ENTERPRISE S URVEYS ENTERPRISE NOTE S ERIES CONTRACTS Navigating the Landscape of Transactions: Understanding Firms’ Transactional Governance Structures 2019 in Six Latin American Countries David C. Francis T his note1 summarizes ndings from a new paper, which lays out a methodology to determine the types of mechanisms—be they the legal system, civil organizations, or shared business interest—that rms use when entering into agreements with their suppliers and customers. Speci cally, the note summarizes the data exploration technique (known as Latent Class Analysis) used ENTERPRISE NOTE No . 38 in the paper and shows some basic results in terms of the attributes of each class. Four basic classes are used for dealing with suppliers (pure bilateralism, bilateralism with private support, bilateralism with legal support, and strong comprehensive) and with customers (pure bilateralism, bilateralism with private support, bilateralism with weak support, and weak comprehensive). Lastly, some basic correlations between the likelihood of class membership and rm-level characteristics are shown. Introduction of the “relative, aggregate importance of each mechanism” in the speci c case of Romania, much more is to be done Forty years ago, Oliver Williamson (1979) pioneered by extending such a framework systematically to other the idea that since it is costly (in terms of time, money, countries, markets, and circumstances. and information) for rms to enter into agreements, those “contracts” are almost always left as incomplete. e A recent paper2 lays out an extension of this details are left to be hashed out between parties using the approach—attempting to classify the relative importance institutions available to them—whether they be of these mechanisms when employed together—by government intervention, the legal system, or the analyzing representative survey data for six Latin intervention of other third parties (such as business American countries across these six mechanisms, for both associations, friends, or goons). is theory of the agreements with suppliers and customers. e paper also ‘governance of contractual relations’ has proven to be a uses a technique that is somewhat novel to the economics powerful framework for understanding the types of literature (though common in other social sciences), agreements rms enter into (and under what particular known as Latent Class Analysis or LCA. is is a statistical WORLD BANK GROUP circumstances). In fact, the work earned Williamson the technique that does not require an a priori theory of the Nobel Prize in Economics in 2009. expected, relative frequency of di erent mechanism types. However, while quite a bit of work has been done to is is both fortunate and necessary: fortunate because explain the speci c mechanisms used for entering and such a comprehensive theory has not been clari ed in the operating within agreements, there is still little, literature and necessary because making sense of the comprehensive understanding of the relative frequency myriad combinations and intensities that rms use to with which rms use those same mechanisms. An enter agreements involves parsing a daunting set of data. exception is Hendley and Murrell (2003), who usefully e result of such analysis is a systematic classi cation of classify the principle mechanisms rms use to make so-called transactional Governance Structures3 that enforceable agreements: in brief, these are 1) personal underlie rms’ relations with both their suppliers and relationships and (perceived) shared trust; 2) mutual customers. interest in an ongoing business relationship; 3) paid, ese results are meant to be informative, but also, to private dispute resolution; 4) assistance from government act as a cartographer’s toolkit for further research applying o cials; 5) the intervention of other third parties; and 6) similar questions in various countries and contexts. the legal system. While their work features an application Indeed, the paper nds that all Governance Structures (for relations with both suppliers and customers) build upon much more space in the accompanying paper is dedicated bilateral mechanisms; and while bilateralism is sometimes to explaining LCA. complemented using more formal institutions, it is never To do this analysis, the paper uses data from a unique supplanted by those mechanisms. e prevalence of set of questions included in the 2017/8 World Bank contractual Governance Structures varies by rm Enterprise Surveys (ES) in six South American countries: characteristics, but these Governance Structures tend to Argentina, Bolivia, Ecuador, Paraguay, Peru, and vary more within countries than between them. Uruguay.7 Data was gathered for over 3,300 observations.8 ese surveys are based on face-to-face From Individual Mechanisms to Governance interviews with top managers and business owners, among Structures o cially registered rms with at least ve employees.9 e data from