TY - JOUR
AU - Sutherland, Andrew
AB - Abstract We study how small and medium enterprise (SME) lenders react to information about their competitors’ contracting decisions. To isolate this learning from lenders’ common reactions to unobserved shocks to fundamentals, we exploit the staggered entry of lenders into an information-sharing platform. Upon entering, lenders adjust their contract terms toward what others offer. This reaction is mediated by the distribution of market shares: lenders with higher shares or that operate in concentrated markets react less. Thus, contract terms are shaped not only by borrower or lender fundamentals but also by the interaction between information availability and competition. Credit markets are characterized by dispersed information. Lenders do not have full information about their counterparties or their competitors’ actions. Strategic and information considerations are thus linked: lenders’ contracts depend on their information about competitors’ actions. Recent advances in information technology have attracted considerable attention from academics and policymakers concerned with its effects on competition.1 In credit markets, information technology has been studied primarily through the lens of learning about borrowers through the revelation of their credit records or the collection of soft information. However, there is also increased scope for learning about competitors, which introduces new issues related to competition, opacity, and the distribution of loan terms. Conceptually, the implications of lenders learning about their competitors are largely unresolved (Vives 2006). Existing theoretical models imply a wealth of empirical predictions that have considerable disagreement over channels, magnitudes, and even the sign of the effects. With imperfect competition, lenders can either mimic rivals if there are strategic complementarities or differentiate themselves through product choice (Shaked and Sutton 1982). There is also a role for information aggregation, in which rivals’ actions partially reveal their private information. Moreover, recent work has shown that the link between information and market outcomes is more complex than previously thought (Murfin and Pratt 2018; Goldstein and Yang 2019). Questions related to information and imperfect competition are notoriously difficult to study empirically. Indeed, the challenge in estimating the effect of learning about competitors is how to isolate variation in agents’ information sets. Specifically, lenders might offer similar terms not because they respond to each other but simply because they respond to the same economic shock. Our paper addresses this challenge by exploiting a unique setting that permits us to observe a direct shift in information that lenders have about rivals. Specifically, we use microlending data around the introduction of a commercial credit information-sharing platform, PayNet, which covers small and medium enterprises (SMEs) in the United States. PayNet launched in 2001; since then it has attracted eight of the 10 largest lenders in the market, a group that includes Bank of America, Wells Fargo, PNC, John Deere, IBM, Volvo, and Caterpillar. The platform provides information on contract terms offered by other lenders that was previously not widely available. We exploit the staggered entry of lenders into the platform to estimate the response to competitors and find that lenders adjust their terms toward what others are offering. Imperfect competition is a key driver of this finding: lenders in the most concentrated markets respond least to others’ offers. Thus, our evidence is most consistent with lenders learning about what it takes to compete as opposed to learning about fundamentals. Finally, we investigate an important consequence of our findings: matching competitors tends to increase delinquencies during the recent crisis, possibly because of the neglect of future risk. We document this evidence in the context of maturity dynamics for SMEs’ equipment financing contracts from 2001 to 2014. With over |${\$}$|1 trillion of annual volume, equipment financing is a major component of corporate investment, particularly for SMEs. Maturity cycles and rollover risk became a concern during the recent crisis and recovery because of their implications for firms’ liquidity and investments. The Survey of Terms of Business Lending shows that maturity on loans lasting over a year fell by 30% between 2007 and 2010, and Kalemli-Ozcan, Laeven, and Moreno (2018) document the dramatic effect of rollover risk on firm investment. Moreover, in our context of financing a specific piece of equipment, it is natural to focus on maturity as it is negotiable, while there is little variation in payment frequency or contract type to study. And by design, interest rates are not shared in the platform, just as they are typically not shared in consumer credit bureaus. Finally, there is evidence consistent with oligopolistic competition in this market (Murfin and Pratt 2019). Our empirical strategy is derived from a simple model of dispersed information and is designed to address the key difficulties associated with estimating the effect of learning about competitors. Specifically, two lenders can offer similar contracts not because they react to what the other is offering but simply because they react to the same shock to fundamentals. This is a crucial issue because it is plausible that at least some of these fundamentals cannot be observed by the econometrician and therefore cannot be controlled for. To address this, we rely on two features of our setting. First, we exploit lenders joining the platform in a staggered fashion to generate variation in information sets within and across lenders over time. Second, for each lender-borrower relationship, we observe contracts made before and after the lender joins the platform.2 Using these features, we study how closely a lender’s contract terms track the bureau average before versus after that lender joins and observes competitor data. Our empirical tests do not take a stand on the direction of the response. The key idea is that, while a lender’s terms may track the bureau average before joining, whether they track it relatively better or worse afterward reveals the sign of the response. For each contract, we model the gap between its maturity and the bureau’s average maturity for similar contracts as a function of whether the lender is a bureau member, contract size, borrower risk, and contract type, as well as lender-borrower relationship and collateral type-year fixed effects. We show that the gap shrinks by 7% after the lender joins the bureau. Lenders’ terms therefore track the bureau average relatively better after joining, consistent with a partial matching of rivals. Economically, this average effect corresponds to a 10% probability of a six-month or larger change in contract maturity. Such changes in maturity are important in their own right considering that in our sample, 18% of borrowers experience a change in their delinquency status over the next six months, substantially affecting rollover risk. Additionally, for a fully amortized loan with median characteristics, our main result is comparable to a 2 percentage point change in APR. Interestingly, the effect is symmetric: sometimes lenders match rivals by increasing maturity, sometimes by shortening it. Finally, we find similar evidence of convergence to competitors when we study contract size instead of maturity, or if we base our empirical strategy on proxies for lenders’ private information. Further evidence suggests that imperfect competition is a key driver of these findings, as opposed to more conventional channels of learning about fundamentals. Our results are strongly mediated by the distribution of market shares. Specifically, lenders competing in concentrated markets (measured by the HHI) or with larger market shares react much less or not at all to observing competitor information. This pattern is not sensitive to the manner in which we define market shares and concentration or to using relationship-switching rates as an alternative proxy for competitive pressure. Our evidence is consistent with oligopoly models in which lenders react to competitors to preserve their market share: joining PayNet gives more information on what it takes to compete.3 Dominant lenders have less incentive to match rivals, as their market share is less sensitive to competing offers. Evidence in favor of learning about fundamentals is less compelling. A first alternative channel is the revelation of a borrower’s repayment history (Pagano and Jappelli 1993). However, we do not find that the effect is smaller for borrowers with a single relationship, for which the credit file contains no new information for the lender. Another possibility is rooted in information aggregation or other social learning models. Rivals’ offers may reveal their private information, which in turn could help lenders learn about fundamentals. Yet we do not find that specialist lenders react less, although they plausibly have better signals about fundamentals and thus would put less weight on others’ terms. Overall, our interpretation is not that fundamentals are irrelevant for lenders’ terms, but instead that rivals’ maturity is not that informative a signal about fundamentals relative to other sources of information available in the market we study. We therefore emphasize a novel channel of learning about competitors, which operates incrementally to more conventional channels. For robustness, we address several remaining threats to identification. Specifically, there could be shocks either to the borrower or lender that coincide with joining the platform and drive maturity independently of observing rivals’ offers. On the borrower side, our results hold when comparing contracts made to the same borrower by two lenders with different information sets: one that has joined the platform and another that has not. Specifically, we include borrower-time fixed effects (Khwaja and Mian 2008) and find that the member lender of PayNet offers a maturity closer to the bureau average than the nonmember lender in the same period. On the lender side, joining the platform might coincide with a business model shift correlated with the propensity to offer specific contract terms. For example, a lender’s joining could accompany their plans to expand or conserve their capacity, which might have its own effect on maturity. To address this concern, we implement two additional tests that exploit the behavior of other lenders. First, we show that our results hold within lender-year across different market segments. Specifically, our information coverage measure is lender-specific in that it counts only contracts shared by rivals and not the lender itself. Thus, coverage for some segments grows faster than others due to the number of new members joining each period, and such joining decisions (and coverage changes) are beyond the incumbent member’s control.4 Including lender-year fixed effects, we show that the maturity of collateral types with higher coverage tracks the bureau average better than collateral types with low coverage. Second, we isolate large shocks to bureau information arising from new members joining and show that incumbent lenders’ contract terms better track those of their rivals once this extra information is available to them. These additional tests support the interpretation that lenders adjust their contract terms in reaction to the information revealed on the platform. Finally, we investigate a key implication of our learning results. While a full welfare analysis is beyond the scope of this paper, we examine the link between learning from competitors and the incidence of delinquencies during the financial crisis. This episode is revealing in that it consisted of a large wave of unexpected delinquencies. For a group of lenders joining the platform before the financial crisis, we compare the crisis-period delinquencies for contracts that originated just before versus just after joining. Controlling for collateral type-quarter, region-quarter, and lender fixed effects, as well as borrower observables, we find that matching competitors is associated with an increase in delinquencies. An interpretation in line with our main findings is that lenders neglected future risk, either because of greater competition or because they relied more on shared information at the expense of their own information collection. In general, these findings echo those of Murfin and Pratt (2018) and Goldstein and Yang (2019), who argue that technologies that increase the availability of competitor information can have unintended consequences. 1. Related Literature This paper relates to a growing body of empirical literature studying how information and lender coordination affect credit market outcomes. Murfin and Pratt (2018) study comparable pricing in the syndicated loan market. They find that past transactions impact new transaction pricing, but a failure to account for the overlap in information across loans leads to pricing mistakes. While our data lack the power to trace out paths of influence as they do, we nevertheless find suggestive evidence that learning about competitors led to more frequent delinquencies during the financial crisis. Hertzberg, Liberti, and Paravisini (2011) illustrate the role of public information in credit market coordination. Lenders react strongly to the public revelation of information they already possess about a borrower. This publicity effect triggers “run-like” behavior by creditors and financial distress for firms with multiple lenders. By comparison, we study the effect of observing information about other lenders and find evidence of a channel independent of creditor runs. Gorton and He (2008) show that public information about rivals can generate credit cycles as banks update their beliefs on the viability of a collusive arrangement. Kang, Loumioti, and Wittenberg-Moerman (2019) study the introduction of loan-level reporting requirements for the ECB repo borrowers that mandate the disclosure of all contract terms, including prices. They find convergence for price and nonprice contract terms across different locations of the same bank. In credit card markets, Liberman et al. (2018) study the equilibrium effects of information deletion on the allocation of credit and risk, while Foley et al. (2018) show the impact of the information environment on competition. Fuster et al. (2018) study the distributional consequences of machine learning techniques for screening borrowers. Compared to these works and much of the earlier literature on information sharing in credit markets, we focus on learning about competitors as opposed to sharing information about borrowers. We also contribute to the literature that studies the drivers of loan terms–maturity in particular. Hertzberg, Liberman, and Paravisini (2018) examine an online consumer lending platform and show that loan maturity can be used to screen borrowers based on their private information. In the auto loan market, Argyle, Nadauld, and Palmer (2020) show that borrowers display a demand for maturity and target low monthly repayments, while Argyle, Nadauld, Palmer, and Pratt (2018) find that loan maturity impacts the pricing of cars. The literature on information sharing and credit bureaus is vast and includes works by Jappelli and Pagano (2006), Doblas-Madrid and Minetti (2013), Sutherland (2018), Liberti, Sturgess, and Sutherland (2020), Giannetti, Liberti, and Sturgess (2017), Brown, Jappelli, and Pagano (2009), Kovbasyuk and Spagnolo (2018), and Balakrishnan and Ertan (2019). Equally vast is the literature studying the role of information in lending markets more broadly (Hertzberg, Liberti, and Paravisini 2010; Liberti, Seru, and Vig 2016; Hauswald and Marquez 2003; Liberti 2018; Berger, Minnis, and Sutherland 2017). Finally, an extensive body of literature has studied the role played by public firms and public markets in diffusing information.5 In contrast, we study private credit markets for which no centralized price exists, making information technology the primary channel of information diffusion. 