesses, women who take them out in developing countries are not always considered to be entrepreneurs, as the majority of them use their loans to cover household expenditures rather than business expansion (Karlan and Morduch, 2010). In contrast, the subjects in my study are active business owners who are working to grow their professional enterprises. A point of difference between my study and many of the other empirical studies that employ the trust game more broadly is the level of trustworthiness among the sample, which determines whether trust is profitable. Among the Mongolian entrepreneurs, the level of 22

trustworthiness is high enough to make a show of trust profitable. In contrast, Karlan (2005) and Barr (2003) both show that trust is not profitable for the subjects in their experiments. Looking beyond those two studies, among 38 trust games summarized in Johnson and Mislin (2011), on average, Player A sends 50.8% of A's endowment to Player B. The 51.53% of their endowment that the Mongolian entrepreneurs in the role of Player A send, on average, to Player B in my study is consistent with these results.

However, in the 38 games

summarized in Johnson and Mislin (2011), the return on trust is only 36.5%—just slightly higher than the 33.3% bar above which trust becomes profitable. Although this result does indicate that most subjects deviated from the noncooperative Nash equilibrium, trust garners a very low gain in the Johnson and Mislin meta-analysis. This is far from the case in my experiment. Player B in my trust game returns to Player A an average of 51.5% of the tripled amount received—a full 15 percentage points more than the average in the other 38 games. This marked difference between my study and those summarized in Johnson and Mislin (2011) comes from the higher level of trustworthiness among the Mongolian subjects. The level of trust itself is the same among the subjects in my sample and other subjects in Johnson and Mislin (2011). But trustworthiness is greater within the Mongolian sample. 17 Because this trustworthiness distribution makes trust highly profitable, the individuals in the Player A group in my experiment have a strong incentive to invest in it. And indeed, the more successful subjects invest in close to the profitmaximizing level of trust by successfully guessing the underlying distribution of trustworthiness of the whole sample. This could very well be because successful entrepreneurs can effectively discern when to trust more.

17

The Mongolian sample shows a lower degree of trustworthiness, however, than the Vietnamese and Thai subjects in Carpenter et al (2004).

23

6. Conclusion Trust in others functions as essential social capital for economic activities. It is especially important in developing countries where legal contracts are often poorly enforced. However, due to the difficulty of simultaneously measuring levels of trust and real-life business variables at the micro level, empirical evidence on the role played by trust in business settings is still quite limited. My paper contributes to filling this gap. I use a game developed in experimental economics to measure the levels of trust exhibited in a lab setting by small business owners from Mongolia, and then link that lab data for these individuals to sales data on their business success in the field. I find that the trust measures obtained in the lab are, in fact, associated with the subjects’ sales volume outside of the lab. These results indicate that trusting others is highly profitable and that successful entrepreneurs are more trusting than less successful ones are. In addition, successful entrepreneurs also invest in close to the optimal amount of trust, according to different measures. In this research, I do not attempt to identify any causality between the show of trust and business outcomes. It is very difficult to clearly identify a causal direction between these two variables, due to the lack of available data and the difficulty of randomly assigning trust to individuals. Indeed, this is a problem that goes well beyond my study: It is due to the nonexistence—or at least the rarity—of simultaneous measures of both variables that the relationship between trust and business outcomes at both the firm and individual levels is generally understudied. To my knowledge, no study has attempted to rigorously identify causality between degrees of trust and success in business precisely because of the considerable challenges involved. Accordingly, my goal in this work was more modest: to investigate in a robust way whether there is an association between trust and successful entrepreneurship at the micro level. My findings do establish such an association, showing

24

that successful entrepreneurs in developing countries utilize displays of trust more advantageously than less successful ones.