these surveys are based on a complex, strati ed De ning the exact mechanisms that rms use when survey design, and taking these aspects into account, can entering and operating within agreements is a di cult be considered as nationally representative.10 Speci cally, task, but one that has fortunately been tackled somewhat the questions asked to what degree each of the six recently in the literature. As noted above, these attempts mechanisms was regarded as e ective when making have generally settled on a comprehensive list of six agreements; the set of questions was asked both for mechanisms.4 While many past studies have applied each suppliers, then for customers.11 Respondents were asked of these mechanisms individually to frame speci c to rate e ectiveness on a ve-point Likert scale from “Not transactions, there is little understanding of the at all” to “Extremely”. On a cursory glance, the usefulness coordinated combination of these mechanisms. A of LCA is apparent: with six questions (each with ve formalized de nition of this combination, that is a possible responses), a total of 15,625 possible response (transactional) Governance Structure, is an “institutional patterns (i.e., speci c combinations of answers) is framework within which the integrity of a transaction is possible. ough a number far lower is observed in the decided.”5 Yet while the use of individual mechanisms has data (711 for suppliers, 631 for customers), without a tool merited substantial discussion, no systematic e orts to such as LCA, one would need to anticipate what possible discern these relative patterns (and in a way consistent Governance Structures and groupings would emerge by across countries) exists. e lack of an understanding of coding or classifying each of these observed patterns. these patterns means that researchers lack the tools to draw any conclusions on whether rms in La Paz, Bolivia, Discovery of Classes: Description of Governance rely more or less on a combination of law and bilateral Structures trust than rms in La Paz, Argentina, or, indeed, La Paz, Arizona. While valuable on its face, this exercise can also As mentioned above, the aim of LCA is to discover shed light on current points of contention: for instance, if classes—here, Governance Structures—as well as to personal relationships and trust are complemented by or estimate the prevalence of rms across these classes (which substituted by legal remedies. will be covered in the next section). Since this discovery is itself a probabilistic exercise, it allows researchers to Data and Analysis express a degree of uncertainty. In fact, the chosen approach rst estimates the likelihood of observing each e lack of a clearly de ned theoretical framework set of question responses given a class membership.12 means that there are not readily available and testable en, each rm can be assigned a probability of speci c hypotheses about the expectations of the relative class membership; and nally, these probabilities of class importance of the mechanisms that govern rms’ membership can be evaluated against speci c rm transactions with suppliers and customers. is absence, characteristics. To give a speci c example: suppose one though, does not render the task impossible; in fact, the observes a rm that says trust and mutual interests are task lends itself to more exploratory statistical techniques. “very much e ective” and all other mechanisms are “not A Governance Structure itself cannot be observed (one at all” e ective.13 LCA rst estimates what is the can only observe the stated intensity with which each rm likelihood of observing this response pattern given considers each mechanism as e ective) and so one must membership in a certain class. To preview later sections: assume those observations are indicative of an underlying the likelihood of observing this set of responses given (or latent) characteristic. As its name implies, then, LCA is membership in a class that is mostly bilateral would be suited for this sort of task. LCA allows researchers to rst comparatively high. In turn, LCA reveals that the identify the ‘classes’ (i.e. Governance Structures) and, likelihood that a rm giving these responses of being in then, to also determine the estimated prevalence of those that class would also be high. Lastly, this likelihood can be classes. Moreover, LCA is well-suited for large amounts of correlated with the characteristics of rms that are more (otherwise nearly intractable) information. However, likely to be in that class. while used fairly widely in other social science disciplines, LCA’s discovery of Governance Structures has the LCA is not deeply rooted in the economics literature6: advantage of not requiring an a priori model of the relative 2 Figure 1 Predicted Response Probabilities (Suppliers) Pure Bilateralism Bilateralism with Private Support 100% 100% 80% 80% 60% 60% 40% 40% 20% 20% 0% 0% Trust Mutual Priv. dispute Gov’t