2. Equipment Financing and PayNet 2.1. The PayNet platform Our data come from PayNet, an information-sharing platform focusing on the U.S. equipment finance market and SMEs. Borrowers in this market seek loans and leases for an array of assets, including agricultural, construction, manufacturing, medical, office, and retail equipment, as well as computers, copiers, and trucks. Lenders include banks, manufacturers (“captives”), and independent finance companies.6 Since PayNet’s 2001 launch, it has attracted eight of the 10 largest lenders in the market as well as several hundred others as members. Like other credit bureaus, PayNet operates on the principle of reciprocity: members must share information, and only members can purchase the credit files, credit scores, and default probability products offered. PayNet gathers its data by directly connecting into lenders’ IT systems, ensuring that the information shared is comprehensive, reliable, and timely. PayNet has developed these products using 25 million contracts for over |${\$}$|1.7 trillion in transactions collected from members. Lenders are anonymous in the system. Prior to PayNet, lenders generally had access to very limited information about new borrowers and other lenders. Competing data providers, such as Experian, offered partial (and rarely timely) information about trade liabilities, which were much smaller than the typical equipment financing contract. Public UCC filings documented the existence of a contract but did not detail whether the borrower paid on time or the terms received. Thus, PayNet provided equipment finance lenders with a source of timely contract-level information about a borrower’s ability to service similar liabilities and details on previous contracts it received. This development was particularly relevant for small borrowers, who typically lacked audited financial statements or public information about their creditworthiness (Berger, Minnis, and Sutherland 2017; Berger and Udell 2006). Jackson (2001) describes PayNet’s value proposition to lenders: “With richer data you get much better predictive models.... There’s no question there is a need for PayNet’s kind of service. The commercial bureaus haven’t done enough to provide data across all the financial industry lenders.” Although PayNet does not allow lenders to mine its data (e.g., by accessing all credit files for a given industry or ZIP code), lenders can observe how their counterparts contract. During the frequent process of accessing individual credit files, they can see the terms other lenders are providing or have provided a given firm in the past. PayNet’s data collection and verification process is further detailed by Doblas-Madrid and Minetti (2013) and the Internet Appendix of Sutherland (2018). Crucially, unlike many consumer credit bureaus, the platform includes detailed information about contracts offered by competitors. Figure 1 illustrates the information available exclusively to PayNet members. The figure displays a snapshot of a (fictitious) borrower’s credit file accessible on the platform in return for a fee. While the first page of the credit file contains a summary of past payments and the borrower’s state, industry, and age (omitted), subsequent pages reveal the terms of past and current contracts with all lender members of PayNet. In Figure 1, the borrower has two lenders and five contracts in total. For each contract, the maturity, amount, and delinquency status are detailed. Figure 1 Open in new tabDownload slide Past contract terms in PayNet credit file This figure illustrates the information contained in a borrower credit file. Contract terms are highlighted. Figure 1 Open in new tabDownload slide Past contract terms in PayNet credit file This figure illustrates the information contained in a borrower credit file. Contract terms are highlighted. However, similar to other credit bureaus (e.g., the consumer bureaus in the United States), PayNet does not collect or distribute interest rate information and takes care that rates are not recoverable from their data, to reduce concerns about both proprietary costs and the potential for collusion. On the one hand, this choice is revealing and supports our hypothesis that information about competitors can have important effects on credit market outcomes. On the other hand, it means that we cannot directly trace the pricing implications of our hypothesis in this setting. As we discuss in Section 3.5.3, we expect prices to respond in a similar manner as maturity. But we cannot empirically verify this using our data, and this represents a limitation of our study. 2.2. Sample We construct our sample from the quarterly credit files of 20,000 borrowers randomly chosen from PayNet’s database. The files contain detailed information for each of the borrower’s current and past contracts with PayNet members. This information includes the contract’s amount, maturity, payment frequency, collateral type, contract type, and delinquency status, as well as the borrower’s state, industry, and age. The data set provides a constant identifier for borrowers and lenders, which we use to track contracting over time. One limitation is that we cannot match lenders and borrowers to external data with this identifier. Importantly, also note that while we have a large amount of information about lenders’ contract choices, we cannot observe the universe of contracts in the bureau. This implies that an estimate of the average of rivals’ contract terms, although unbiased, is measured with error. Such measurement error can, in general, reduce the statistical significance of our results. We restrict the sample of contracts used for our main analysis to a relatively short window around the lender joining PayNet. We include contracts that originated between the four quarters before and four quarters after the lender joins the bureau. We only study lenders with at least one contract before and one contract after joining the bureau in the given collateral type. This sample selection has little effect on the distribution of loan terms in our sample. 2.2.1 Sample statistics Table 1 describes the lenders and borrowers that meet our regression sample requirements described above. We have 2,076 unique borrowers and 44 unique lenders involved in 8,194 credit relationships with 54,290 contracts. Relationships can span multiple contracts because a borrower’s needs for capital grow over time, and old assets depreciate and new ones with updated features are released. The typical borrower maintains two relationships; however, because borrowers occasionally switch lenders, we observe more relationships across the full sample period. Lenders on average maintain 94 relationships; this understates their true scope, given that we only observe a random snapshot of their clients. Borrowers maintain multiple relationships, in part because lenders can specialize by collateral type. A given firm may, for example, require both computers and forklifts and can access different lenders to finance each. The average lender is exposed to just over six collateral types and the average borrower to 1.7 collateral types. Table IA1 in the Internet Appendix illustrates the distribution of collateral types in the sample. The five most common collateral types are copiers, trucks, construction and mining equipment, computers, and agricultural equipment. Table 1 Sample description No. of borrowers 2,076 No. of lenders 44 No. of relationships 8,194 No. of contracts 54,290 No. of collateral types 23 No. of relationships per lender 94.0 No. of relationships per borrower 2.0 No. of collateral types per lender 6.1 No. of collateral types per borrower 1.7 No. of borrowers 2,076 No. of lenders 44 No. of relationships 8,194 No. of contracts 54,290 No. of collateral types 23 No. of relationships per lender 94.0 No. of relationships per borrower 2.0 No. of collateral types per lender 6.1 No. of collateral types per borrower 1.7 This table presents summary statistics for the borrowers and lenders in our Table 4 regression sample. Open in new tab Table 1 Sample description No. of borrowers 2,076 No. of lenders 44 No. of relationships 8,194 No. of contracts 54,290 No. of collateral types 23 No. of relationships per lender 94.0 No. of relationships per borrower 2.0 No. of collateral types per lender 6.1 No. of collateral types per borrower 1.7 No. of borrowers 2,076 No. of lenders 44 No. of relationships 8,194 No. of contracts 54,290 No. of collateral types 23 No. of relationships per lender 94.0 No. of relationships per borrower 2.0 No. of collateral types per lender 6.1 No. of collateral types per borrower 1.7 This table presents summary statistics for the borrowers and lenders in our Table 4 regression sample. Open in new tab 2.2.2 Oligopolistic competition As in other credit markets for durable goods (cars, real estate, etc.), borrowers in the equipment financing market transact at regular intervals and search for and negotiate with lenders. For this reason, these markets tend not to be defined by a single market-clearing price (Argyle, Nadauld, Palmer, and Pratt 2018). Relationships are prevalent, and lenders can exercise some degree of market power, with the degree of competition affecting borrowers (Rice and Strahan 2010). Nevertheless, market power likely varies across market segments. There is evidence of product market power in equipment sales (Murfin and Pratt 2019; Mian and Smith Jr. 1992; Bodnaruk, O’Brien, and Simonov 2016), which potentially can lead to financing market power if producers are captives or tend to work with a limited number of lenders. Consistent with financing market power, the equipment finance market is highly concentrated.7 Defining market segments as Census district-collateral type pairs (henceforth “region-collateral type pairs”), the median probability that a new contract is issued with a previous lender is 70%, the 25th percentile is 55%, and the 75th percentile is 92%.8 The median number of lenders in each segment is 12, with an interquartile range of 5 to 31. 2.3. Contract terms Table 2 describes the terms for the typical contract in our regression sample. The median (average) contract size is |${\$}$|20,300 (⁠|${\$}$|101,000). The median maturity is 37 months from origination; the average is 44.3 months. Eighty-one percent of contracts are some form of lease (including true leases, conditional sales, and rental leases), while the remaining 19% are loans.9 The overwhelming majority of contracts require fixed monthly payments. The levels of these contract terms are broadly similar before and after a lender joins the platform, although these levels are affected by changes in lender and borrower composition over time. Table 2 Contract characteristics . . All contracts . . Post=1 . . Post=0 . Contract characteristics . . N . Mean . Median . SD . . N . Mean . Median . SD . . N . Mean . Median . SD . Loan size (thousands |${\$}$|⁠) 54,290 101 20.3 593 37,333 104 20.7 589 16,957 93 19.7 605 Maturity (months) 54,290 44.3 37 17 37,333 44.5 39 17 16,957 43.8 37 16 Lease (indicator) 54,290 0.81 1 0.39 37,333 0.81 1 0.39 16,957 0.82 1 0.39 Monthly repayment (indicator) 51,568 0.91 1 0.28 35,410 0.90 1 0.29 16,158 0.92 1 0.26 |$|$|Maturity gap|$|$| (months) 54,290 13.9 11.3 12.8 37,333 14.0 11.4 13.5 16,957 13.5 11.1 11.4 . . All contracts . . Post=1 . . Post=0 . Contract characteristics . . N . Mean . Median . SD . . N . Mean . Median . SD . . N . Mean . Median . SD . Loan size (thousands |${\$}$|⁠) 54,290 101 20.3 593 37,333 104 20.7 589 16,957 93 19.7 605 Maturity (months) 54,290 44.3 37 17 37,333 44.5 39 17 16,957 43.8 37 16 Lease (indicator) 54,290 0.81 1 0.39 37,333 0.81 1 0.39 16,957 0.82 1 0.39 Monthly repayment (indicator) 51,568 0.91 1 0.28 35,410 0.90 1 0.29 16,158 0.92 1 0.26 |$|$|Maturity gap|$|$| (months) 54,290 13.9 11.3 12.8 37,333 14.0 11.4 13.5 16,957 13.5 11.1 11.4 This table summarizes the terms for the contracts in our Table 4 regression sample. The unit of observation is contract. Open in new tab Table 2 Contract characteristics . . All contracts . . Post=1 . . Post=0 . Contract characteristics . . N . Mean . Median . SD . . N . Mean . Median . SD . . N . Mean . Median . SD . Loan size (thousands |${\$}$|⁠) 54,290 101 20.3 593 37,333 104 20.7 589 16,957 93 19.7 605 Maturity (months) 54,290 44.3 37 17 37,333 44.5 39 17 16,957 43.8 37 16 Lease (indicator) 54,290 0.81 1 0.39 37,333 0.81 1 0.39 16,957 0.82 1 0.39 Monthly repayment (indicator) 51,568 0.91 1 0.28 35,410 0.90 1 0.29 16,158 0.92 1 0.26 |$|$|Maturity gap|$|$| (months) 54,290 13.9 11.3 12.8 37,333 14.0 11.4 13.5 16,957 13.5 11.1 11.4 . . All contracts . . Post=1 . . Post=0 . Contract characteristics . . N . Mean . Median . SD . . N . Mean . Median . SD . . N . Mean . Median . SD . Loan size (thousands |${\$}$|⁠) 54,290 101 20.3 593 37,333 104 20.7 589 16,957 93 19.7 605 Maturity (months) 54,290 44.3 37 17 37,333 44.5 39 17 16,957 43.8 37 16 Lease (indicator) 54,290 0.81 1 0.39 37,333 0.81 1 0.39 16,957 0.82 1 0.39 Monthly repayment (indicator) 51,568 0.91 1 0.28 35,410 0.90 1 0.29 16,158 0.92 1 0.26 |$|$|Maturity gap|$|$| (months) 54,290 13.9 11.3 12.8 37,333 14.0 11.4 13.5 16,957 13.5 11.1 11.4 This table summarizes the terms for the contracts in our Table 4 regression sample. The unit of observation is contract. Open in new tab Our analyses study contract maturity, for two reasons. First, maturity impacts firms’ liquidity and investments. During the financial crisis, maturities on loans lasting over a year fell by 30% between 2007 and 2010, before recovering slowly according to the Survey of Terms of Business Lending. Figures 4 and 5 show that contracts in our sample also display considerable time variation throughout the business cycle. Kalemli-Ozcan, Laeven, and Moreno (2018) provide evidence that short maturities and rollover risk were responsible for a large share of the drop in firm investment during the financial crisis. Milbradt and Oehmke (2015) also argue that loan maturity has real effects by distorting firms’ decisions toward inefficiently short-term investments. Second, maturities are regularly negotiated and play an important role in managing risk and allocating credit. Shorter maturities may protect lenders from a deterioration in the borrower’s financial position, but can impose liquidity costs on borrowers. Indeed, the corporate finance literature has shown that, in the presence of frictions, nonprice loan terms are key to credit access, with maturity being a prominent example.10 By contrast, in our context of financing a specific piece of equipment, observed payment frequencies and contract types vary little. And even though interest rates are not reported in our data, contract maturity is relevant since maturity and prices are not perfect substitutes. As for contract sizes, we cannot observe the specific model or quantity of equipment being financed, but offer evidence in Section 4.1 that contract amounts and maturities respond similarly to competitor information. Moreover, maturity choices appear to be far from mechanical and display substantial unexplained variation in the cross-section of borrowers and lenders over our sample period. The raw standard deviation is 17 months, a little less than half of the sample mean. Table AII in the Internet Appendix shows that only about a third of this variation can be explained by collateral type, year, and lender-borrower fixed effects. In the analysis below, we analyze the dispersion in contract terms by computing, for each contract, the gap between its maturity and the bureau’s average maturity (excluding the lender’s own contracts) for that collateral type in the previous quarter. The median gap in our sample is 11 months, which is a substantial fraction of the underlying variation in maturity choice. 2.4. Lender participation in PayNet When a lender joins PayNet, it gains access to information about others’ contracts but must share information about its own contracts, including past ones. This is enforced through PayNet’s direct access to lenders’ IT systems and extensive audit and testing procedures. This back-fill requirement is crucial to our empirical design: we can observe contracts made before and after the lender joins. This allows us to study changes in contracting between the same firm and lender during a relatively short window around the lender joining PayNet. Another key feature of our setting is that lenders join in a staggered pattern over the sample period. This variation offers two benefits. First, the platform information is not publicly revealed: in the same period, some lenders have access to it, while others competing in the same market do not. This within-market-period, across-lender variation allows us to distinguish the effects of the new information from other events affecting lenders or borrowers in a given year. Second, the information revealed to entrants by the platform varies over time as a function of what other lenders are offering. Indeed, lenders often specialize by collateral type; therefore, the bureau coverage across collateral types evolves in a nonsystematic pattern. Thus, members regularly experience shocks to the information coverage in their markets driven by other lenders joining, which is outside of their control.11 We leverage these additional sources of variation in our main specification and robustness tests. Table 3 shows the variation in entry timing for lenders meeting our sample criteria, described in Section 2.2. Sample lenders join in all years between 2002 and 2014 except one. While large lenders tend to join earlier than small ones, in most years, a variety of lenders join. At the same time, joining PayNet is voluntary, and the timing of joining the platform is not randomly assigned. In Section 4.3 below, we perform a series of tests to ensure that our results are not driven by lender or borrower shocks coinciding with the timing of joining. Note also that Liberti, Sturgess, and Sutherland (2020) study in detail the decision to join PayNet. Their main finding is that, when deciding to participate, lenders trade off the greater ability to enter new markets against the threat of losing existing borrowers. In other words, PayNet helps lenders screen new borrowers, which presents both benefits (it reduces adverse selection problems associated with expansion) and costs (it increases the likelihood that the lender loses clients). Because our tests are conducted within the lender-borrower relationship in the short period around the lender’s entry to PayNet, we abstract away from these extensive margin effects. Note also that our sample of lenders differs from that of Liberti, Sturgess, and Sutherland (2020) given our narrow event window and the sample requirements described in Section 2.2. Table 3 Lender entry to bureau . Lenders . Lender size quartile . Year . . 1 . 2 . 3 . 4 . 2002 2 2 2003 1 1 2004 9 1 1 2 5 2005 2 1 1 2006 2 1 1 2007 4 1 3 2008 4 1 3 2009 3 2 1 2010 0 2011 4 3 1 2012 7 1 2 4 2013 6 5 1 Total 44 11 11 11 11 . Lenders . Lender size quartile . Year . . 1 . 2 . 3 . 4 . 2002 2 2 2003 1 1 2004 9 1 1 2 5 2005 2 1 1 2006 2 1 1 2007 4 1 3 2008 4 1 3 2009 3 2 1 2010 0 2011 4 3 1 2012 7 1 2 4 2013 6 5 1 Total 44 11 11 11 11 This table displays the year of joining PayNet for lenders in our Table 4 regression sample according to the size of the lender. Lender size quartiles are assigned according to total credit upon joining the bureau. Open in new tab Table 3 Lender entry to bureau . Lenders . Lender size quartile . Year . . 1 . 2 . 3 . 4 . 2002 2 2 2003 1 1 2004 9 1 1 2 5 2005 2 1 1 2006 2 1 1 2007 4 1 3 2008 4 1 3 2009 3 2 1 2010 0 2011 4 3 1 2012 7 1 2 4 2013 6 5 1 Total 44 11 11 11 11 . Lenders . Lender size quartile . Year . . 1 . 2 . 3 . 4 . 