25

FIGURE 1 Average amount and percentage of t2 for any t1 0.7

Trustworthiness at any given trust level

10000 9000

0.6

8000

0.5

7000 6000

0.4

5000 0.3

4000

0.2

Trustworthiness by Amount Sent Back on Secondary Axis

0.1

Trustworthiness by Percentage of Tripled Amount on Primary Axis

0

500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000 Trust, Amount of T1

3000 2000 1000 0

Note: t1 is on the X axis and t2 is on the Y axis, the percentage of the tripled amount in MNT on the primary Y axis and the amount returned by the second player out of the tripled amount on the secondary Y axis. The average of all subjects is plotted for any given t1. The actual amount is shown by the dotted line and the percentage returned is shown by the solid line. The solid horizontal line at 33.3% of the tripled amount is the boundary for trust to be profitable. Since all squares on the solid line for trustworthiness lie above the horizontal dotted line at 33%, it is clear that, on average, trust is always profitable.

26

FIGURE 2 Expected earnings for any given t1

Expected Earnings in MNT

12000 10000 8000 6000 4000 95% CI for Entrepreneurs 2000 0

95% CI for Entrepreneurs 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000

Note: t1(trust) is on the X axis and expected earnings in MNT are on the Y axis. The graph shows expected earnings at each t1 (amount sent), averaged over all subjects, assuming that the counterpart’s responses (t2 as function of t1) are identical to those of the sender.

27

FIGURE 3

0

10000

20000 30000 40000 50000

Correlation between trust and sales in 1000 MNTs (~1 USD)

0

2000 4000 trust measure, the amount sent as first mover in trust game Monthly Real World Sales in 1000MNT (~$1)

6000

Fitted values

Note: Figure 3 shows the correlation between trust and real-world business sales: The greater the trust, the higher the real-world business sales.

28

FIGURE 4

0

10000

20000 30000 40000 50000

Correlation between sales and absolute deviation from the optimal trust

0

2000 4000 deviation from optimal trust, average t2p used Monthly Real World Sales in 1000MNT (~$1)

6000 Fitted values

Notes: The figure shows the negative correlation between real-world business sales and the absolute value of deviation from the optimal level of trust: The more the subject deviates from the optimal trust level, the lower the sales.

29

TABLE 1.A Summary Statistics

Variables

Obs

Mean

Std. Dev.

Min

Max

sales: Monthly Sales in 1000 MNT t1: Trust Measure t2p: Average Percentage of T2 t2ave: Average Amount of T2 risk: Risk Measure 1-6

116 119 119 119 85

11891 3092 51.50 4913 2.75

8752 1733 21.21 2226 1.97

150 0 0 0 1

52516 6000 100 9750 6

ws: Wholesaler Dummy male: Gender Dummy age bus_duration: Length of Business loan: Amount of Loan from Banks bank_rate: Monthly Interest Rate

120 120 118 116 97 97

0.44 0.32 37 7.25 5635 0.78

0.50 0.47 10 4.82 11948. 1.09

0 0 19 1 0 0

1 1 67 24 55000 3.5

30

TABLE 1.B: Correlation between Variables

Sales Sales 1 t1: Trust 0.34* t2(%): Trustworthiness 0.06 Risk Aversion -0.13 Age 0.11 Male 0.25* Amount of Loan from Banks 0.16 Length of Business in Years 0.12 Wholesaler Dummy 0.32* Notes: * indicates 5% significance.

t1

t2(%)

Risk

Age

Male

Loan

Length

ws

1 0.39* -0.07 0.04 0.25* 0.16 0.15 0.11

1 0.06 0.18* 0.11 0.25* 0.17* 0.19*

1 -0.05 0.06 0.02 -0.15 -0.19

1 -0.04 0.18* 0.43* 0.23*

1 0.17* 0.11 0.08

1 0.43* 0.46*

1 0.29*

1

31

TABLE 2 Trust and Sales: OLS regressions VARIABLES t1: Trust (amount sent in MNT)

Monthly Sales in 1000MNT, May 2010 data used 1.71***

1.88***

1.91***

1.67***

1.64***

(0.44)