officials Other 3rd Legal system Trust Mutual Priv. dispute Gov’t officials Other 3rd Legal system interests resolution party interests resolution party Bilateralism with Legal Support Strong Comprehensive 100% 100% 80% 80% 60% 60% 40% 40% 20% 20% 0% 0% Trust Mutual Priv. dispute Gov’t officials Other 3rd Legal system Trust Mutual Priv. dispute Gov’t officials Other 3rd Legal system interests resolution party interests resolution party Extremely Very much Moderately Slightly Not at all importance of each mechanism and how these t stands out immediately is how pervasive these bilateral together. However, this does not mean that researchers mechanisms are: they are nearly ever-present and are using LCA are not required to make choices. e paper generally complemented by (and not substituted for) lays out several of the key assumptions applied, including other mechanisms. Indeed, while there was a long-held various tests of those assumptions. Two points merit view that well-heeled legal mechanisms may cause rms to mention here. First, the number—in this case four—of substitute away from bilateral ones, this does not appear classes itself is a choice.14 LCA allows researchers to choose to be the case.16 between imposing any number of classes. ese classes a. Pure Bilateralism (B): considering this nding, the should provide meaningful insights, such that adding (or rst class is the most self-evident, and is characterized removing) a class does not obscure the usefulness of the by a near-exclusive reliance on only bilateral exercise.15 Four classes are presented as this number allows mechanisms. Figure 1, panel a. shows the estimated for a more digestible classi cation, which provides a probabilities for this class. generally well- t model. Second, the paper’s application of LCA imposes no further assumptions based on b. Bilateralism with Private Support (BP): a second underlying, rm characteristics. is includes, for governance structure class can be described as example, the country where a rm is located. at is, the bilateralism, but with paid private dispute resolution underlying Governance Structures are assumed to be followed by the legal system as the supplementary common (an assumption also inherent in any pooled mechanisms. is is consistent with how paid private regression analysis); what can vary are rms’ likelihood of third parties often work in practice. Arbitration being in each class, contingent on, say, country or other mechanisms always need the backing of formal legal characteristics. enforcement; the job of goons is often simply to forcefully remind miscreants of the possibility of legal Governance Structures - Suppliers sanctions. For the patterns attributable to each class to be well-understood, it is helpful if they are also well-named. c. Bilateralism with Legal Support (BL): a third class of In looking at relations with suppliers, one stark nding is governance structures is best called bilateralism with that all classes rely heavily on both bilateral trust and legal support when dealing with suppliers. is class is mutual interest (together these two mechanisms as termed denoted by a somewhat weaker reliance on bilateralism here as “bilateral”). Governance Structures di er mostly and a stronger regard for the legal system (though with on the dimensions in how they elaborate beyond this some complementarity with paid, private dispute bilateralism, if at all. To preview the discussion, what resolution). 3 Figure 2 Predicted Response Probabilities (Customers) Pure Bilateralism Bilateralism with Private Support 100% 100% 80% 80% 60% 60% 40% 40% 20% 20% 0% 0% Trust Mutual Priv. dispute Gov’t offcials Other 3rd Legal system Trust Mutual Priv. dispute Gov’t offcials Other 3rd Legal system interests resolution party interests resolution party Bilateralism with Weak Support Weak Comprehensive 100% 100% 80% 80% 60% 60% 40% 40% 20% 20% 0% 0% Trust Mutual Priv. dispute Gov’t offcials Other 3rd Legal system Trust Mutual Priv. dispute Gov’t offcials Other 3rd Legal system interests resolution party interests resolution party Extremely Very much Moderately Slightly Not at all d. Strong Comprehensive (SC): e last governance reported e ectiveness of mechanisms other than trust structure on the suppliers’ side is best described as and mutual interest. comprehensive (with a degree of higher e ectiveness of all mechanisms), and strongly so. Prevalence of Classes: Which Firms are More Likely to be in which Classes Governance Structures – Customers Of course, beyond the discovery of what classes of Regarding relationships with customers, the observed Governance Structures best describe patterns of how rms patterns can be similarly described as those governing how are operating, it is useful to see what the relative rms deal with suppliers. In fact, these patterns are so similar prevalence of those classes is as well. e probabilities of for the rst two classes (a. and b.) that they are