2002 2 2 2003 1 1 2004 9 1 1 2 5 2005 2 1 1 2006 2 1 1 2007 4 1 3 2008 4 1 3 2009 3 2 1 2010 0 2011 4 3 1 2012 7 1 2 4 2013 6 5 1 Total 44 11 11 11 11 This table displays the year of joining PayNet for lenders in our Table 4 regression sample according to the size of the lender. Lender size quartiles are assigned according to total credit upon joining the bureau. Open in new tab 3. Hypothesis Development and Empirical Strategy Lenders’ optimal contract terms depend on a variety of factors. These include not just fundamentals, such as the borrower’s credit risk or the lender’s portfolio performance, but also rivals’ offers. With imperfect competition, a lender’s optimal contract depends on what competitors are offering. Moreover, the information environment plays a role. Lenders have access to public information (industry reports, macroeconomic news) as well as private signals about both fundamentals and their rivals’ offers. Conceptually, joining PayNet can be thought of as receiving an additional signal that is informative about the distribution of rivals’ terms. This learning about competitors can operate through two broad, non-mutually exclusive channels. 3.1. Learning about fundamentals First, lenders may react to rivals’ terms because they reveal private information about the state of the economy or industry. The rational expectations version of this information aggregation effect is canonical in the context of financial markets (Hellwig 1980) but has been much less explored within credit markets. Importantly, at this stage, we want to include under this broad channel other social learning models that are less “rational” or “efficient” in nature, such as information cascades or naïve herding (Murfin and Pratt 2018), as well as rational models with endogenous information acquisition (Goldstein and Yang 2017).12 This class of models would predict convergence in terms across lenders. However, the predicted impact of market structure is less clear, as these models are typically framed in a competitive market or a sequence of decision-making problems. 3.3. Learning about what it takes to compete A second potential channel relates to imperfect competition. The market for financing equipment is not centralized, and not all lenders offer the same contract terms in equilibrium. Instead, buyers search for good deals, and lenders’ choice of terms is driven by attracting or retaining borrowers. The profit-maximizing contract terms balance a higher probability that a contract is accepted with a lower profit margin on that contract. Learning about competitors helps lenders determine the appropriate contract terms and preserve or grow market shares. The sign of the effect, however, is ambiguous. On the one hand, contract terms can be strategic complements: matching rivals’ offers is necessary to attract demand, and we can expect convergence in terms. On the other hand, the industrial organization literature has also raised the possibility that rivals choose to differentiate themselves through product choice, as in Shaked and Sutton (1982). Our empirical tests do not take a stand on the direction of the response and can tease out whether strategic complementarities or the differentiation motive dominates. Nevertheless, the predictions regarding market structure are more clear-cut: lenders in dominant positions have weaker incentives to respond as their market share is less sensitive to rivals’ offers. 3.3. An illustrative framework To illustrate our empirical strategy, we sketch a simple framework in which lenders have dispersed information about their borrowers as well as their competitors. We use the model to describe the effect of joining the platform on contract maturity, as well as how we empirically account for some important confounders. However, because of data limitations, we do not explicitly provide microfoundations for the market game or for the joint optimization of maturity and pricing. The main text is limited to notation and key ideas, while the Internet Appendix contains more details.13 A lender’s optimal contract terms depend on both fundamentals |$\phi$|⁠, such as borrower credit risk and the lender’s risk tolerance, as well as the lender’s competitors’ terms |$m_{-l}$|⁠, due to imperfect competition. A lender’s information |$I_l$| consists of some public information as well as private signals about fundamentals and competitors’ terms. We can decompose lender |$l$|’s choice of maturity |$m$| to firm |$f$|⁠, which is part of a group of similar firms |$g$|⁠, linearly as follows: $$\begin{equation} m^f_l=\underbrace{m_0^g}_{\substack{\text{public} \\ \text{information}}} + \underbrace{\mathbb{E}[\phi^g | I_l]}_{\substack{\text{borrower} \\ \text{fundamentals}}} + \underbrace{\alpha\mathbb{E}[m^g_{-l} | I_l]}_{\substack{\text{competitors}' \\ \text{terms}}} +\underbrace{\eta_{l,f}}_{\substack{\text{idiosyncratic to} \\ \text{relationship}}} \text{.} \label{model} \end{equation}$$(1) The degree to which lenders respond to their competitors’ terms is denoted by |$\alpha$| and summarizes the nature and degree of competition faced by the lender for this borrower. Strategic complements imply that |$\alpha >0$|⁠, while |$\alpha <0$| if the differentiation effect dominates. Finally, the idiosyncratic term |$\eta_{l,f}$| captures factors that are specific to this lender-borrower relationship. Crucially, lenders are uncertain about both fundamentals and their competitors’ actions. Before joining PayNet, lenders have two sources of information in |$I_l$|⁠: (i) public information about fundamentals or competitors’ terms that can be gleaned from, for instance, forecasts of local and national economic conditions or industry reports, summarized in |$m_0 =(m^\phi_0, m^m_0)$|⁠, and (ii) private signals |$s_l=(s^\phi_l, s^m_l)$|⁠, reflecting the lender’s own effort to determine the appropriate contract maturity. After joining the platform, lenders can also observe an additional signal: the average terms offered by competitors (⁠|$\bar{m}^g_{-l}$|⁠) to similar borrowers.14 In equilibrium, the maturity choice depends on the information available to the lender at the time. Before joining, lenders put some weight on their own private signals, depending on their respective precision. After joining, lenders place less weight on their own private signals and place some weight on the bureau average. Importantly, note that reacting to the information in PayNet implies two things: lenders care about rivals’ offers, and they did not have complete information about them before joining. The framework also makes clear that lenders can react to the information through the two non-mutually exclusive channels described above. First, the bureau average |$\bar{m}$| contains information about fundamentals |$\phi$| because each lender’s term is partly influenced by their private signal |$s^\phi_l$| about fundamentals. Second, |$\bar{m}$| is information about rivals’ terms |$m_{-l}$|⁠, which the lender cares about for strategic reasons. The Illustrative Theoretical Framework section of the Internet Appendix further develops this model to more formally illustrate the empirical strategy introduced below. Section 5 will re-examine our evidence in the context of the model and discuss the potential efficiency implications. 3.4. Empirical strategy The main identification threat in isolating the effect of learning about competitors is the existence of unobserved common shocks. Maturity choices are naturally correlated across agents due to public information |$m_0$| as well as private signals |$\{s_l\}$|⁠, independent of the information revealed by the bureau. Then lenders might start offering certain terms at the same time, not because they respond to each other but simply because they react to the same news about fundamentals. The main contribution of our empirical strategy is to specifically account for these unobserved common components. To address this challenge, we exploit the time dimension associated with the lender joining PayNet. Joining leads to a shift in the lender’s information set. Importantly, lenders join in a staggered fashion over 14 years. Our main specification measures how maturity changes within a relationship over a short window around the lender joining. While a lender’s terms may track the bureau average before joining, we ask whether they track it relatively better or worse afterward. Figure 2 provides a graphical illustration of this idea, focusing on the case of convergence for simplicity. Because of common shocks, the lender’s terms are correlated with competitors’ terms even before joining the platform. However, they track the bureau average relatively better after joining. This would be consistent with lenders mimicking competitors. A divergence in terms would generate the opposite pattern, with lenders’ terms generally tracking the bureau average worse after joining. In the data, we can follow lender-borrower relationships over time, including the time before the lender joined the platform. We can also observe rivals’ offers before and after the lender joins. This allows us to test this prediction directly within a fixed-effect regression framework. To illustrate, Figure A3 in the Internet Appendix studies the average maturity for a large lender in the retail equipment market and shows how this lender better tracks the bureau average after they join. Figure 2 Open in new tabDownload slide Empirical strategy: Illustration This figure illustrates our empirical strategy. We assess whether a lender’s maturity tracks the bureau average better or worse after they join. The case of convergence (better tracking) is illustrated. Figure 2 Open in new tabDownload slide Empirical strategy: Illustration This figure illustrates our empirical strategy. We assess whether a lender’s maturity tracks the bureau average better or worse after they join. The case of convergence (better tracking) is illustrated. Other empirical strategies are consistent with our illustrative framework. For example, if there exists a good proxy for a lender’s private or public information, one could test whether the lender’s contracts are less sensitive to this proxy after they join. By comparison, our approach does not require specifying the type or functional form of public or private information. Instead, we test whether lenders’ contracts are more similar to their rivals’ contracts after joining the platform, which is consistent with lenders placing less weight on their other information sources. To estimate learning, our approach only requires constructing a proxy for what lenders can observe in the platform. Given that we observe contracts by other PayNet participants, we can compute a natural proxy: the characteristics of competitor contracts for the relevant collateral type in the bureau. This is significantly less noisy than constructing a proxy for the public or private information lenders observe outside the platform. Nevertheless, we show below that our main results are similar when using a proxy for lenders’ private information instead. 3.5. Addressing confounders By construction, our empirical strategy is not confounded by the existence of a number of factors: public information unobservable to the econometrician |$m_0$|⁠, other sources of information outside of the platform |$s_l$|⁠, or idiosyncratic loan terms |$\eta_{l,f}$|⁠. Indeed, all of these forces exist in the framework above, and our tests based on comparing maturities before and after joining are valid. This is the main advantage of our approach. However, a necessary assumption for identification is that shocks to other sources of public information or to idiosyncratic loan terms are uncorrelated with individual lenders’ decision to join. While this assumption is considerably weaker than assuming that no such shocks exist, it cannot be taken for granted. Specifically, the identification strategy creates the possibility of two important confounders. Lenders’ responses might be driven by (i) information in the platform other than rivals’ offers, namely, the revelation of the borrower’s repayment history, or (ii) shocks unrelated to the platform information but whose timing coincides with the decision to join. We take both concerns seriously and design our main specification as well as additional tests to address them as best we can. 3.5.1 Revelation of borrower past repayment history The contract terms offered to a borrower can be influenced by what a lender learns from the borrower’s PayNet credit file. Note, however, that we restrict attention to lending to previous borrowers, for which the credit file is not necessarily informative. Additionally, we show that our main result holds for borrowers with a single relationship, for which the credit file carries no additional information. 3.5.2 Other shocks correlated with joining PayNet The decision to join the platform is voluntary and can therefore depend on a number of factors that could affect maturity, independent of the information revealed by the bureau. On this front, note that Liberti, Sturgess, and Sutherland (2020) show that lenders joining PayNet are motivated by a desire to enter new markets. However, our main test is exclusively within existing markets. In addition, in Section 4.3 we conduct within borrower-time tests (Khwaja and Mian, 2008) and show that borrower shocks coinciding with the timing of joining cannot explain our results. We also exploit the decision of other lenders to join PayNet in order to address any bias coming from a business model shift, such as a plan to expand or conserve lending capacity. 3.5.3 Price adjustments Finally, one limitation of our empirical setting is that interest rates are not shared in the platform. This implies that we cannot measure price adjustments when we trace the effect on maturity. In this section, we investigate the effect of rate changes in a setting in which firms compete on both rates and maturity by offering a menu of different contracts. Under standard assumptions about credit market competition, our tests based on maturity alone are valid even if rates also react. Figure 3 provides an illustration.15 We assume that the rate-maturity pairs offered by a lender (dashed line) potentially differ from what rivals are offering (solid line) due to dispersed information and market power. We also make a standard assumption for commercial credit markets: a positive relation between rates and maturity (an upward-sloping yield curve) and that, all else being equal, borrowers prefer longer maturities and lower rates. Figure 3 Open in new tabDownload slide Price adjustments This figure illustrates how lenders would adjust interest rates and maturities upon joining the bureau under the assumptions discussed in Section 3.5.3. Panel A presents the average maturity and interest rate for a given lender before they join (dashed line), alongside the same averages for bureau members (solid line). Panel B adds the lender’s new average maturity and interest rate (dashed red line) after they join. Figure 3 Open in new tabDownload slide Price adjustments This figure illustrates how lenders would adjust interest rates and maturities upon joining the bureau under the assumptions discussed in Section 3.5.3. Panel A presents the average maturity and interest rate for a given lender before they join (dashed line), alongside the same averages for bureau members (solid line). Panel B adds the lender’s new average maturity and interest rate (dashed red line) after they join. This lender’s market share is larger for segments in which it offers a better rate. In the example depicted in the figure, the lender’s long-term contracts are more appealing, and in equilibrium, their maturity is larger than the rivals’ maturity. After joining, the lender updates his entire menu offering by adjusting both rates and maturities in the same direction. A convergence in menus implies more competitive short-term contracts, leading to a reduction in the maturity gap. Other reactions seem less plausible: lenders that reduce maturities but raise rates after joining will lose clients, while those that increase maturities and cut rates would leave money on the table. In other words, inferences based on maturity are still likely to be valid with unobservable price adjustments. Figure 4 Open in new tabDownload slide Contract originations in PayNet This figure displays the distribution of contract originations by year for our random sample of PayNet contracts. The sample includes all contracts in our data. Figure 4 Open in new tabDownload slide Contract originations in PayNet This figure displays the distribution of contract originations by year for our random sample of PayNet contracts. The sample includes all contracts in our data. Figure 5 Open in new tabDownload slide Contract maturity by origination year This figure displays the average maturity of the contracts in our regression sample according to origination year. Figure 5 Open in new tabDownload slide Contract maturity by origination year This figure displays the average maturity of the contracts in our regression sample according to origination year. 4. The Effects of Learning about Competitors 4.1. Main specification and findings We design our main specification to answer the following question: does the contract maturity for the same borrower track the lender’s rivals’ maturities more or less after the lender joins the bureau? For each contract, the dependent variable is a measure of the “gap” |$| m_{i}^*- \bar{m}|$| between the maturity offered by the lender and what rivals are offering for similar transactions. The variable of interest is a “Post” indicator, equal to 0 for contracts issued before the lender joins PayNet and 1 for those issued after. A negative coefficient on |$\delta_{post}$| implies that lenders react to the bureau information by offering terms more similar to those of competitors. Specifically, the main specification estimates the following regression: $$\begin{equation} \log|m_{l,f,c,t} - \overline{m_{c, t-1}}| = \delta_{post}+ \eta_{l,f}+ \alpha_{t} + \nu_{contract}+ \varepsilon_{l,f,c,t}. \label{eq:main} \end{equation}$$(2) The unit of observation is a contract originated between firm |$f$| and lender |$l$| at quarter |$t$| to finance a piece of equipment. The dependent variable is the log of the absolute value of the gap between the contract maturity at origination and the bureau average maturity for that collateral type in the previous quarter |$\overline{m_{c, t-1}}$|⁠, excluding the lender’s own contracts.16 We show robustness to