(0.48)

(0.48)

(0.48)

(0.47)

25.02

-36.11

-48.31

-51.96

-66.55*

(38.57)

(39.48)

(40.26)

(39.67)

(38.56)

103.2

115.2

70.88

(76.0)

(75.1)

(73.8)

3,618**

3,133*

(1,706)

(1,653)

t2(%): Trustworthiness (percentage returned) Age Male Wholesaler Dummy

4,687*** (1,543)

Constant

6,593*** 10,613*** 7,908***

4,671

4,006

4,577

(1,559)

(2,132)

(2,122)

(3,301)

(3,265)

(3,154)

Observations

116

116

116

114

114

114

R-squared

0.11

0.01

0.12

0.14

0.17

0.24

Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1. Risk-aversion measure uses the first measure of risk: the higher the risk , the more the subject loves risk.

32

TABLE 3 Correlation between Sales and Absolute Deviation from the Optimal Trust; Own Answers Used

VARIABLES Absolute Deviation from the Optimal Trust (Own answers used to calculate) Wholesaler dummy Male dummy Trustworthiness as percentage out of 100% Constant

Monthly Sales in 1000MNT (~$1) 1.31*** (0.48)

1.25*** (0.46)

1.09** (0.45)

5,371*** (1518)

5,051*** (1500) 3,560** (1618)

1.19** (0.47)

5,307*** (1529) 3,667** (1624) -33.10 (37.05) 15064*** 12,558*** 11,175*** 12,956*** (1,400) (1,511) (1,613) (2,565)

Observations 116 116 R-squared 0.06 0.16 Standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1

33

116 0.19

116 0.20

TABLE 4 Correlation between Sales and Absolute Deviation from the Optimal Trust; Average t2p Used

VARIABLES Absolute Deviation from the Optimal Trust (Average values used to calculate) Wholesaler dummy Male dummy Trustworthiness as percentage out of 100% Constant

Monthly Sales in 1000MNT (~$1) -1.75*** (0.49)

-1.53*** (0.48)

-1.31*** (0.48)

4,811*** (1517)

4,613*** (1502) 3,143* (1,633)

-1.57*** (0.52)

4,935*** (1513) 3,195* (1,627) -53.60 (38.26) 16,295*** 13,622*** 12,153*** 15,401*** (1,454) (1,633) (1,785) (2,921)

Observations 116 116 R-squared 0.10 0.17 Standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1

34

116 0.20

116 0.21

TABLE 5 Risk Measure and its Correlation with Other Variables

(1)

(3)

2.29***

2.25***

1.61***

(0.60)

(0.66)

(0.46)

Absolute Deviation from the Optimal Trust (Average values used to calculate) Wholesaler Dummy

Gender (Male=1)

Trustworthiness (in percentage)

(5)

(6)

-2.19***

-2.10***

-1.57***

(0.69)

(0.75)

(0.52)

2,541

2,562

4,830***

2,276

2,315

4,668***

(1,877)

(1,927)

(1,524)

(1,936)

(1,985)

(1,546)

3,130

3,149

2,703

3,265

3,316

2,888*

(1,940)

(2,010)

(1,648)

(1,994)

(2,063)

(1,666)

-97.90*

-94.71*

-63.45*

-81.64

-75.85

-57.71

(49.88)

(53.22)

(38.08)

(50.56)

(53.65)

(38.58)

Consistent in Risk Measure

Risk Measure

(4)

Monthly Sales in 1000MNT (~$1)

VARIABLES

Trust

(2)

1,402

1,497

(1,688)

(1,709)

-209.84

-227.98

35

(1 to 6, 1 being the most risk averse)

(475.09)

risk==1

risk==2

risk==4

risk==5

risk==6

Constant

Observations R-squared

(487.53) 2,175

2,880

(3,590)

(3,660)

758

1,583

(4,717)

(4,814)

1,836

1,999

(4,438)