likewise class membership are shown in Figure 3, together with named “pure bilateralism” and “bilateralism with private standard errors. What is immediately notable is the support”, respectively. e third and fourth classes di er pervasiveness of bilateralism. An estimated two-thirds notably, however, and these are, in turn, given di erent proportion of rms use a governance structure of almost names than for those dealing with suppliers (Figure 2). exclusively trust and mutual interest in their agreements Speci cally, these two classes are: with supplier as well as with customers. c. Bilateralism with Weak Support (BW): the third class on dealing with customers has some regard for Figure 3 Probabilities of Class Membership bilateralism, but to a lesser degree than other classes, and only uses other, additional mechanisms to a Standard errors in brackets lukewarm degree. In fact, among all eight governance 100% 2% [0.4%] 5% [0.8%] structures, this class has the lowest rated aggregate 16% [1.7%] 15% [2.5%] e ect of the 6 mechanisms. One might alternatively 75% 17% [2.2%] 24% [0.0%] call this class ‘ine ective governance’ as it re ects a 66% [2.4%] general lack of importance attributed to all 50% 57% [3.2%] mechanisms by the rms in this class. 25% d. Weak Comprehensive (WC): lastly on making agreements with customers, the fourth class is also 0% Suppliers Customers characterized by its comprehensiveness, but this is B BP BL SC B BP BW WC denoted as ‘weak’ due to the lesser intensity of the 4 Table 1 Predicted Class Probabilities, by Country Suppliers Argentina Bolivia Ecuador Paraguay Peru Uruguay Pure Bilateralism (B) 67% 52% 65% 71% 73% 61% Bilateralism, Private Support (BP) 19% 24% 12% 13% 12% 15% Bilateralism, Legal Support (BL) 12% 22% 23% 15% 15% 20% Strong Comprehensive (SC) 2% 2% 1% 1% 0% 4% Customers Argentina Bolivia Ecuador Paraguay Peru Uruguay Pure Bilateralism (B) 62% 44% 44% 48% 62% 69% Bilateralism, Private Support (BP) 19% 35% 33% 22% 29% 22% Bilateralism, Weak Support (BW) 14% 9% 21% 22% 6% 5% Weak Comprehensive (WC) 5% 12% 2% 9% 3% 3% It is also worth looking at how class probabilities vary of Pure Bilateralism); in Piura, Peru, this shift is one across and within countries (Table 1). Bolivia (though the toward support from private mechanisms (i.e. Bilateralism least developed of the six), is the country where rms are with Private Support).18 Again, these comparisons are the most likely to have governance structures that involve only cursory and have not been thoroughly tested; mechanisms beyond pure bilateralism; this seems in part however, the marked di erences within countries is novel due to a higher tendency for governance structures that and something previously under-explored by researchers. also engage private support. is pattern holds for both e ES data allow one to analyze patterns over other agreements with suppliers and with customers. While intriguing variables.19 For example, aligning with some some patterns may be somewhat surprising, such intuitive sense, both foreign-owned and exporting rms relationships are at best speculative and invite more (Figure 5) tend to use governance structures that use investigation. mechanisms beyond bilateralism (this, both in dealing with e richness of the data allows one to analyze whether suppliers and customers). In each, this shift from nearly some of these patterns can be parsed by looking at exclusively bilateral mechanisms is generally toward private variation within each country, for instance, by location. support and sometimes legal support. Similarly, large (100-399 For instance, the analysis reveals fewer, stand-out employees), and particularly, very large (400+ employees) rms di erences between countries, but yet there are marked are less likely to use purely bilateral structures. Better managed di erences within countries. Take, for example, the rms (Figure 6), also show a similar pattern, either when supplier-side class of Pure Bilateralism. In Argentina, dealing with suppliers or customers. Ecuador, Paraguay, and Peru17 there are quite sizable di erences between the cities or regions with the lowest Estimated Share of Figure 4 “Pure Bilateralism”, Suppliers and highest estimated shares of “Pure Bilateralism” governance for making agreements with suppliers. Figure 100% 4 shows the regions in each of these countries with both the lowest and highest estimated probability of Pure 75% Bilateralism in dealing with suppliers (only those countries where location is signi cant are shown; in 50% Bolivia and Uruguay, intra-country, location di erences are not signi cant). It is worth noting that in these 25% particular regions, other class membership proportions 0% may be informative. For example, in Argentina, Ecuador, Cordoba Rosario Guayas Azuay Asuncion Central Piura Chiclayo Argentina Ecuador Paraguay Peru and Paraguay, the region with the lowest probability of purely bilateral governance (Cordoba, Guayas, Asuncion, respectively), seems to turn toward legal support beyond Note: Horizontal bars show country averages. Only countries where sub-national (regional) differences are significant are shown. That is, these regional differences are bilateralism (relative to the region with the highest share not statistically significant in Bolivia, nor in Uruguay. 