using different measures of the bureau average below. The parameter of interest is the coefficient |$\delta_{post}$|⁠. To control for heterogeneous deviations from the average maturity, we add a series of fixed effects. |$\eta_{l,f}$| is a lender-borrower fixed effect that accounts for idiosyncratic time-invariant maturity at the relationship level, including industry and regional variation. Given that lenders join at different times, we include time fixed effects |$\alpha_{t}$| to allow for aggregate time series patterns in the maturity gap. Finally, we include controls |$\nu_{contract}$| for each of the three contract size categories, whether the contract is classified as a lease or a loan, and each borrower risk category based on prior delinquencies.17 Because the decision to join PayNet is made at the lender level, we cluster our standard errors at the lender level.18 Table 4 presents the main result of estimating Equation (2). It shows that, upon joining PayNet, the gap between a lender’s maturity and the bureau average falls by 6% to 7% in absolute value. This effect reveals that observing new information about competitors leads lenders to offer maturities closer to what others are offering. The effect is virtually unchanged whether we use quarter, year, or collateral-year fixed effects to account for aggregate time variation. Table 4 Information sharing and contract maturity: Main specification . Log |$|$| gap |$|$| . . (1) . (2) . (3) . (4) . Post -0.069** -0.069** -0.067** -0.059** [-2.30] [-2.34] [-2.12] [-2.30] Year FE Yes Yes No No Borrower-lender FE Yes Yes Yes Yes Quarter FE No No Yes No Collateral-year FE No No No Yes Controls No Yes Yes Yes N 54,290 54,290 54,290 54,290 Adj. |$R^2$| 0.521 0.522 0.524 0.524 . Log |$|$| gap |$|$| . . (1) . (2) . (3) . (4) . Post -0.069** -0.069** -0.067** -0.059** [-2.30] [-2.34] [-2.12] [-2.30] Year FE Yes Yes No No Borrower-lender FE Yes Yes Yes Yes Quarter FE No No Yes No Collateral-year FE No No No Yes Controls No Yes Yes Yes N 54,290 54,290 54,290 54,290 Adj. |$R^2$| 0.521 0.522 0.524 0.524 This table displays the regression results from estimating Equation (2). The unit of observation is contract. The sample includes contracts that originated between the four quarters before and four quarters after the lender joins the bureau. We study only lenders with at least one contract before and one contract after joining the bureau in the given collateral type. The dependent variable is the log absolute value of the gap between the contract maturity and the bureau average maturity for that collateral type in the previous quarter (excluding the lender’s own contracts). Post is an indicator variable equal to one in quarters after the lender has joined the bureau. Controls include indicators for contract size categories, leases, and the borrowers’ risk category. Standard errors are clustered by lender, and t-statistics are in parentheses. *** |$p<$|⁠.01, ** |$p<$|⁠.05, * |$p<$|⁠.10. Open in new tab Table 4 Information sharing and contract maturity: Main specification . Log |$|$| gap |$|$| . . (1) . (2) . (3) . (4) . Post -0.069** -0.069** -0.067** -0.059** [-2.30] [-2.34] [-2.12] [-2.30] Year FE Yes Yes No No Borrower-lender FE Yes Yes Yes Yes Quarter FE No No Yes No Collateral-year FE No No No Yes Controls No Yes Yes Yes N 54,290 54,290 54,290 54,290 Adj. |$R^2$| 0.521 0.522 0.524 0.524 . Log |$|$| gap |$|$| . . (1) . (2) . (3) . (4) . Post -0.069** -0.069** -0.067** -0.059** [-2.30] [-2.34] [-2.12] [-2.30] Year FE Yes Yes No No Borrower-lender FE Yes Yes Yes Yes Quarter FE No No Yes No Collateral-year FE No No No Yes Controls No Yes Yes Yes N 54,290 54,290 54,290 54,290 Adj. |$R^2$| 0.521 0.522 0.524 0.524 This table displays the regression results from estimating Equation (2). The unit of observation is contract. The sample includes contracts that originated between the four quarters before and four quarters after the lender joins the bureau. We study only lenders with at least one contract before and one contract after joining the bureau in the given collateral type. The dependent variable is the log absolute value of the gap between the contract maturity and the bureau average maturity for that collateral type in the previous quarter (excluding the lender’s own contracts). Post is an indicator variable equal to one in quarters after the lender has joined the bureau. Controls include indicators for contract size categories, leases, and the borrowers’ risk category. Standard errors are clustered by lender, and t-statistics are in parentheses. *** |$p<$|⁠.01, ** |$p<$|⁠.05, * |$p<$|⁠.10. Open in new tab Table 5 shows that the effect is symmetric in that maturity itself does not change on average–only the gap relative to rivals changes. Column 2 confirms that lenders adjust terms in both directions. Panel A of Figure 6 plots the coefficients of a version of Equation (2) in which each quarter before and after joining has its own indicator variable. The omitted category is the quarter prior to joining and is labeled time 0. The plot shows that the change in maturity happens the quarter after the lender joins PayNet and is unlikely to be driven by pre-trends. Figure 6 Open in new tabDownload slide Joining PayNet and contract maturity: dynamic coefficient plots This figure plots the coefficients from estimating a piecewise version of Equation (2) using event quarter indicators. For this plot, we extend the Table 4 sample to include contracts that originated between the eight quarters before and eight quarters after the lender joins the bureau. The dashed lines plot 90% level confidence intervals. Panel A considers the entire sample. For panels B and C, the sample is split according to the median HHI of the collateral type-region-contract size category. Figure 6 Open in new tabDownload slide Joining PayNet and contract maturity: dynamic coefficient plots This figure plots the coefficients from estimating a piecewise version of Equation (2) using event quarter indicators. For this plot, we extend the Table 4 sample to include contracts that originated between the eight quarters before and eight quarters after the lender joins the bureau. The dashed lines plot 90% level confidence intervals. Panel A considers the entire sample. For panels B and C, the sample is split according to the median HHI of the collateral type-region-contract size category. Table 5 Information sharing and contract maturity: Symmetry . (1) . (2) . . Log maturity . Log |$|$|gap|$|$| . Post 0.024 [1.16] Post |$\times$| Positive gap|$_{t-1}$| -0.103^* [-1.68] Post |$\times$| Negative gap|$_{t-1}$| -0.055^* [-1.91] Year FE Yes Yes Borrower-lender FE Yes Yes Controls Yes Yes N 54,290 54,290 Adj. |$R^2$| 0.666 0.522 . (1) . (2) . . Log maturity . Log |$|$|gap|$|$| . Post 0.024 [1.16] Post |$\times$| Positive gap|$_{t-1}$| -0.103^* [-1.68] Post |$\times$| Negative gap|$_{t-1}$| -0.055^* [-1.91] Year FE Yes Yes Borrower-lender FE Yes Yes Controls Yes Yes N 54,290 54,290 Adj. |$R^2$| 0.666 0.522 This table displays the regression results from estimating a modified version of Equation (2). The unit of observation is contract. In column 1, the dependent variable is log contract maturity. In column 2, the dependent variable is the log absolute value of the gap between the contract maturity and the bureau average maturity for that collateral type in the previous quarter (excluding the lender’s own contracts). Positive gap|$_{t-1}$| and Negative gap|$_{t-1}$| are defined based on the last contract in the lender-borrower relationship before the lender joins PayNet. Controls include indicators for contract size categories, leases, and the borrowers’ risk category. Standard errors are clustered by lender, and t-statistics are in parentheses. *** |$p<$|⁠.01, ** |$p<$|⁠.05, * |$p<$|⁠.10. Open in new tab Table 5 Information sharing and contract maturity: Symmetry . (1) . (2) . . Log maturity . Log |$|$|gap|$|$| . Post 0.024 [1.16] Post |$\times$| Positive gap|$_{t-1}$| -0.103^* [-1.68] Post |$\times$| Negative gap|$_{t-1}$| -0.055^* [-1.91] Year FE Yes Yes Borrower-lender FE Yes Yes Controls Yes Yes N 54,290 54,290 Adj. |$R^2$| 0.666 0.522 . (1) . (2) . . Log maturity . Log |$|$|gap|$|$| . Post 0.024 [1.16] Post |$\times$| Positive gap|$_{t-1}$| -0.103^* [-1.68] Post |$\times$| Negative gap|$_{t-1}$| -0.055^* [-1.91] Year FE Yes Yes Borrower-lender FE Yes Yes Controls Yes Yes N 54,290 54,290 Adj. |$R^2$| 0.666 0.522 This table displays the regression results from estimating a modified version of Equation (2). The unit of observation is contract. In column 1, the dependent variable is log contract maturity. In column 2, the dependent variable is the log absolute value of the gap between the contract maturity and the bureau average maturity for that collateral type in the previous quarter (excluding the lender’s own contracts). Positive gap|$_{t-1}$| and Negative gap|$_{t-1}$| are defined based on the last contract in the lender-borrower relationship before the lender joins PayNet. Controls include indicators for contract size categories, leases, and the borrowers’ risk category. Standard errors are clustered by lender, and t-statistics are in parentheses. *** |$p<$|⁠.01, ** |$p<$|⁠.05, * |$p<$|⁠.10. Open in new tab Economically, the effect on borrowers is sizeable. While the coefficient implies roughly a one-month change in contract maturity for the average borrower, maturities are typically set in six-month increments. Our results correspond to a 10% probability of a six-month or larger change in contract maturity. This is notable considering that in our sample, 18% (24%) of borrowers experience a change in their delinquency status over the next six months (one year). Moreover, for the subset of borrowers with fully amortizing contracts, a back-of-the-envelope calculation suggests a 2% change in monthly payments, equivalent to a 2 percentage point change in APR.19 Table 6 subjects our main result to a series of robustness tests. To account for heterogeneous shocks to collateral types across regions, column 1 calculates the bureau average by collateral type-region-quarter, instead of collateral type-quarter, and yields a similar estimate. Table AV in the Internet Appendix also reveals convergence not just toward the mean, but also to other central moments of the distribution of rivals’ terms. We then perform two placebo tests. First, in column 2, we calculate the bureau average using contracts from one year ago instead of current contracts. We expect lenders to react less to stale information. Second, in column 3, we calculate the bureau average using an unrelated collateral type, based on the relatedness measure introduced in Liberti, Sturgess, and Sutherland (2018). For both placebo tests, we find null results. Table 6 Robustness . Bureau average . Bureau . Bureau average . Contract size: . Alternative . . by collateral . average for . for unrelated . |$|$|Size gap|$|$| . specification: . type-quarter-region previous year collateral type Log maturity (1) (2) (3) (4) (5) Post -0.047** -0.042 0.059 -0.238*** [-1.82] [-1.03] [0.42] [-3.29] Borrower risk |$\times$| Relationship length 0.056 [1.15] Post |$\times$| Borrower risk |$\times$| Relationship length -0.055 [-1.54] Year FE Yes Yes Yes Yes Yes Borrower-lender FE Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes N 53,231 41,540 22,484 54,290 54,290 Adj. |$R^2$| 0.510 0.553 0.477 0.332 0.666 . Bureau average . Bureau . Bureau average . Contract size: . Alternative . . by collateral . average for . for unrelated . |$|$|Size gap|$|$| . specification: . type-quarter-region previous year collateral type Log maturity (1) (2) (3) (4) (5) Post -0.047** -0.042 0.059 -0.238*** [-1.82] [-1.03] [0.42] [-3.29] Borrower risk |$\times$| Relationship length 0.056 [1.15] Post |$\times$| Borrower risk |$\times$| Relationship length -0.055 [-1.54] Year FE Yes Yes Yes Yes Yes Borrower-lender FE Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes N 53,231 41,540 22,484 54,290 54,290 Adj. |$R^2$| 0.510 0.553 0.477 0.332 0.666 This table displays the regression results from estimating variations of Equation (2). In columns 1–3, the dependent variable is the log absolute value of the gap between the contract maturity and the bureau average maturity for that collateral type in the previous quarter (excluding the lender’s own contracts). Column 1 calculates the bureau average within collateral type-quarter- region, instead of within collateral type-quarter. Column 2 calculates the bureau average from four quarters ago instead of one quarter ago. Column 3 uses the bureau average for an unrelated collateral type, chosen as the median of the relatedness measure defined by Liberti, Sturgess, and Sutherland (2018). In column 4, the dependent variable is the contract-bureau average gap for contract sizes. In column 5, the dependent variable is the log contract maturity, and the specification includes all main effects and interaction terms. Relationship length is an indicator equal to one if it has been more than three years since the lender-borrower relationship began. In all specifications, the unit of observation is contract. Controls include indicators for contract size categories, leases, and the borrowers’ risk category. Standard errors are clustered by lender, and t-statistics are in parentheses. *** |$p<$|⁠.01, ** |$p<$|⁠.05, * |$p<$|⁠.10. Open in new tab Table 6 Robustness . Bureau average . Bureau . Bureau average . Contract size: . Alternative . . by collateral . average for . for unrelated . |$|$|Size gap|$|$| . specification: . type-quarter-region previous year collateral type Log maturity (1) (2) (3) (4) (5) Post -0.047** -0.042 0.059 -0.238*** [-1.82] [-1.03] [0.42] [-3.29] Borrower risk |$\times$| Relationship length 0.056 [1.15] Post |$\times$| Borrower risk |$\times$| Relationship length -0.055 [-1.54] Year FE Yes Yes Yes Yes Yes Borrower-lender FE Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes N 53,231 41,540 22,484 54,290 54,290 Adj. |$R^2$| 0.510 0.553 0.477 0.332 0.666 . Bureau average . Bureau . Bureau average . Contract size: . Alternative . . by collateral . average for . for unrelated . |$|$|Size gap|$|$| . specification: . type-quarter-region previous year collateral type Log maturity (1) (2) (3) (4) (5) Post -0.047** -0.042 0.059 -0.238*** [-1.82] [-1.03] [0.42] [-3.29] Borrower risk |$\times$| Relationship length 0.056 [1.15] Post |$\times$| Borrower risk |$\times$| Relationship length -0.055 [-1.54] Year FE Yes Yes Yes Yes Yes Borrower-lender FE Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes N 53,231 41,540 22,484 54,290 54,290 Adj. |$R^2$| 0.510 0.553 0.477 0.332 0.666 This table displays the regression results from estimating variations of Equation (2). In columns 1–3, the dependent variable is the log absolute value of the gap between the contract maturity and the bureau average maturity for that collateral type in the previous quarter (excluding the lender’s own contracts). Column 1 calculates the bureau average within collateral type-quarter- region, instead of within collateral type-quarter. Column 2 calculates the bureau average from four quarters ago instead of one quarter ago. Column 3 uses the bureau average for an unrelated collateral type, chosen as the median of the relatedness measure defined by Liberti, Sturgess, and Sutherland (2018). In column 4, the dependent variable is the contract-bureau average gap for contract sizes. In column 5, the dependent variable is the log contract maturity, and the specification includes all main effects and interaction terms. Relationship length is an indicator equal to one if it has been more than three years since the lender-borrower relationship began. In all specifications, the unit of observation is contract. Controls include indicators for contract size categories, leases, and the borrowers’ risk category. Standard errors are clustered by lender, and t-statistics are in parentheses. *** |$p<$|⁠.01, ** |$p<$|⁠.05, * |$p<$|⁠.10. Open in new tab Next, we measure the contract-bureau gap for contract sizes rather than maturity. In our setting, there is drastically more variation in contract sizes than maturity or payment frequency. In some collateral types (such as copiers), contract sizes are often just a few thousand dollars, whereas other (such as aircraft) contracts regularly exceed seven figures. Moreover, within collateral type, there is significant heterogeneity in borrower size, and therefore contract size. To mitigate the influence of outliers on our estimation, we assign each contract’s size to a decile within a collateral type-year and calculate the absolute gap between each contract’s decile and the bureau’s average decile. In column 4, we find that after the lender joins the bureau the contract size gap shrinks by roughly 10% of its preperiod mean. Finally, column 5 implements an empirical approach based on a proxy for lenders’ private information: lender-borrower relationship length. According to our theoretical framework, lenders should put less weight on their private information after joining the bureau.20 To test this, we regress Maturity on Post, Borrower Risk, Relationship Length, and their interactions. We find evidence consistent with our main results. In the prejoining period, Relationship Length mediates the link between Borrower Risk and Maturity (risky borrowers can get longer maturity if their lender has known them longer), but this link disappears in the postjoining period. While the statistical power is limited by the noisiness of our private information and credit risk proxies, this alternative strategy adds credibility to our main results. We make two comments regarding the implications of our main results. First, note that for econometric reasons, our tests are restricted to existing lender-borrower relationships. However, in principle, this effect would apply to new borrowers as well. For example, better information about competitors’ offers is particularly valuable when trying to poach borrowers. Nevertheless, cleanly isolating this effect for new borrowers is particularly difficult. Second, it is plausible that the change in lender behavior upon joining PayNet in turn affects other lenders, implying that there can be knock-on effects that propagate, either through competition or learning. For instance, Murfin and Pratt (2018) document pricing mistakes by tracing out “paths of influence” across syndicated loans and show how these mistakes are propagated across time. However, the fact that we do not observe the universe of lenders and contracts limits our ability to study propagation. 