(4,552)

239

792

(4,451)

(4,555)

1,426

2,207

(4,058)

(4,137)

8,829***

6,647

6,172***

20,736***

17,441***

14,762***

(3,019)

(4,486)

(2,133)

(4,100)

(5,952)

(3,014)

81

81

116

81

81

116

0.24

0.25

0.24

0.20

0.21

0.22

Note: The risk measure has 5 choices and a dummy is used for each choice: risk==1 is a dummy for subjects who chose the first risky asset (the most risk averse) etc. Standard errors in parentheses: *** p<0.01, ** p<0.05, * p<0.1

36

References:

Aldrich, H. E. (2000). “Entrepreneurial Strategies in New Organizational Populations.” Entrepreneurship: The Social Science View, 211-228. Anderson, A.R. and Jack, S.L. (2002). “The Articulation of Social Capital in Entrepreneurial Networks: A glue or a lubricant?” Entrepreneurship and Regional Development, 14, 193–210. Ayyagari, M., Beck, T., and Demirguc-Kunt, A. (2007). “Small and Medium Enterprises across the Globe.” Small Business Economics, 29(4), 415-434. Barr, A. M. (2003). “Trust and Expected Trustworthiness: Experimental Evidence from Zimbabwean Villages.” The Economic Journal, 113:614-30 Barr, A. and M. Lindelow (2005). “Strategy Choice and Cognitive Ability in Field Experiments.” University of Oxford, Department of Economics, Economics Series Working Papers: GPRG-WPS-034. Bardsley, N., R. Cubitt, G. Loomes, P. Mo_att, C. Starmer, and R. Sugden. 2009. “Experimental Economics: Rethinking the Rules.” Princeton University Press. Batsaikhan, M. (2014). “Growth for Micro-enterprises: Evidence from Phone Card Wholesalers in Mongolia.” Working Paper, Georgetown University Belot, M., Duch, R., and Miller, L. (2015). “A comprehensive comparison of students and non-students in classic experimental games.” Journal of Economic Behavior and Organization, 113, 26-33. Berg, J., J. Dickhaut; and K. McCabe (1995). “Trust, reciprocity, and social-history.” Games and Economic Behavior, 10:122–142. Bloom, N., F. Sadun, and J. V. Reenen (2012). “Organization of Firms Across Countries.” The Quarterly Journal of Economics, 127:1663-1705.

37

Bohnet, I., F. Greig, B. Herrmann, and R. Zeckhauser (2008). “Betrayal aversion: Evidence from Brazil, China, Oman, Switzerland, Turkey, and the United States.” American Economic Review, 98: 294–310. Brandts, J., and G. Charness (2011). "The strategy versus the direct-response method: a first survey of experimental comparisons." Experimental Economics, 14:375-98 Butler, J., P. Guiliano, and L. Guiso (2014). “The Right Amount of Trust.” NBER Working Paper No. 15344 Caliendo, M., Fossen, F., and Kritikos, A. (2012). “Trust, positive reciprocity, and negative reciprocity: Do these traits impact entrepreneurial dynamics?” Journal of Economic Psychology, 33(2), 394-409 Camerer, C. (2015). “The promise and success of lab-field generalizability in experimental economics: A critical reply to Levitt and List.” Handbook of Experimental Economic Methodology. G. Frechette and A. Schotter. Oxford, UK, Oxford University Press. CÁRDENAS, J. C., CHONG, A., ÑOPO, H., Horowitz, A. W., and Lederman, D. (2009). “To what extent do latin americans trust, reciprocate, and cooperate? evidence from experiments in six latin american countries [with comments]”. Economía, 9(2), 45-94 Carpenter, J. P., A. G. Daniere, and L. M. Takahashi (2004). "Cooperation, trust, and social capital in Southeast Asian urban slums." Journal of Economic Behavior & Organization, 55: 533-51 Cleave, B. L., Nikiforakis, N., and Slonim, R. (2013). Is there selection bias in laboratory experiments? The case of social and risk preferences. Experimental Economics, 16(3), 372-382.