5 Figure 5 Foreign Ownership, Export Status, and Governance Structures Suppliers Customers Suppliers Customers 100% 100% 75% 75% 50% 50% 25% 25% 0% 0% Domestic owned Foreign owned Domestic owned Foreign owned Non exporter Exporter Non exporter Exporter B BP BL SC B BW BP WC B BP BL SC B BW BP WC Figure 6 Management Practices and Governance Structures Suppliers Customers Probabiilty of Class Membership Probabiilty of Class Membership 100% 100% 80% 80% 60% 60% 40% 40% 20% 20% 0% 0% 0.0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1 Management Practices Index Management Practices Index Bilateralism, legal support Bilateralism Bilateralism Bilateralism, weak support Bilateralism, private support Strong comprehensive Bilateralism, private support Weak comprehensive Conclusion: Discussion and Policy Consid- alter their supply chains and target new customers. at is, erations the mere chance for rms to take these sorts of actions may not be enough; rms almost certainly require a clear is note summarizes the main results of a formal knowledge or reliable mechanisms before entering into paper that provides a systematic way to classify and measure the prevalence of di erent Governance these sorts of contracts and agreements. e preliminary Structures for how rms regard agreements with suppliers ndings are suggestive of these patterns, but researchers may and customers. is research addresses a topic that has not want to test their own hypotheses using these presented been formally studied before due to the lack of data, results or they can similarly use an adapted version of an which has hindered the literature from providing a LCA method. It is worth noting that in the particular way comprehensive, cross-country analysis of how rms use LCA was used in the WB Working Paper, each possible these mechanisms together. e use of LCA, presented in response pattern is given a posterior probability of detail, will also be valuable to economists and other membership in each class. In other words, the only thing researchers wishing to apply such exploratory techniques needed to calculate the likelihood of class membership to large and rather complex data. would be responses to the survey questions used in the ES. e ndings themselves are useful: the prevalence of ese probabilities can then be easily estimated by anyone bilateralism, for instance, may have implications for rms gathering such data and can be extended to later data wishing to break into new markets—including foreign collection. For those interested, the data are available via rms wishing to operate in these domestic markets—or to https://www.enterprisesurveys.org/en/survey-datasets. 6 Notes correlations does seem advisable. e preferred model choice for suppliers was a four-class model with correlation structure 1 is work bene tted from nancial support provided by the 1-2,2-3,3-4,3-5,4-5. For customers the preferred model is a Research and Development Center in Chile - Development four-class 1-2,4-5,4-6,5-6 model. Economics Vice Presidency. 16 For a long time, the dominant view in the literature was that 2 Francis, D., N. Karalashvili, and P. Murrell, 2018. “Mapping the formal legal arrangements for transactions were inconsistent with Landscape of Transactions: e Governance of Business Relations personalized relationships based on trust: the formality eroded the in Latin America.” World Bank Policy Research Working Paper, trust. But this view has been moderated somewhat especially after No. 8564. is Enterprise Note is a summary work by one of the Poppo and Zenger's (2002) seminal contribution. Our results are authors. consistent with the changing view but are based on a broader 3 e paper uses the term Governance Structures (a slightly more overview of existing governance structures than any current succinct term for Transactional Governance Structure), as this is in contribution to the literature. line with the literature on the topic and will be familiar to some 17 In these four countries sub-national, regional di erences were readers. It should not be confused with corporate governance, found to be signi cant. Here, and in the paper, only statistically however, which can be summarized as the (often internal) rules signi cant di erences are presented. that rms use to conduct their operations, including corporate 18 ese di erences are the greatest pair-wise supplier-side class structures, bylaws, etc. Governance Structures are certainly not di erence after Pure Bilateralism (suppliers). at is, after noting wholly separate from corporate governance, but such structures that the