4.2. The role of market structure A natural question is whether the effect of learning about competitors is mediated by market structure. To investigate this, we construct different measures capturing the distribution of market shares. We first measure market concentration using the HHI. We define a “market” either at the collateral type-contract size category level or at the collateral type-contract size category-region level because lenders might compete locally or nationally. To alleviate concerns that local market concentration is directly affected by information sharing, we compute market concentration at the beginning of 2001, before PayNet was introduced. There is considerable variation in concentration across market segments: moving from the 25th to the 75th percentile of the distribution implies a 0.15–0.20 increase in the HHI. We also use relationship-switching rates as an alternative measure of market competitiveness. Some market segments see more relationship switching than others, presumably because of their unique degree of product differentiation, specialization, or other switching costs. Finally, we also construct a within-market across-lender measure that flags lenders that are among the five largest in a collateral type-region-quarter. This classification allows us to distinguish between dominant lenders and the competitive fringe. Table AIII presents summary statistics for these measures. Table 7 shows that our learning results are mediated by market structure. All market structure measures point in the same direction. The first two columns show that the effect is driven by markets with low levels of concentration. In these less-concentrated markets, the gap between the lender’s maturity and the bureau average falls by 8% after joining, while it is statistically unchanged in markets with high concentration levels.21 Column 3 confirms these findings by showing that the effect is driven by markets with high relationship-switching rates. Finally, column 4 suggests the same interpretation: lenders in the competitive fringe are more responsive to information about their competitors, although the distinction is statistically weaker in this specification. Table 7 Information sharing and contract maturity: Split by market structure . Log |$|$|gap|$|$| . . (1) . (2) . (3) . (4) . . Collateral-region- . Collateral- . switching . Top 5 . . contract size HHI . contract size HHI . Rate . lender . Post |$\times$| High HHI -0.030 -0.036 [-0.93] [-1.01] Post |$\times$| Low HHI -0.116*** -0.104*** [-2.91] [-3.93] Post |$\times$| High switching -0.115*** [-4.28] Post |$\times$| Low switching -0.022 [-0.58] Post |$\times$| Top 5 -0.063** [-2.11] Post |$\times$| Not top 5 -0.101** [-2.20] Borrower-lender FE Yes Yes Yes Yes Year FE Yes Yes Yes Yes Controls Yes Yes Yes Yes N 53,305 54,101 54,290 54,290 Adj. |$R^2$| 0.523 0.522 0.523 0.523 . Log |$|$|gap|$|$| . . (1) . (2) . (3) . (4) . . Collateral-region- . Collateral- . switching . Top 5 . . contract size HHI . contract size HHI . Rate . lender . Post |$\times$| High HHI -0.030 -0.036 [-0.93] [-1.01] Post |$\times$| Low HHI -0.116*** -0.104*** [-2.91] [-3.93] Post |$\times$| High switching -0.115*** [-4.28] Post |$\times$| Low switching -0.022 [-0.58] Post |$\times$| Top 5 -0.063** [-2.11] Post |$\times$| Not top 5 -0.101** [-2.20] Borrower-lender FE Yes Yes Yes Yes Year FE Yes Yes Yes Yes Controls Yes Yes Yes Yes N 53,305 54,101 54,290 54,290 Adj. |$R^2$| 0.523 0.522 0.523 0.523 This table displays the regression results from estimating an augmented version of Equation (2) that considers various market structure measures. In columns 1 and 2, market structure is defined according to the median HHI of the collateral type-region-contract size category and collateral type-contract size category, respectively. Column 3 uses the relationship-switching rate, defined as the fraction of relationships in the market last quarter that no longer exist this quarter. Column 4 uses an indicator for whether the lender is among the five largest in this particular collateral type-region-quarter combination. The unit of observation is contract. The dependent variable is the log absolute value of the gap between the contract maturity and the bureau average maturity for that collateral type in the previous quarter (excluding the lender’s own contracts). Controls include indicators for contract size categories, leases, and the borrower’s risk category. Standard errors are clustered by lender, and t-statistics are in parentheses. *** |$p<$|⁠.01, ** |$p<$|⁠.05, * |$p<$|⁠.10. Open in new tab Table 7 Information sharing and contract maturity: Split by market structure . Log |$|$|gap|$|$| . . (1) . (2) . (3) . (4) . . Collateral-region- . Collateral- . switching . Top 5 . . contract size HHI . contract size HHI . Rate . lender . Post |$\times$| High HHI -0.030 -0.036 [-0.93] [-1.01] Post |$\times$| Low HHI -0.116*** -0.104*** [-2.91] [-3.93] Post |$\times$| High switching -0.115*** [-4.28] Post |$\times$| Low switching -0.022 [-0.58] Post |$\times$| Top 5 -0.063** [-2.11] Post |$\times$| Not top 5 -0.101** [-2.20] Borrower-lender FE Yes Yes Yes Yes Year FE Yes Yes Yes Yes Controls Yes Yes Yes Yes N 53,305 54,101 54,290 54,290 Adj. |$R^2$| 0.523 0.522 0.523 0.523 . Log |$|$|gap|$|$| . . (1) . (2) . (3) . (4) . . Collateral-region- . Collateral- . switching . Top 5 . . contract size HHI . contract size HHI . Rate . lender . Post |$\times$| High HHI -0.030 -0.036 [-0.93] [-1.01] Post |$\times$| Low HHI -0.116*** -0.104*** [-2.91] [-3.93] Post |$\times$| High switching -0.115*** [-4.28] Post |$\times$| Low switching -0.022 [-0.58] Post |$\times$| Top 5 -0.063** [-2.11] Post |$\times$| Not top 5 -0.101** [-2.20] Borrower-lender FE Yes Yes Yes Yes Year FE Yes Yes Yes Yes Controls Yes Yes Yes Yes N 53,305 54,101 54,290 54,290 Adj. |$R^2$| 0.523 0.522 0.523 0.523 This table displays the regression results from estimating an augmented version of Equation (2) that considers various market structure measures. In columns 1 and 2, market structure is defined according to the median HHI of the collateral type-region-contract size category and collateral type-contract size category, respectively. Column 3 uses the relationship-switching rate, defined as the fraction of relationships in the market last quarter that no longer exist this quarter. Column 4 uses an indicator for whether the lender is among the five largest in this particular collateral type-region-quarter combination. The unit of observation is contract. The dependent variable is the log absolute value of the gap between the contract maturity and the bureau average maturity for that collateral type in the previous quarter (excluding the lender’s own contracts). Controls include indicators for contract size categories, leases, and the borrower’s risk category. Standard errors are clustered by lender, and t-statistics are in parentheses. *** |$p<$|⁠.01, ** |$p<$|⁠.05, * |$p<$|⁠.10. Open in new tab Figure 6 illustrates the full dynamics of the effect across subsamples with high and low market concentrations, respectively. Panel B shows that, in the most concentrated markets, the gap between a lender’s terms and the bureau average is unaffected by joining. Panel C presents a different pattern for the least concentrated markets. After joining, there is a significant and persistent fall in the gap, implying that lenders adjust their terms toward what others are offering. The gradual reduction in the gap is intuitive: because lenders cannot mine the database, it takes time to aggregate and use the information about rivals contained in individual credit files.22 4.3. Other shocks coinciding with the lender loining PayNet Joining PayNet is voluntary and not randomly assigned. Therefore, we cannot ignore the possibility that our results are due to factors that drive both the decision to join and equilibrium maturities. Recall that access to new markets is the key driver of lenders’ joining PayNet (Liberti, Sturgess, and Sutherland 2020). However, our main test is exclusively within existing markets: it includes lender-borrower fixed effects and is restricted to lenders with contracts in a given collateral type before and after joining. Note also that Figure 6, panel A, reveals no discernible pretrends in our dependent variable prior to joining. Moreover, Table AIV shows that the effect is similar in magnitude across cohorts of lenders joining PayNet in different years. Nevertheless, we leverage the granularity of our data and conduct a number of robustness tests to directly address this threat to identification. 4.3.1 Accounting for borrower shocks On the borrower side, we exploit the fact that, in a given period, some lenders to the same borrower have access to the platform, while others do not. We can use this across-lender variation to distinguish the effects of the new information from other events affecting a given borrower in a given year. Specifically, we include borrower-year fixed effects for the subset of borrowers with multiple lenders: $$\begin{equation} \log|m_{l,f,c, t} - \overline{m_{c, t-1}}| = \delta_{post}+ \eta_{l,f}+ \zeta_{f,t} + \nu_{contract}+ \varepsilon_{l,f,c, t}. \label{eq:borrowerFE} \end{equation}$$(3) Table 8 shows the results of this extended specification. As before, the gap between a lender’s maturity and the bureau average falls after joining in competitive market segments but is unchanged in others. The coefficient reflects the reduction in the gap after joining relative to other lenders of the firm in the post period. This more stringent specification alleviates the concern that our results are driven by shocks to borrower demand or creditworthiness that coincide with the lender’s decision to join PayNet. Table 8 Accounting for borrower shocks . Log |$|$|gap|$|$| . . (1) . (2) . . High HHI . Low HHI . Post 0.048 -0.044* [0.89] [-1.79] Borrower-year FE Yes Yes Borrower-lender FE Yes Yes Controls Yes Yes N 17,615 18,175 Adj. |$R^2$| 0.523 0.561 . Log |$|$|gap|$|$| . . (1) . (2) . . High HHI . Low HHI . Post 0.048 -0.044* [0.89] [-1.79] Borrower-year FE Yes Yes Borrower-lender FE Yes Yes Controls Yes Yes N 17,615 18,175 Adj. |$R^2$| 0.523 0.561 This table displays the regression results from estimating Equation (3). In addition to our Table 4 sample restrictions, these tests are also limited to borrowers with at least two outstanding relationships. The unit of observation is contract. The dependent variable is the log absolute value of the gap between the contract maturity and the bureau average maturity for that collateral type in the previous quarter (excluding the lender’s own contracts). Controls include indicators for contract size categories, leases, and the borrowers’ risk category. Standard errors are clustered by lender, and t-statistics are in parentheses. *** |$p<$|⁠.01, ** |$p<$|⁠.05, * |$p<$|⁠.10. Open in new tab Table 8 Accounting for borrower shocks . Log |$|$|gap|$|$| . . (1) . (2) . . High HHI . Low HHI . Post 0.048 -0.044* [0.89] [-1.79] Borrower-year FE Yes Yes Borrower-lender FE Yes Yes Controls Yes Yes N 17,615 18,175 Adj. |$R^2$| 0.523 0.561 . Log |$|$|gap|$|$| . . (1) . (2) . . High HHI . Low HHI . Post 0.048 -0.044* [0.89] [-1.79] Borrower-year FE Yes Yes Borrower-lender FE Yes Yes Controls Yes Yes N 17,615 18,175 Adj. |$R^2$| 0.523 0.561 This table displays the regression results from estimating Equation (3). In addition to our Table 4 sample restrictions, these tests are also limited to borrowers with at least two outstanding relationships. The unit of observation is contract. The dependent variable is the log absolute value of the gap between the contract maturity and the bureau average maturity for that collateral type in the previous quarter (excluding the lender’s own contracts). Controls include indicators for contract size categories, leases, and the borrowers’ risk category. Standard errors are clustered by lender, and t-statistics are in parentheses. *** |$p<$|⁠.01, ** |$p<$|⁠.05, * |$p<$|⁠.10. Open in new tab 4.3.2 Accounting for lender shocks On the lender side, joining PayNet might coincide with a business model shift, which is potentially correlated with the propensity to offer specific contract maturities. For example, lenders may alter the maturities they offer to support their efforts to increase total lending. This possibility is important to consider because we do not observe lender identities, and therefore cannot rely on public financial statements or other sources to develop business model controls. To address this concern, we design two additional tests that exploit the behavior of other lenders. Specifically, the information coverage in the bureau depends on contracts originated by others and thus varies by collateral type over time in a way that is not directly driven by one’s own decision to join. For example, after lenders join, they have no control over how the bureau’s membership or collateral market coverage evolves. Any given year could see nonsystematic changes in bureau coverage across collateral types based on who else joins, and these coverage changes affect the precision of the bureau average. In the first test, we leverage this variation driven by other lenders to check whether our result holds within lender-year across different collateral types. We can ask whether the maturity of collateral types with higher coverage tracks the bureau average better than collateral types with low coverage. Concretely, we augment Equation (2) as follows: $$\begin{align} & \log|m_{l,f,c, t} - \overline{m_{c, t-1}}|\nonumber\\ &\quad = \delta_{post}*Volume_{c,t-1} + \gamma Volume_{c,t-1}+ \eta_{l,f}+ \xi_{lt} + \nu_{contract}+ \varepsilon_{l,f,c, t}\label{eq:lenderFE}. \end{align}$$(4) The main coefficient of interest is now the Post|$\times$|Volume interaction, where Volume is defined as the number of open contracts in the bureau of the same collateral type as of the previous quarter. We include a lender-year fixed effect |$\xi_{lt}$| that absorbs any change in lenders’ credit supply that is constant across collateral types within a year. Panel A of Table 9 shows the results for this extended specification. The estimated coefficients are consistent with our main finding. For a given lender joining in a specific quarter, the maturity of collateral types with higher coverage tracks the bureau average better than collateral types with low coverage and only so in the most competitive market segments. Columns 3 and 4 also include borrower-year fixed effects for robustness and arrive at the same results. Table 9 Accounting for lender shocks A. Volume tests . . Log |$|$|gap|$|$| . . (1) . (2) . (3) . (4) . . HHI . HHI . . High . Low . High . Low . Post |$\times$| Volume -0.002 -0.011^* 0.002 -0.008** [-0.59] [-1.67] [0.38] [-2.09] Lender-year FE Yes Yes Yes Yes Borrower-lender FE Yes Yes Yes Yes Borrower-year FE No No Yes Yes Controls Yes Yes Yes Yes N 26,142 27,163 17,607 18,163 Adj. |$R^2$| 0.553 0.574 0.525 0.560 A. Volume tests . . Log |$|$|gap|$|$| . . (1) . (2) . (3) . (4) . . HHI . HHI . . High . Low . High . Low . Post |$\times$| Volume -0.002 -0.011^* 0.002 -0.008** [-0.59] [-1.67] [0.38] [-2.09] Lender-year FE Yes Yes Yes Yes Borrower-lender FE Yes Yes Yes Yes Borrower-year FE No No Yes Yes Controls Yes Yes Yes Yes N 26,142 27,163 17,607 18,163 Adj. |$R^2$| 0.553 0.574 0.525 0.560 B. Other lenders’ entry tests . . Log |$|$|gap|$|$| . . (1) . Post large info shock -0.064*** [-2.88] Year FE Yes Borrower-lender FE Yes Controls Yes N 30,498 Adj. |$R^2$| 0.482 B. Other lenders’ entry tests . . Log |$|$|gap|$|$| . . (1) . Post large info shock -0.064*** [-2.88] Year FE Yes Borrower-lender FE Yes Controls Yes N 30,498 Adj. |$R^2$| 0.482 Panel A displays the regression results from estimating Equation (4). Volume (main effect not tabulated for brevity) is defined as the number of contracts in the bureau of the same collateral type in the previous quarter. The sample in columns 1–4 is split according to the median HHI of the collateral type-region-contract size category measured at the contract level. Panel B displays the regression results from estimating a variant of Equation (2) in which the Post indicator is defined with respect to when the lender experiences a large information shock for its primary collateral type after it has joined the bureau. In both panels, the unit of observation is contract. The dependent variable is the log absolute value of the gap between the contract maturity and the bureau average maturity for that collateral type in the previous quarter (excluding the lender’s own contracts). Controls include indicators for contract size categories, leases, and the borrowers’ risk category. Standard errors are clustered by lender, and t-statistics are in parentheses. *** |$p<.01$|⁠, ** |$p<.05$|⁠, * |$p<.10$|⁠. Open in new tab Table 9 Accounting for lender shocks A. Volume tests . . Log |$|$|gap|$|$| . . (1) . (2) . (3) . (4) . . HHI . HHI . . High . Low . High . Low . Post |$\times$| Volume -0.002 -0.011^* 0.002 -0.008** [-0.59] [-1.67] [0.38] [-2.09] Lender-year FE Yes Yes Yes Yes Borrower-lender FE Yes Yes Yes Yes Borrower-year FE No No Yes Yes Controls Yes Yes Yes Yes N 26,142 27,163 17,607 18,163 Adj. |$R^2$| 0.553 0.574 0.525 0.560 A. Volume tests . . Log |$|$|gap|$|$| . . (1) . (2) . (3) . (4) . . HHI . HHI . . High . Low . High . Low . Post |$\times$| Volume -0.002 -0.011^* 0.002 -0.008** [-0.59] [-1.67] [0.38] [-2.09] Lender-year FE Yes Yes Yes Yes Borrower-lender FE Yes Yes Yes Yes Borrower-year FE No No Yes Yes Controls Yes Yes Yes Yes N 26,142 27,163 17,607 18,163 Adj. |$R^2$| 0.553 0.574 0.525 0.560 B. Other lenders’ entry tests . . Log |$|$|gap|$|$| . . (1) . Post large info shock -0.064*** [-2.88] Year FE Yes Borrower-lender FE Yes Controls Yes N 30,498 Adj. |$R^2$| 0.482 B. Other