Charness, G., U. Gneezy, and A. Imas (2013). "Experimental methods: Eliciting risk preferences." Journal of Economic Behavior & Organization, 87: 43– 51

38

De Mel, S., D. McKenzie, and C. Woodruff (2008). "Returns to Capital in Microenterprises: Evidence from a Field Experiment" The Quarterly Journal of Economics, 123: 1329-1372 Eckel, C. C. and R. K. Wilson (2004). "Is trust a risky decision?" Journal of Economic Behavior & Organization, 55: 447-65 Eckel,C. C. and R. K. Wilson (2010). “Trust and Social Exchange.” Cambridge Handbook of Experimental Political Science, James N. Druckman, Donald P. Green, James H. Kuklinski, Arthur Lupia, eds., Boston: Cambridge University Press Falk, A., Meier, S., and Zehnder, C. (2013). “Do lab experiments misrepresent social preferences? the case of self-selected student samples.” Journal of the European Economic Association, 11(4), 839-852 Fehr, E., and List, J. A. (2004). “The hidden costs and returns of incentives? Trust and trustworthiness among CEOs.” Journal of the European Economic Association, 2(5), 743-771 Galizzi, M. M. and D. N. Martinez (2015). “On the external validity of social-preference games: A systematic lab-field study.” Department of Economics and Business, Universitat Pompeu Fabra, Economics Working Papers Goel, S., and R. Karri. “Entrepreneurs, Effectual Logic, and overTrust.”Entrepreneurship Theory and Practice (2006): 477-93. Guiso, L., P. Sapienza. and L. Zingales (2004). "The Role of Social Capital in Financial Development" American Economic Review, 94: 526-56 Guiso, L., P. Sapienza. and L. Zingales (2009). “Cultural Biases in Economic Exchange?" The Quarterly Journal of Economics, 124: 1095-1131 Harrison, G. W. and J. A. List (2004). “Field Experiments.” Journal of Economic Literature, 42: 1009-55

39

Henrich, J., Heine, S. J., and Norenzayan, A. (2010). “The weirdest people in the world?.” Behavioral and brain sciences, 33(2-3), 61-83 Hoffman, M., and Morgan, J. (2015). “Who's naughty? who's nice? experiments on whether pro-social workers are selected out of cutthroat business environments.” Journal of Economic Behavior and Organization, 109, 173-187 Holt, C. A. and S. K. Laury (2002). “Risk Aversion and Incentive Effects.” American Economic Review, 92: 1644–55. Howorth, C. and Moro, A. (2006). “Trust within entrepreneur bank relationships: Insights from Italy.” Entrepreneurship Theory and Practice, 30(4), 495–517. Hsu, S. H., C. Y. Huang, AND C. T. Tang (2007). “Minimax Play at Wimbledon: Comment.” American Economic Review, 97(1), 517–523 Jackson, C. (2012). "Internal and External Validity in Experimental Games: A Social Reality Check." European Journal of Development Research 24(1), 71-88. Johnson, N. D. and A. Mislin (2011). “Trust Games: A Meta-Analysis.” Journal of Economic Psychology, 32: 865-89 Karlan, D. S. (2005). "Using Experimental Economics to Measure Social Capital and Predict Financial Decisions." American Economic Review, 95: 1688-99 Karlan D. S. and J. Morduch (2009). “Access to Finance” in Handbook of Development Economics, Vol 5, by Rodrik, D., and Rosenzweig, M., North Holland Publications. Keefer, P. and S. Knack (1997). “Does Social Capital Have an Economic Payoff? A Cross-Country Investigation." The Quarterly Journal of Economics, 112: 125188 Knack, S. and P. J. Zak (2001). "Trust and Growth." The Economic Journal, 111: 295– 321