largest di erence in Pure Bilateralism in Argentina is also include informal or unwritten interactions as well. To between Cordoba and Rosario, the next largest di erence is in distinguish the two terms, Governance Structure is capitalized predicted membership in Bilateralism with Legal Support. throughout this note. 19 To avoid a shing expedition, the paper also addresses the multiple 4 Hendley and Murrell, 2003 comparisons problem of looking at a wide number of co-variates to 5 Francis, Karalashvili, and Murrell, 2018 nd the signi cant ones. e paper addresses this, including by 6 Mike and Kiss (2018) are an exception, as their recent paper shows applying such criteria as family-wise error rates or false discovery LCA is well-suited for addressing the issue of classifying rates. e results presented here and, in the paper, hold up to these governance structures. criteria. 7 Data collection e orts bene ted from generous funding from the World Bank Group’s Knowledge for Change (KCP) program. Due to funding restrictions, data collection was only possible in these References six countries. A survey with similar, but modi ed and, thus, incomparable questions, was implemented in Colombia. Alesina, A., Devleeschauwer, A., Easterly, W., Kurlat, S. and Wacziarg, 8 Precisely, the number of observations was n=3,350 for relations R., 2003. Fractionalization. Journal of Economic Growth, 8(2), with suppliers and n=3,339 for relations with customers. Numbers pp.155-194. di er due to item non-response, which was minimal. For a more Bakk, Zsuzsa, Oberski, Daniel L., and Vermunt, Jeroen K. 2014. complete explanation of treatment, please refer to the main paper. Relating Latent Class Assignments to External Variables: Standard 9 For more information on the ES data collection and methodology, Errors for Correct Inference. Political Analysis, 22, 520-540. see: www.enterprisesurveys.org. Francis, D., N. Karalashvili, and P. Murrell, “Mapping the Landscape 10 is aspect sets these data apart from previous e orts. of Transactions: e Governance of Business Relations in Latin 11 e paper features the exact English wording of the questions. America.” World Bank Policy Research Working Paper, No. 8564. 12 Researchers employing LCA have generally used either what is Hendley, Kathryn, and Murrell, Peter. 2003. Which mechanisms called a three-step or one-step method, our paper using the former. support the ful llment of sales agreements? Asking decision-makers In the three-step method, one rst calculates maximum likelihood in rms. Economics Letters, 78(1), pp 49-54 estimates of observing response patterns given class membership; Latin American Public Opinion Project. 2016/7. “ e Americas then posterior probabilities of class membership (based on Barometer.” Latin American Public Opinion Project (LAPOP), response patterns) are determined; and lastly these probabilities can Vanderbilt University. http://www.LapopSurveys.org (accessed be correlated with co-variates, such as rm characteristics. See June 12, 2018). Vermunt (2010) or Bakk et al (2014). Mike, Károly and Kiss, Gábor. 2018. "Combining formal and informal 13 is is, in fact, the most frequently occurring pattern vis-à-vis contract enforcement in a developed legal system. A latent class suppliers, accounting for 12% of responses. approach.", Journal of Institutional Economics, forthcoming. 14 e number of classes was arrived at through a series of exercises Poppo, L. and Zenger T. 2002. “Do formal contracts and relational and criteria laid out in the paper, which also includes testing over governance function as substitutes or complements?”, Strategic various numbers of classes. Management Journal, 23(8): 707-725. 15 LCA relies upon an assumption that the responses to any two (or Williamson, Oliver E. 1979. "Transaction-Cost Economics: e more questions) themselves are un-related, this being known as Governance of Contractual Relations" e Journal of Law & “local independence”. However, allowing certain responses to be Economics, 22(2), Oct.: 233-261. correlated (for instance between trust and mutual), may improve Vermunt, Jeroen K., 2010. Latent class modeling with covariates: Two the t of these models. e adjustment of these assumptions is improved three-step approaches. Political Analysis, 18(4), explored and explained fully in the paper; indeed, allowing some pp.450-469. e Enterprise Note Series presents short research reports to encourage the exchange of ideas on business environment issues. e notes present evidence on the relationship between government policies and the ability of businesses to create wealth. e notes carry the names of the authors and should be cited accordingly. e ndings, interpretations, and conclusions expressed in this note are entirely those of the authors. ey do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its a liated organizations, or those of the Executive Directors of the World Bank or the governments they represent. 7