lenders’ entry tests . . Log |$|$|gap|$|$| . . (1) . Post large info shock -0.064*** [-2.88] Year FE Yes Borrower-lender FE Yes Controls Yes N 30,498 Adj. |$R^2$| 0.482 Panel A displays the regression results from estimating Equation (4). Volume (main effect not tabulated for brevity) is defined as the number of contracts in the bureau of the same collateral type in the previous quarter. The sample in columns 1–4 is split according to the median HHI of the collateral type-region-contract size category measured at the contract level. Panel B displays the regression results from estimating a variant of Equation (2) in which the Post indicator is defined with respect to when the lender experiences a large information shock for its primary collateral type after it has joined the bureau. In both panels, the unit of observation is contract. The dependent variable is the log absolute value of the gap between the contract maturity and the bureau average maturity for that collateral type in the previous quarter (excluding the lender’s own contracts). Controls include indicators for contract size categories, leases, and the borrowers’ risk category. Standard errors are clustered by lender, and t-statistics are in parentheses. *** |$p<.01$|⁠, ** |$p<.05$|⁠, * |$p<.10$|⁠. Open in new tab In the second test, we ask whether lenders react to large information shocks due to others joining PayNet. We implement this test in three steps. First, for each lender, we identify the primary collateral type, defined as the one that the lender most frequently finances. Second, for each lender, we identify an event quarter after the lender joined when the bureau experiences the largest increase in contract coverage for the primary collateral type. Although some lenders will share primary collateral types, their staggered joining results in different event quarters. Third, for each lender, we estimate a variant of Equation (2) around the event quarter, where the Post indicator is now defined relative to each lender’s event quarter. Panel B of Table 9 shows the results for this alternative specification. Consistent with our interpretation that lenders react to information about competitors contained in the platform, contract maturities are closer to rivals’ average following a large information inflow after the lender has joined PayNet. Overall, these additional results alleviate the concern that our main findings are purely driven by factors behind the decision to join. 5. Interpreting the Findings and Implications 5.1. Conventional channels of information sharing The previous section provides robust evidence that lenders react to learning about their competitors. In this section, we put this result into perspective with more conventional channels of information sharing in credit markets. We do not claim that these channels are not at play in general; in fact, previous work using PayNet data suggests that some of them are operating in our setting (Doblas-Madrid and Minetti 2013; Sutherland 2018; Liberti, Sturgess, and Sutherland 2020). We argue only that our findings cannot be fully explained by these conventional channels. 5.1.1 Revelation of credit history A key role of credit bureaus is to create credit files that reduce information asymmetries between lenders and borrowers. The revelation of borrowers’ payment histories affects the amount of credit and contract terms. Part of this channel works through a change in the composition of borrowers: worse borrowers are screened out or offered harsher terms, while better borrowers receive better offers (Foley et al. 2018). However, by design, our tests keep the composition of lender-borrower pairs constant by including relationship fixed effects. The effect we document is therefore a change in maturity within a relationship. The revelation of credit histories can affect an existing relationship if a borrower has multiple lenders. Accessing the bureau can reveal negative information to the lender that the borrower previously tried to keep secret. If this channel were driving our result, we expect that it would be smaller or absent for borrowers with (i) a single relationship, because for them the credit file would contain no new information, and (ii) a good credit history. However, Table 10 reveals that there is no significant difference in the effect for single relationship borrowers or borrowers with bad credit records. Table 10 Information sharing and contract maturity: Borrower heterogeneity . (1) . (2) . . Log |$|$|gap|$|$| . Log |$|$|gap|$|$| . Post -0.055** -0.085*** [-2.12] [-2.66] Post |$\times$| Single relationship -0.070 [-1.13] Post |$\times$| Past 90+ days delinquency 0.036 [1.22] Collateral-year FE Yes Yes Borrower-lender FE Yes Yes Controls Yes Yes N 54,290 54,290 Adj. |$R^2$| 0.545 0.545 . (1) . (2) . . Log |$|$|gap|$|$| . Log |$|$|gap|$|$| . Post -0.055** -0.085*** [-2.12] [-2.66] Post |$\times$| Single relationship -0.070 [-1.13] Post |$\times$| Past 90+ days delinquency 0.036 [1.22] Collateral-year FE Yes Yes Borrower-lender FE Yes Yes Controls Yes Yes N 54,290 54,290 Adj. |$R^2$| 0.545 0.545 This table displays the regression results from estimating Equation (2) by borrower type. The interaction in column 1 flags borrowers with one lender at the time of contract origination. The interaction in column 2 flags borrowers whose worst delinquency in the previous three years exceeds 90 days. The unit of observation is contract. The dependent variable is the log absolute value of the gap between the contract maturity and the bureau average maturity for that collateral type in the previous quarter (excluding the lender’s own contracts). Controls include indicators for contract size categories, leases, and the borrowers’ risk category. Standard errors are clustered by lender, and t-statistics are in parentheses. *** |$p<$|⁠.01, ** |$p<$|⁠.05, * |$p<$|⁠.10. Open in new tab Table 10 Information sharing and contract maturity: Borrower heterogeneity . (1) . (2) . . Log |$|$|gap|$|$| . Log |$|$|gap|$|$| . Post -0.055** -0.085*** [-2.12] [-2.66] Post |$\times$| Single relationship -0.070 [-1.13] Post |$\times$| Past 90+ days delinquency 0.036 [1.22] Collateral-year FE Yes Yes Borrower-lender FE Yes Yes Controls Yes Yes N 54,290 54,290 Adj. |$R^2$| 0.545 0.545 . (1) . (2) . . Log |$|$|gap|$|$| . Log |$|$|gap|$|$| . Post -0.055** -0.085*** [-2.12] [-2.66] Post |$\times$| Single relationship -0.070 [-1.13] Post |$\times$| Past 90+ days delinquency 0.036 [1.22] Collateral-year FE Yes Yes Borrower-lender FE Yes Yes Controls Yes Yes N 54,290 54,290 Adj. |$R^2$| 0.545 0.545 This table displays the regression results from estimating Equation (2) by borrower type. The interaction in column 1 flags borrowers with one lender at the time of contract origination. The interaction in column 2 flags borrowers whose worst delinquency in the previous three years exceeds 90 days. The unit of observation is contract. The dependent variable is the log absolute value of the gap between the contract maturity and the bureau average maturity for that collateral type in the previous quarter (excluding the lender’s own contracts). Controls include indicators for contract size categories, leases, and the borrowers’ risk category. Standard errors are clustered by lender, and t-statistics are in parentheses. *** |$p<$|⁠.01, ** |$p<$|⁠.05, * |$p<$|⁠.10. Open in new tab 5.1.2 Creditor runs Alternatively, lenders can react to observing others’ terms due to the fear of a creditor run.23 For instance, Hertzberg, Liberti, and Paravisini (2011) illustrate the effect of information sharing on lender coordination. In the context of maturity choice, Brunnermeier and Oehmke (2013) emphasize the risk of a “maturity rat race,” in which new lenders offer short maturities in an effort to front-run existing creditors. In general, these incentives to run lead to strategic complementarities in maturity choice that could explain a convergence in maturities after joining the bureau. Several factors suggest that a run-based explanation is not behind our results. First, the institutional setting is not conducive to front-running: contracts are attached to a specific piece of equipment and typically have monthly payments. Second, recall from Table 5 that lenders do not shorten their maturities systematically upon joining: lenders adjust their terms toward what others are offering, in both directions. Third, the aforementioned findings in Table 10 contradict a run interpretation; the effect is equally strong for borrowers with good credit records or with a single relationship for which the incentives to run are muted. 5.2. Learning about competitors: Revisiting the channels The illustrative framework presented in Section 3.3 suggests two potential nonexclusive channels of learning about competitors: learning about what it takes to compete and learning about fundamentals. We revisit them in light of our main results and offer further tests aiming to differentiate between the two. Nevertheless, discriminating among all alternative models is difficult, given that market power and beliefs are not directly observable. 5.2.1 Learning about what it takes to compete Under this channel, lenders respond to competitors’ offers to preserve or grow their market share. Interestingly, industrial organization models can disagree on the sign of the effect. Lenders might try to preserve the demand for their contracts by matching rivals’ terms (strategic complementarities) or by trying to differentiate themselves (Shaked and Sutton 1982). Our evidence of lenders adjusting maturities toward what others are offering is consistent with the first view. This result can have important implications because it is well known that strategic complementarities help propagate shocks throughout the economy (Angeletos and Lian 2016). Strategic complementarities are also crucial to determining the total effect of lifting barriers to entry, as they dictate the strength of incumbents’ response to entrants contesting the market. Our findings that maturity adjustments are mediated by market structure are in line with this channel. Lenders in a dominant position and whose market share is less sensitive to competitors face little pressure to respond to what others are offering. Conceptually, lenders’ “market power” should predict the strength of the effect (Bebchuk and Goldstein 2011).24 5.2.2 Learning about fundamentals There can also be an inference effect: competitors’ actions partly reveal their private signals, which are informative about fundamentals such as credit risk or borrower demand in the economy. As opposed to learning about a specific borrower’s credit file, this channel postulates that lenders look at the bureau information to extrapolate to similar borrowers. The rational expectations version of this effect has been studied extensively, but at this stage, other social learning models, such as information cascades or naïve herding (Murfin and Pratt 2018), are equally plausible. The unifying theme is that learning about competitors reduces the lender’s reliance on its own private information. We cannot directly measure lenders’ beliefs, so instead we look for additional cross-sectional evidence consistent with learning about fundamentals. Conceptually, this learning channel does not suggest a clear-cut prediction with respect to the role of market structure. The main reason is that the canonical models tend to be cast in terms of a competitive financial market or through a sequence of decision-making problems. Recent work has incorporated elements of strategic behavior, and a consensus has yet to emerge (Vives 2011; Bernhardt and Taub 2015; Rostek and Weretka 2015). Empirically, Bustamante and Fresard (2017) study how public firms respond to peers’ investment, and document that, consistent with their learning model, peers only influence public firms’ investment in more concentrated industries. This prediction of a learning channel is the opposite of the pattern we document above. We perform several additional tests to look for direct evidence in support of learning about fundamentals. First, we compare the behavior of specialist lenders to others joining the platform. Although specialization is an imperfect proxy for differences in information, the idea is that specialist lenders may have more precise private information and thus put less weight on others’ terms when deciding what to offer.25 We include five definitions of lender specialization, with the intent of capturing lenders that have expertise in a market segment. The first two define specialization based on the number of quarters since the lender’s first contract originated in this collateral type or collateral type-region category. The next two define a lender as a specialist for a specific collateral type if that collateral type is either the most common or one of the top three originated by that lender. Finally, we define a lender as a specialist for a collateral type if that collateral type constitutes at least 30% of its lending portfolio. A potential concern is that specialization measures could be mechanically correlated with market concentration measures, because to some extent they both rely on lenders’ origination volume in different markets. To address this, our first two measures of specialization do not rely on volume; instead, they measure experience length. Additionally, recall that earlier we present results using the relationship-switching rate as an alternative proxy for market power that is not constructed based on volume. Moreover, the average univariate correlation between the HHI and our five specialist measures is below 0.1. This is consistent with specialization and market concentration not being mechanically related. A lender can be specialized in a collateral type without competing in a concentrated market. Likewise, a diversified lender can compete in concentrated markets without being a specialist in any. Overall, we acknowledge that our market structure and specialization measures are imperfect and assess the sensitivity of our results to approaches that do not directly depend on origination volume. If learning about fundamentals drives our main results, then specialists should adjust their terms relatively less upon observing others’ terms. However, the specialist interaction is typically small, of the wrong sign, and insignificant, as displayed in Table 11. Table 11 Information sharing and contract maturity: Lender specialization . Log |$|$|gap|$|$| . . (1) . (2) . (3) . (4) . (5) . Specialist . Quarters since . Quarters since . Lender’s most . In lender’s . Collateral type . definition . 1st contract in . 1st contract in . common . top 3 . >30% of lender’s . . collateral type . collateral type-region . collateral type . collateral types . portfolio . Post |$\times$| Specialist -0.002 -0.000 -0.045 -0.019 -0.038 [-0.90] [-0.14] [-1.08] [-0.49] [-0.79] Post -0.050 -0.086 -0.037 -0.062** -0.040 [-0.75] [-1.33] [-0.78] [-2.01] [-0.75] Specialist 0.017 0.015 0.056 -0.250*** 0.076 [1.19] [1.56] [0.51] [-2.83] [0.52] Borrower-lender FE Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes N 54,290 54,290 54,290 54,290 54,290 Adj. |$R^2$| 0.523 0.524 0.523 0.525 0.523 . Log |$|$|gap|$|$| . . (1) . (2) . (3) . (4) . (5) . Specialist . Quarters since . Quarters since . Lender’s most . In lender’s . Collateral type . definition . 1st contract in . 1st contract in . common . top 3 . >30% of lender’s . . collateral type . collateral type-region . collateral type . collateral types . portfolio . Post |$\times$| Specialist -0.002 -0.000 -0.045 -0.019 -0.038 [-0.90] [-0.14] [-1.08] [-0.49] [-0.79] Post -0.050 -0.086 -0.037 -0.062** -0.040 [-0.75] [-1.33] [-0.78] [-2.01] [-0.75] Specialist 0.017 0.015 0.056 -0.250*** 0.076 [1.19] [1.56] [0.51] [-2.83] [0.52] Borrower-lender FE Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes N 54,290 54,290 54,290 54,290 54,290 Adj. |$R^2$| 0.523 0.524 0.523 0.525 0.523 This table displays the regression results from augmenting Equation (2) with different specialist lender measures and their interaction with Post. Columns 1 and 2 define specialization as the number of quarters since the lender’s first contract originated in this collateral type or collateral type-region category. Columns 3 and 4 define a lender as a specialist for a specific collateral type if that collateral type is either its most common or one of its three most common in terms of originations. Column 5 defines a lender as a specialist for a specific collateral type if that collateral type makes up at least 30% of its lending portfolio. The dependent variable is the log absolute value of the gap between the contract maturity and the bureau average maturity for that collateral type in the previous quarter (excluding the lender’s own contracts). Controls include indicators for contract size categories, leases, and the borrower’s risk category. Standard errors are clustered by lender, and |$t$|-statistics are in parentheses. *** |$p<$|⁠.01, ** |$p<$|⁠.05, * |$p<$|⁠.10. Open in new tab Table 11 Information sharing and contract maturity: Lender specialization . Log |$|$|gap|$|$| . . (1) . (2) . (3) . (4) . (5) . Specialist . Quarters since . Quarters since . Lender’s most . In lender’s . Collateral type . definition . 1st contract in . 1st contract in . common . top 3 . >30% of lender’s . . collateral type . collateral type-region . collateral type . collateral types . portfolio . Post |$\times$| Specialist -0.002 -0.000 -0.045 -0.019 -0.038 [-0.90] [-0.14] [-1.08] [-0.49] [-0.79] Post -0.050 -0.086 -0.037 -0.062** -0.040 [-0.75] [-1.33] [-0.78] [-2.01] [-0.75] Specialist 0.017 0.015 0.056 -0.250*** 0.076 [1.19] [1.56] [0.51] [-2.83] [0.52] Borrower-lender FE Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes N 54,290 54,290 54,290 54,290 54,290 Adj. |$R^2$| 0.523 0.524 0.523 0.525 0.523 . Log |$|$|gap|$|$| . . (1) . (2) . (3) . (4) . (5) . Specialist . Quarters since . Quarters since . Lender’s most . In lender’s . Collateral type . definition . 1st contract in . 