40

Krawczyk, M., (2011). "What brings your subjects to the lab? A field experiment," Experimental Economics, 14(4), 482-489 Levitt, S. D., and List, J. A. (2007). “What do laboratory experiments measuring social preferences reveal about the real world?” The Journal of Economic Perspectives,21(2), 153-174. Levitt, S. D., J. A. List, and D. H. Reiley (2010). “What Happens in the Field Stays in the Field: Exploring Whether Professionals Play Minimax in Laboratory Experiments.” Econometrica, 78(4), 1413–1434 Liao, J. and Welsch, H. (2005). “Roles of social capital in venture creation: Key dimensions and research implications.” Journal of Small Business Management, 43(4), 345–362. List, J. A., and Cherry, T. L. (2008). “Examining the role of fairness in high stakes allocation decisions.” Journal of Economic Behavior and Organization, 65(1), 1-8. Ostrom, E. and J. Walker (2005). (eds) “Trust and Reciprocity: Interdisciplinary Lessons for Experimental Research.” Russel Sage Foundation Series on Trust, Political Science Palacios-Huerta, I. (2003). "Professionals Play Minimax." Review of Economic Studies, 70: 395-415. Palacios-Huerta, I. and O. Volij (2008). "Experientia Docet: Professionals Play Minimax in Laboratory Experiments." Econometrica, 76: 71-115 Stoop, J., Noussair, C. N., and Van Soest, D. (2012). “From the lab to the field: Cooperation among fishermen.” Journal of Political Economy, 120(6), 1027-1056. Sutter, M., and Kocher, M. G. (2007). “Trust and trustworthiness across different age groups.” Games and Economic Behavior, 59(2), 364-382.

41

Tabellini, G. (2008). “Culture and Institutions: Economic Development in the Regions of Europe.” Journal of the European Economic Association, 8: 677-716 Welter, F., and Smallbone, D. (2006). “Exploring the role of trust in entrepreneurial activity.” Entrepreneurship Theory and Practice, 30(4), 465-475. Zahra, S., Yavuz, R.I., and Ucbasaran, D. (2006). “How much do you trust me? The dark side of relational trust in new business creation in established companies.” Entrepreneurship Theory and Practice, 30(4), 541–559.

Appendix material 1: Instructions for Trust game (Back-translated) Your first decisions in today’s experiment will determine part of your earnings. From this first portion, you can earn at least 6000MNT, but depending on your decision and the decision of another participant with whom you will be paired for this part, you might earn less or more than that (anything between 0 and 24000). You will indicate your choices for this portion by circling numbers on a form. In all, you must circle one number on each of 14 different lines. First, we will explain the general idea of this part of the experiment. You and the participant with whom you are paired will each begin this portion being credited with money in the amount of 6000MNT. One of you will end up being assigned to the role of first decision maker, or Participant A. The other will be second decision maker, or Participant B. A makes only one decision: whether to send to B some, all, or none of A’s 6000MNT. If A sends some amount of money to B, the experimenter triples that amount of money, causing B to receive three times what A sent. For example, if A decides to send 2000MNT, B receives 6000MNT. If A decides to send 5000MNT, B receives 15000MNT. B then decides how much of the money received B sends back to A. B can keep the entire amount, return the entire amount, or send back some of the amount received. Amounts that B sends back to A are not tripled, so if B sends back to A, say 2000MNT, A receives 2000MNT. Given the decisions made by A and B, A will earn from this portion of the experiment any part of A’s initial 6000MNT that A does not send to B plus any part of the tripled amount that B sends back to A. B will earn B’s initial 6000MNT plus the tripled amount, if anything, sent by A, minus any part of the tripled amount that B chooses to send back to A. Each individual will engage in this kind of interaction only one time. The money that you earn from this interaction will be added to other money you earn today and paid to you in cash at the end of the experiment. The results of this first portion of today’s experiment will not be announced to you until all parts of the experiment are over. (Even then, you will never learn the identity of the participant with whom you were paired.) Therefore, an individual in the B role does not know in advance how much (if anything) their counterpart in the A role has decided to send. Instead, each individual makes a decision for every possible amount that A might send. B decides how much, if anything, to send back to A if A sends B 500MNT (and B receives 1500MNT); how much, if anything, to send back to B if A sends B 1000MNT (and B receives

42

3000MNT); and so on. (If A sends 0, there is no real decision to be made by B.) Which decision of B’s actually ends up determining B’s earnings depends on what participant A he is randomly paired with and on what that participant A has decided to do. Whether you are going to be in role A or in role B is going to be determined randomly after you fill out the decision form. Therefore, every participant is asked to complete the form in its entirety, without knowing whether you are in role A or role B. You are to begin by selecting an amount (if anything) you want to send to your counterpart B if you end up being in role A. Then you will select an amount (if anything) you want to send back to your counterpart A for each possible amount that A might send if you end up being in role B. Before you begin, please look at the decision sheet. Then raise your hand if you have any questions, and I will come to you. Do not begin until you are sure that you understand your task.

Decision Sheet for 1st Portion of Experiment

Instructions: For each numbered line (except 2.), circle clearly only one of the possible items.

1.

If I am in role A, I wish to send to my counterpart in role B (circle one number)

0

500 1000 1500 2000 2500 3000 3500 4000 4500 50000 5500 6000

2.

(If I am in role B and my counterpart A sends me 0, I have no decision to make.)

3. If I am in role B and my counterpart A sends me 500 (I receive 1500), I choose to return to A (circle one number)

0

500 1000 1500

4. If I am in role B and my counterpart A sends me 1000 (I receive 3000), I choose to return to A (circle one number)

0

500 1000 1500 2000 2500 3000

43

5. If I am in role B and my counterpart A sends me 1500 (I receive 4500), I choose to return to A (circle one number)

0

500 1000 1500 2000 2500 3000 3500 4000 4500

6. If I am in role B and my counterpart A sends me 2000 (I receive 6000), I choose to return to A (circle one number)

0

500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000

7. If I am in role B and my counterpart A sends me 2500 (I receive 7500), I choose to return to A (circle one number)

0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000 6500 7000 7500

8. If I am in role B and my counterpart A sends me 3000 (I receive 9000), I choose to return to A (circle one number) 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000 6500 7000 7500 8000 8500 9000 9. If I am in role B and my counterpart A sends me 3500 (I receive 10500), I choose to return to A (circle one number) 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000 6500 7000 7500 8000 8500 9000 9500 10000 10500

10. If I am in role B and my counterpart A sends me 4000 (I receive 12000), I choose to return to A (circle one number)

0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000 6500 7000 7500 8000 8500 9000 9500 10000 10500 11000 11500 12000

11. If I am in role B and my counterpart A sends me 4500 (I receive 13500), I choose to return to A (circle one number)

44

0

500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000 6500 7000 7500 8000 8500 9000 9500 10000 10500 11000 11500 12000 12500 13000 13500

12. If I am in role B and my counterpart A sends me 5000 (I receive 15000), I choose to return to A (circle one number)

0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000 6500 7000 7500 8000 8500 9000 9500 10000 10500 11000 11500 12000 12500 13000 13500 14000 14500 15000

13. If I am in role B and my counterpart A sends me 5500 (I receive 16500), I choose to return to A (circle one number)

0

500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000 6500 7000 7500 8000 8500 9000 9500 10000 10500 11000 11500 12000 12500 13000 13500 14000 14500 15000 15500 16000 16500 14. If I am in role B and my counterpart A sends me 6000 (I receive 18000), I choose to return to A (circle one number) 1 2

0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000 6500 7000 7500 8000 8500 9000 9500 10000 10500 11000 11500 12000 12500 13000 13500 14000 14500 15000 15500 16000 16500 17000 17500 18000

45