1st contract in . common . top 3 . >30% of lender’s . . collateral type . collateral type-region . collateral type . collateral types . portfolio . Post |$\times$| Specialist -0.002 -0.000 -0.045 -0.019 -0.038 [-0.90] [-0.14] [-1.08] [-0.49] [-0.79] Post -0.050 -0.086 -0.037 -0.062** -0.040 [-0.75] [-1.33] [-0.78] [-2.01] [-0.75] Specialist 0.017 0.015 0.056 -0.250*** 0.076 [1.19] [1.56] [0.51] [-2.83] [0.52] Borrower-lender FE Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes N 54,290 54,290 54,290 54,290 54,290 Adj. |$R^2$| 0.523 0.524 0.523 0.525 0.523 This table displays the regression results from augmenting Equation (2) with different specialist lender measures and their interaction with Post. Columns 1 and 2 define specialization as the number of quarters since the lender’s first contract originated in this collateral type or collateral type-region category. Columns 3 and 4 define a lender as a specialist for a specific collateral type if that collateral type is either its most common or one of its three most common in terms of originations. Column 5 defines a lender as a specialist for a specific collateral type if that collateral type makes up at least 30% of its lending portfolio. The dependent variable is the log absolute value of the gap between the contract maturity and the bureau average maturity for that collateral type in the previous quarter (excluding the lender’s own contracts). Controls include indicators for contract size categories, leases, and the borrower’s risk category. Standard errors are clustered by lender, and |$t$|-statistics are in parentheses. *** |$p<$|⁠.01, ** |$p<$|⁠.05, * |$p<$|⁠.10. Open in new tab Second, learning about fundamentals suggests that lenders might not converge toward the raw average of all other comparable contracts, but instead to other types of averages. For instance, a natural model of information aggregation implies convergence to the average of others’ average. Moreover, specialists or lenders with the best portfolio performance might act as “thought leaders” that other lenders try to mimic. Table 12, however, finds little evidence in favor of these hypotheses. The convergence to the average of others’ average is strong, but not statistically different from our baseline effect (repeated in column 1 for convenience). In addition, there is no statistical convergence toward the average of specialists or low-delinquency lenders. Recall that lender identities are not disclosed to PayNet members, making it difficult to distinguish, say, specialists from others. So, in our setting, it may be less practical for lenders to track the average term of specific rivals than the term from all contracts they observe from a given market. Table 12 Learning about fundamentals: Tracking different averages . Log |$|$|gap|$|$| . . (1) . (2) . (3) . (4) . . Other lenders’ . Average of . Specialists’ . Low-delinquency . . average . others’ average . average . lenders’ average . Post -0.069** -0.090** -0.051 -0.049 [-2.34] [-2.26] [-1.21] [-1.59] Year FE Yes Yes Yes Yes Borrower-lender FE Yes Yes Yes Yes Controls Yes Yes Yes Yes N 54,290 54,290 52,601 53,462 Adj. |$R^2$| 0.522 0.515 0.560 0.531 . Log |$|$|gap|$|$| . . (1) . (2) . (3) . (4) . . Other lenders’ . Average of . Specialists’ . Low-delinquency . . average . others’ average . average . lenders’ average . Post -0.069** -0.090** -0.051 -0.049 [-2.34] [-2.26] [-1.21] [-1.59] Year FE Yes Yes Yes Yes Borrower-lender FE Yes Yes Yes Yes Controls Yes Yes Yes Yes N 54,290 54,290 52,601 53,462 Adj. |$R^2$| 0.522 0.515 0.560 0.531 This table displays the regression results from estimating Equation (2) using different versions of the bureau average for the benchmark, as labeled in the column header. Specialist lenders are those for whom the collateral type is one of their three largest. Low-delinquency lenders are those who have a lower-than-average record of delinquencies in that collateral type. The unit of observation is contract. The dependent variable is the log absolute value of the gap between the contract maturity and the respective bureau average maturity for that collateral type in the previous quarter. Controls include indicators for contract size categories, leases, and the borrower’s risk category. Standard errors are clustered by lender, and t-statistics are in parentheses. *** |$p<$|⁠.01, ** |$p<$|⁠.05, * |$p<$|⁠.10. Open in new tab Table 12 Learning about fundamentals: Tracking different averages . Log |$|$|gap|$|$| . . (1) . (2) . (3) . (4) . . Other lenders’ . Average of . Specialists’ . Low-delinquency . . average . others’ average . average . lenders’ average . Post -0.069** -0.090** -0.051 -0.049 [-2.34] [-2.26] [-1.21] [-1.59] Year FE Yes Yes Yes Yes Borrower-lender FE Yes Yes Yes Yes Controls Yes Yes Yes Yes N 54,290 54,290 52,601 53,462 Adj. |$R^2$| 0.522 0.515 0.560 0.531 . Log |$|$|gap|$|$| . . (1) . (2) . (3) . (4) . . Other lenders’ . Average of . Specialists’ . Low-delinquency . . average . others’ average . average . lenders’ average . Post -0.069** -0.090** -0.051 -0.049 [-2.34] [-2.26] [-1.21] [-1.59] Year FE Yes Yes Yes Yes Borrower-lender FE Yes Yes Yes Yes Controls Yes Yes Yes Yes N 54,290 54,290 52,601 53,462 Adj. |$R^2$| 0.522 0.515 0.560 0.531 This table displays the regression results from estimating Equation (2) using different versions of the bureau average for the benchmark, as labeled in the column header. Specialist lenders are those for whom the collateral type is one of their three largest. Low-delinquency lenders are those who have a lower-than-average record of delinquencies in that collateral type. The unit of observation is contract. The dependent variable is the log absolute value of the gap between the contract maturity and the respective bureau average maturity for that collateral type in the previous quarter. Controls include indicators for contract size categories, leases, and the borrower’s risk category. Standard errors are clustered by lender, and t-statistics are in parentheses. *** |$p<$|⁠.01, ** |$p<$|⁠.05, * |$p<$|⁠.10. Open in new tab Finally, we test whether the effect is stronger in market segments in which fundamentals are more persistent. The idea is that rivals’ actions are more informative when there is more persistence. We calculate the persistence of both maturity and delinquencies for each collateral market. Specifically, we fit separate AR(1) models of the quarterly average of maturities and delinquencies on the previous quarterly average for the same collateral type. We obtain the AR(1) coefficient and sort collateral types according to the coefficient. However, Table 13 reveals no significant difference across markets with different persistence for either maturity or delinquencies.26 Of course, one limitation of this test is that persistence is measured with error. Table 13 Learning about fundamentals: Market persistence . Log |$|$|gap|$|$| . . (1) . (2) . (3) . (4) . . Persistence of maturity . Persistence of delinquencies . Post 0.252 -0.072** -0.105 -0.050 [0.57] [-2.12] [-0.47] [-1.58] Post |$\times$| Persistence -0.327 0.041 [-0.70] [0.16] Post |$\times$| High persistence indicator 0.006 -0.043 [0.20] [-1.39] Year FE Yes Yes Yes Yes Borrower-lender FE Yes Yes Yes Yes Controls Yes Yes Yes Yes N 54,290 54,290 54,290 54,290 Adj. |$R^2$| 0.530 0.524 0.523 0.520 . Log |$|$|gap|$|$| . . (1) . (2) . (3) . (4) . . Persistence of maturity . Persistence of delinquencies . Post 0.252 -0.072** -0.105 -0.050 [0.57] [-2.12] [-0.47] [-1.58] Post |$\times$| Persistence -0.327 0.041 [-0.70] [0.16] Post |$\times$| High persistence indicator 0.006 -0.043 [0.20] [-1.39] Year FE Yes Yes Yes Yes Borrower-lender FE Yes Yes Yes Yes Controls Yes Yes Yes Yes N 54,290 54,290 54,290 54,290 Adj. |$R^2$| 0.530 0.524 0.523 0.520 This table displays the regression results from estimating an augmented version of Equation (2). The interactions in columns 1 and 3 measure within-collateral type persistence in maturity and delinquency, respectively, as the AR(1) coefficient of quarterly regressions. The interactions in columns 2 and 4 use an indicator for collateral types with above-median AR(1) coefficients. The unit of observation is contract. The dependent variable is the log absolute value of the gap between the contract maturity and the respective bureau average maturity for that collateral type in the previous quarter. Controls include indicators for contract size categories, leases, and the borrowers’ risk category. Standard errors are clustered by lender, and |$t$|-statistics are in parentheses. *** |$p<$|⁠.01, ** |$p<$|⁠.05, * |$p<$|⁠.10. Open in new tab Table 13 Learning about fundamentals: Market persistence . Log |$|$|gap|$|$| . . (1) . (2) . (3) . (4) . . Persistence of maturity . Persistence of delinquencies . Post 0.252 -0.072** -0.105 -0.050 [0.57] [-2.12] [-0.47] [-1.58] Post |$\times$| Persistence -0.327 0.041 [-0.70] [0.16] Post |$\times$| High persistence indicator 0.006 -0.043 [0.20] [-1.39] Year FE Yes Yes Yes Yes Borrower-lender FE Yes Yes Yes Yes Controls Yes Yes Yes Yes N 54,290 54,290 54,290 54,290 Adj. |$R^2$| 0.530 0.524 0.523 0.520 . Log |$|$|gap|$|$| . . (1) . (2) . (3) . (4) . . Persistence of maturity . Persistence of delinquencies . Post 0.252 -0.072** -0.105 -0.050 [0.57] [-2.12] [-0.47] [-1.58] Post |$\times$| Persistence -0.327 0.041 [-0.70] [0.16] Post |$\times$| High persistence indicator 0.006 -0.043 [0.20] [-1.39] Year FE Yes Yes Yes Yes Borrower-lender FE Yes Yes Yes Yes Controls Yes Yes Yes Yes N 54,290 54,290 54,290 54,290 Adj. |$R^2$| 0.530 0.524 0.523 0.520 This table displays the regression results from estimating an augmented version of Equation (2). The interactions in columns 1 and 3 measure within-collateral type persistence in maturity and delinquency, respectively, as the AR(1) coefficient of quarterly regressions. The interactions in columns 2 and 4 use an indicator for collateral types with above-median AR(1) coefficients. The unit of observation is contract. The dependent variable is the log absolute value of the gap between the contract maturity and the respective bureau average maturity for that collateral type in the previous quarter. Controls include indicators for contract size categories, leases, and the borrowers’ risk category. Standard errors are clustered by lender, and |$t$|-statistics are in parentheses. *** |$p<$|⁠.01, ** |$p<$|⁠.05, * |$p<$|⁠.10. Open in new tab Overall, we do not find strong evidence in favor of the learning about fundamentals channel. Our interpretation is not that fundamentals are irrelevant for lenders’ terms, but that rivals’ maturity is not that informative a signal about fundamentals relative to other sources of information available to lenders. 5.3. Implications Our findings provide a new perspective on how information shocks are transmitted within credit markets. First, we show that information about rivals matters beyond simply information about borrower or lender fundamentals. The economic magnitudes we document are not small. They imply significant changes in rollover risk, which can have a large impact on this population of SMEs with particularly volatile cash flows. Our estimates imply a 10% probability of a six-month or greater change in maturity. This is notable because in our sample 18% of borrowers see a change in their delinquency category over the next six months. Therefore, changing the borrower’s maturity could affect their ability to make required payments on time, which in turn may affect the terms they will receive on their next contract or whether they get credit at all. Second, we argue that the competitive environment is key for the transmission of these information shocks. While most existing works emphasize the aggregation of fundamental information through market-clearing prices, credit markets tend to be decentralized and imperfectly competitive. Interestingly, the rise in big data and algorithm developments across many markets is making learning about competitors increasingly easier. Debates over the effects of information technology on market competition have therefore resurfaced recently. Our findings speak, in a novel way, to this interaction. The equilibrium effects of information flows are inextricably linked to the underlying competitive environment. Conversely, any change in the competitive landscape will influence the transmission of new information (or lack thereof) to the real economy. Much work remains before we fully understand the implications of learning about competitors. The economic forces and welfare considerations at play are subtle. On one hand, pooling information can be beneficial: it can improve production efficiency or remove barriers to competition. On the other hand, information from competitors could facilitate collusion. Moreover, having access to more information can backfire if “mistakes” are propagated as opposed to corrected when information is shared. For instance, Goldstein and Yang (2019) argue that in general, the market quality implications of information disclosure are subtle and can crowd out the production of private information. Murfin and Pratt (2018) document in detail how the use of comparables leads to pricing mistakes in the syndicated loan market, showing, for example, that learning from others leads to incorporating stale information. This in turn implies that market terms will be slow to react to market fundamentals in both downturns and upturns. As often described anecdotally, this can lead to aggressive lending in the downturns, but also, not-aggressive-enough lending in upturns. As shown in Figure IA2, the maturity dynamics across PayNet members and nonmembers appear to broadly follow this pattern. 5.4. Delinquencies during the financial crisis To relate the above implications to our setting, we examine delinquencies during the financial crisis. This is an interesting episode, as it led to a wave of delinquencies that was difficult to predict. Broadly speaking, there are two potential and not mutually exclusive channels that could increase delinquencies. First, enhanced competition can lead lenders to neglect risk as they aggressively compete to preserve their market share. Second, reliance on hard information, such as credit reports and scores, exposes lenders to significant losses caused by negative shocks that are not anticipated by the hard information. Rajan, Seru, and Vig (2015) document this phenomenon in the market for securitized subprime mortgages during this period.27 We exploit the staggered timing of lenders’ joining and study how contracts that originated prior to the crisis performed during the crisis. Specifically, for each lender joining between 2005 and 2007, we study the 2008–2009 performance of contracts that originated shortly before joining, compared to contracts that originated shortly after joining. Our assumption, based on our prior tests, is that lenders do more firm-specific screening before joining, and rely more on shared information after by reacting to what rivals are offering. In addition to lender fixed effects, our tests include indicators for the quarter of origination for each collateral type and the quarter of origination for each borrower region. These fixed effects ensure that our results are not driven by lending to different cohorts with different (and potentially region-specific) default risks. Note however, that unlike our main specification, we cannot control for lender-borrower fixed effects or impose the same sample restrictions. Table 14 shows that contracts that originated just after the lender joined experienced more crisis-period delinquencies than the contracts originated by the same lender just before. Specifically, the postjoin contracts experienced approximately 0.3 more quarters of delinquency from 2008 to 2009 than the prejoin contracts. One interpretation is that a desire to match competitors can backfire if lenders overlook fundamental sources of risk. Table 14 Information sharing and delinquencies during the Financial Crisis . Number of quarters delinquent in 2008–2009 . . (1) . (2) . (3) . (4) . (5) . (6) . . All . High-HHI . Low-HHI . Housing . Other . Exclude . . contracts . market . market . crisis states . states . lenders entering . . . . . . . new markets . Post 0.299** -0.430 0.501** 0.594*** 0.113 0.233** [2.54] [-1.60] [2.73] [3.41] [0.73] [2.10] Lender FE Yes Yes Yes Yes Yes Yes Collateral type- quarter FE Yes Yes Yes Yes Yes Yes Region-quarter FE Yes Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes N 3,236 1,676 1,485 1,324 1,912 3,189 Adj. |$R^2$| 0.211 0.230 0.246 0.247 0.232 0.210 . Number of quarters delinquent in 2008–2009 . . (1) . (2) . (3) . (4) . (5) . (6) . . All . High-HHI . Low-HHI . Housing . Other . Exclude . . contracts . market . market . crisis states . states . lenders entering . . . . . . . new markets . Post 0.299** -0.430 0.501** 0.594*** 0.113 0.233** [2.54] [-1.60] [2.73] [3.41] [0.73] [2.10] Lender FE Yes Yes Yes Yes Yes Yes Collateral type- quarter FE Yes Yes Yes Yes Yes Yes Region-quarter FE Yes Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes N 3,236 1,676 1,485 1,324 1,912 3,189 Adj. |$R^2$| 0.211 0.230 0.246 0.247 0.232 0.210 This table shows the effect of joining PayNet on delinquencies during the crisis. The sample is restricted to (1) lenders joining between 2005 and 2007 and (2) contracts that originated no later than 2006 and that were still open in 2008–2009. The unit of observation is contract. The dependent variable is the number of delinquencies for the contract during 2008 and 2009. HHI is the credit-weighted Herfindahl-Hirschman Index for the market measured in 2001, before the bureau’s inception. Housing crisis states are defined as those states with a greater than 30% housing price decline from peak, according to the FHFA index. Standard errors are clustered by lender, and t-statistics are in parentheses. *** |$p<$|⁠.01, ** |$p<$|⁠.05, * |$p<$|⁠.10. Open in new tab Table 14 Information sharing and delinquencies during the Financial Crisis . Number of quarters delinquent in 2008–2009 . . (1) . (2) . (3) . (4) . (5) . (6) . . All . High-HHI . Low-HHI . Housing . Other . Exclude . . contracts . market . market . crisis states . states . lenders entering . . . . . . . new markets . Post 0.299** -0.430 0.501** 0.594*** 0.113 0.233** [2.54] [-1.60] [2.73] [3.41] [0.73] [2.10] Lender FE Yes Yes Yes Yes Yes Yes Collateral type- quarter FE Yes Yes Yes Yes Yes Yes Region-quarter FE Yes Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes N 3,236 1,676 1,485 1,324 1,912 3,189 Adj. |$R^2$| 0.211 0.230 0.246 0.247 0.232 0.210 . Number of quarters delinquent in 2008–2009 . . (1) . (2) . (3) . (4) . (5) . (6) . . All . High-HHI . Low-HHI . Housing . Other . Exclude . . contracts . market . market . crisis states . states . lenders entering . . . . . . . new markets . Post 0.299** -0.430 0.501** 0.594*** 0.113 0.233** [2.54] [-1.60] [2.73] [3.41] [0.73] [2.10] Lender FE Yes Yes Yes Yes Yes Yes Collateral type- quarter FE Yes Yes Yes Yes Yes Yes Region-quarter FE Yes Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes N 3,236 1,676 1,485 1,324 1,912 3,189 Adj. |$R^2$| 0.211 0.230 0.246 0.247 0.232 0.210 This table shows the effect of joining PayNet on delinquencies during the crisis. The sample is restricted to (1) lenders joining between 2005 and 2007 and (2) contracts that originated no later than 2006 and that were still open in 2008–2009. The unit of observation is contract. The dependent variable is the number of delinquencies for the contract during 2008 and 2009. HHI is the credit-weighted Herfindahl-Hirschman Index for the market measured in 2001, before the bureau’s inception. Housing crisis states are defined as those states with a greater than 30% housing price decline from peak, according to the FHFA index. Standard errors are clustered by lender, and t-statistics are in parentheses. *** |$p<$|⁠.01, ** |$p<$|⁠.05, * |$p<$|⁠.10. Open in new tab Admittedly, this is not the only possible explanation, and although our data cannot reject alternatives with absolute confidence, we offer additional supporting evidence. First, in line with our prior results, we also find that the delinquency increase is entirely driven by more competitive markets, as shown in columns 2 and 3. Second, we identify states with the largest drop in housing prices during the crisis, where a substitution from screening to mimicking should result in worse contract outcomes.28 Even after controlling for region-origination quarter fixed effects, columns 4 and 5 show more delinquencies for postjoining contracts only in large housing price drop states. To mitigate concerns that our delinquency evidence stems from new market entry by lenders, column 6 limits the sample to the lender’s existing markets. Our results are unaffected. Several additional untabulated results support our inference that the rise in delinquencies relates to learning about competitors. We find a reduction in the average gap between the lender’s contract maturity and rivals’ maturity after joining PayNet, but this decline is most pronounced for contracts that end up delinquent. Finally, lenders do not seem to target riskier borrowers after joining. On the contrary, if anything, borrowers’ credit records improve, consistent with the canonical information effect of credit bureaus (Doblas-Madrid and Minetti 2013). Accordingly, we find that the effect is large for existing borrowers. Because the set of lenders joining PayNet a few years before the financial crisis instead of in other periods is small and potentially selected, we take this evidence as suggestive. Nevertheless, it supports the idea that incentives to match competitors can have a cost if they lead to the neglect of fundamental risk. 6. Conclusion We show how contract terms and outcomes are shaped by the availability of competitor information. Using microdata from the introduction of an information-sharing platform, we find that upon joining, lenders adjust their terms toward what others are offering. Further tests reveal that this effect is unlikely to be driven by lenders learning about fundamentals. Instead, we argue that imperfect competition plays a key role: information about rivals allows lenders to learn about what it takes to compete. These findings imply a new perspective on how information shocks are transmitted within credit markets and, ultimately, to the real economy. Information about rivals matters beyond simply information about borrower or lender fundamentals, and market structure is key for the transmission of these information shocks. These results speak directly to recent trends that have attracted considerable attention from academics and policymakers. Learning about competitors is becoming increasingly easier given the rise of large pooled databases and improvements in data mining, in credit markets and beyond. Many works have also documented a rise in credit market concentration. The implications for consumer welfare, production efficiency, and policy design are open questions worthy of further investigation. Finally, one limitation of our evidence is that, like most information-sharing systems (e.g., the consumer bureaus in the United States), the bureau we study does not collect interest rates. While we point to several reasons why rates and maturities should respond similarly to competitor information in Section 3.5.3, we cannot directly test this. Therefore, future research investigating rate dynamics is warranted. Authors have furnished an Internet Appendix, which is available on the Oxford University Press Web site next to the link to the final published paper online Acknowledgement For helpful comments, we thank Hassan Afrouzi, Marieke Bos, Mariassunta Gianetti, Xavier Giroud, Andrew Hertzberg, Jennifer La’O, Andres Liberman, Erik Loualiche, Andrew MacKinlay (discussant), Jonathan Parker, Tomasz Piskorski, Ryan Pratt (discussant), Johannes Stroebel, David Thesmar, Laura Veldkamp, Rodrigo Verdi, Dong Yan, Edison Yu, and seminar participants at Columbia Business School, the Federal Reserve Bank of Philadelphia, the Federal Reserve Bank of New York, MIT Sloan, NYU Stern, the Stockholm School of Economics, the University of Bonn, Yale SOM, the 2019 Midwest Finance Association conference, and the 2019 Western Finance Association conference. Supplementary data can be found on The Review of Financial Studies web site. Footnotes 1 In 2016, European Commissioner for Competition Margrethe Vestager argued that “The future of big data is not just about technology. It’s about things like. . . competition” at the EDPS-BEUC Conference on Big Data and Competition. 2 Joining involves an invasive implementation process in which PayNet establishes access to the lenders’ IT systems to ensure complete and truthful sharing. PayNet uses shared information to create credit scores and reports for members. Nonmembers cannot access the system or its scores and reports. 3 For example, see Li (1985) for an early model of oligopoly and information sharing. 4 For example, after a truck captive joins, there is a large increase in the platform’s coverage of truck contracts but no new contracts for copiers. Thus, lenders who had joined before this truck captive experience an information shock for reasons beyond their control (they have no say over the truck captive joining) and only to the extent that they lend against trucks. 5 See, for instance, Sockin and Xiong (2015), Chen, Goldstein, and Jiang (2006), Foucault and Fresard (2014), Dessaint, Foucault, Frésard, and Matray (2019), Kurlat and Veldkamp (2015), Veldkamp (2006), Leary and Roberts (2014), Badertscher, Shroff, and White (2013) Bustamante and Fresard (2017), Iyer, Khwaja, Luttmer, and Shue (2015), Broecker (1990), and Breuer, Leuz, and Vanhaverbeke (2019). 6 Murfin and Pratt (2019) provide an explanation for the existence of captives in equipment financing. 7 According to a 2018 industry report, the top five (10, 25) lenders in the equipment finance market constitute 40% (55%, 82%) of industry assets. 8 Throughout this paper, we use the term “region” to refer to one of the nine Census divisions, described at https://factfinder.census.gov/help/en/division.htm. 9 The borrower’s choice between a lease or a loan can depend on many considerations, including cost, tax or financial reporting treatment, different services offered under each contract type, the borrower’s credit risk and liquidity, and obsolescence risk. For our purposes, these contracts function similarly. In the context of captive financing, Murfin and Pratt (2019) highlight the fundamental similarities of leases and loans. 10 Hertzberg, Liberman, and Paravisini (2018) document that demand for maturity is heterogeneous in consumer credit markets and that maturity can be used to screen applicants. We abstract from screening by focusing on repeat borrowers. 11 Figure IA1 in the Internet Appendix shows that there is considerable time variation in the volume of contracts in the bureau across collateral types. 12 Note also the difference from the canonical effect of credit files, in which lenders learn about a specific borrower from its payment history. Instead, here lenders use the bureau information to extrapolate to other similar borrowers (e.g., with respect to size, sector, or collateral type). 13 Our mathematical notation borrows from canonical “beauty contest” models exemplified by Morris and Shin (2002). Note, however, that we use it for a different purpose and the underlying economics and microfoundations differ. 14 Concretely, lenders can learn about others’ terms by purchasing individual credit files from PayNet. This makes it unlikely that they can learn the entire distribution of competitors’ terms or that they can leak this information easily. 15 We thank our discussant Andrew MacKinlay for suggesting this illustration. 16 Excluding the lender’s own contracts and using a lag helps address the mechanical aspects of the reflection problem of Manski (1993). 17 Specifically, the three contract size categories are: small ticket (below |${\$}$|250k), medium ticket (between |${\$}$|250k and |${\$}$|5M), and big ticket (above |${\$}$|5M). The three delinquencies categories are: no missed payments, missed payments 90 or fewer days late, and default or missed payments over 90 days late, all measured over the last three years. 18 Our results are similar if we double-cluster by lender and collateral type-year. 19 For example, the median contract is for |${\$}$|20,000 and 37 months, which corresponds to a |${\$}$|678 payment per month. Reducing maturity to 36 months increases monthly payments to |${\$}$|693, roughly comparable to increasing the interest rate from 15% to 17% (⁠|${\$}$|698). Since we cannot directly observe interest rates or any embedded options in our data, this calculation relies on the Schalheim and Zhang (2017) estimate of the mean annualized interest rate of 15% during this period. 20 We thank an anonymous reviewer for suggesting this alternative approach. 21 We also find a similar pattern for other terms including contract size: there is more convergence in competitive markets. 22 In principle, the speed of learning might be slower in less competitive markets. However, when we extend our event window, we continue to find null results for these markets. 23 More broadly, a number of papers, such as Morris and Shin (1998), Bebchuk and Goldstein (2011), Goldstein, Ozdenoren, and Yuan (2011), and Goldstein and Pauzner (2005) have emphasized the role of information in explaining run-like behavior. 24 Ideally, we would also use data on applications to measure directly how the take-up rate of a lender’s offer depends on rivals’ maturity, as in Argyle, Nadauld, Palmer, and Pratt (2018). Unfortunately, PayNet does not collect data on applications. 25 Stroebel (2016), Kurlat and Stroebel (2015), and Loutskina and Strahan (2011) also exploit heterogeneity in expertise in the context of real estate markets. 26 A countervailing force is that persistent fundamentals can also lead to strong priors, muting the reaction from observing rivals. 27 More generally, this is related to the Lucas (1976) critique. See also Farboodi, Matray, and Veldkamp (2018) for a recent discussion of how the use of information by the stock market can deviate from the social optimum. 28 Housing crisis states are defined as those with a greater than 30% housing price decline from their peak, according to the FHFA index (14 states). References Angeletos, G.-M. , and Lian C.. 2016 . Incomplete information in macroeconomics: Accommodating frictions in coordination. In Handbook of Macroeconomics , Volume 2 , ed. Goldstein Itay, pp. 1065 – 240 . Oxford, UK : Elsevier . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Argyle, B. , Nadauld T., and Palmer C.. 2020 . Monthly payment targeting and the demand for maturity . The Review of Financial Studies . Google Scholar OpenURL Placeholder Text WorldCat Argyle, B. , Nadauld T., Palmer C., and Pratt R.. 2018 . The capitalization of consumer financing into durable goods prices . Working paper , Brigham Young University . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Badertscher, B. , Shroff N., and White H. D.. 2013 . Externalities of public firm presence: Evidence from private firms’ investment decisions . Journal of Financial Economics 109 : 682 – 706 . Google Scholar Crossref Search ADS WorldCat Balakrishnan, K. , and Ertan A.. 2019 . Credit information sharing and loan loss recognition . Working paper , London Business School . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Bebchuk, L. A. , and Goldstein I.. 2011 . Self-fulfilling credit market freezes . The Review of Financial Studies 24 : 3519 – 55 . Google Scholar Crossref Search ADS WorldCat Berger, A. N. , and Udell G. F.. 2006 . A more complete conceptual framework for SME finance . Journal of Banking & Finance 30 : 2945 – 66 . Google Scholar Crossref Search ADS WorldCat Berger, P. G. , Minnis M., and Sutherland A.. 2017 . Commercial lending concentration and bank expertise: Evidence from borrower financial statements . Journal of Accounting and Economics 64 : 253 – 77 . Google Scholar Crossref Search ADS WorldCat Bernhardt, D. , and Taub B.. 2015 . Learning about common and private values in oligopoly . The RAND Journal of Economics 46 : 66 – 85 . Google Scholar Crossref Search ADS WorldCat Bodnaruk, A. , O’Brien W., and Simonov A.. 2016 . Captive finance and firm’s competitiveness . Journal of Corporate Finance 37 : 210 – 28 . Google Scholar Crossref Search ADS WorldCat Breuer, M. , Leuz C., and Vanhaverbeke S.. 2019 . Mandated financial reporting and corporate innovation . Working paper , Columbia University . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Broecker, T. 1990 . Credit-worthiness tests and interbank competition . Econometrica 58 : 429 – 52 . Google Scholar Crossref Search ADS WorldCat Brown, M. , Jappelli T., and Pagano M.. 2009 . Information sharing and credit: Firm-level evidence from transition countries . Journal of Financial Intermediation 18 : 151 – 72 . Google Scholar Crossref Search ADS WorldCat Brunnermeier, M. K. , and Oehmke M.. 2013 . The maturity rat race . The Journal of Finance 68 : 483 – 521 . Google Scholar Crossref Search ADS WorldCat Bustamante, M. C. , and Fresard L.. 2017 . Does firm investment respond to peers’ investment? Working paper , University of Maryland . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Chen, Q. , Goldstein I., and Jiang W.. 2006 . Price informativeness and investment sensitivity to stock price . The Review of Financial Studies 20 : 619 – 50 . Google Scholar Crossref Search ADS WorldCat Dessaint, O. , Foucault T., Frésard L., and Matray A.. 2019 . Noisy stock prices and corporate investment . The Review of Financial Studies 32 : 2625 – 72 . Google Scholar Crossref Search ADS WorldCat Doblas-Madrid, A. , and Minetti R.. 2013 . Sharing information in the credit market: Contract-level evidence from us firms . Journal of Financial Economics 109 : 198 – 223 . Google Scholar Crossref Search ADS WorldCat Farboodi, M. , Matray A., and Veldkamp L.. 2018 . Where has all the big data gone? Working paper , MIT Sloan . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Foley, F. , Hurtado A., Liberman A., and Sepulveda A.. 2018 . The effects of information on credit market competition: Evidence from credit cards . Working paper , Harvard Business School . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Foucault, T. , and Fresard L.. 2014 . Learning from peers’ stock prices and corporate investment . Journal of Financial Economics 111 : 554 – 77 . Google Scholar Crossref Search ADS WorldCat Fuster, A. , Goldsmith-Pinkham P., Ramadorai T., and Walther A.. 2018 . Predictably unequal? The effects of machine learning on credit markets . Working paper , Swiss National Bank . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Giannetti, M. , Liberti J., and Sturgess J.. 2017 . Information sharing and rating manipulation . The Review of Financial Studies 30 : 3269 – 304 . Google Scholar Crossref Search ADS WorldCat Goldstein, I. , Ozdenoren E., and Yuan K.. 2011 . Learning and complementarities in speculative attacks . The Review of Economic Studies 78 : 263 – 92 . Google Scholar Crossref Search ADS WorldCat Goldstein, I. , and Pauzner A.. 2005 . Demand–deposit contracts and the probability of bank runs . The Journal of Finance 60 : 1293 – 327 . Google Scholar Crossref Search ADS WorldCat Goldstein, I. , and Yang L.. 2017 . Information disclosure in financial markets . Annual Review of Financial Economics 9 : 101 – 25 . Google Scholar Crossref Search ADS WorldCat Goldstein, I. , and Yang L.. 2019 . Good disclosure, bad disclosure . Journal of Financial Economics 131 : 118 – 38 . Google Scholar Crossref Search ADS WorldCat Gorton, G. B. , and He P.. 2008 . Bank credit cycles . The Review of Economic Studies 75 : 1181 – 214 . Google Scholar Crossref Search ADS WorldCat Hauswald, R. , and Marquez R.. 2003 . Information technology and financial services competition . The Review of Financial Studies 16 : 921 – 48 . Google Scholar Crossref Search ADS WorldCat Hellwig, M. F. 1980 . On the aggregation of information in competitive markets . Journal of Economic Theory 22 : 477 – 98 . Google Scholar Crossref Search ADS WorldCat Hertzberg, A. , Liberman A., and Paravisini D.. 2018 . Screening on loan terms: evidence from maturity choice in consumer credit . The Review of Financial Studies 31 : 3532 – 67 . Google Scholar Crossref Search ADS WorldCat Hertzberg, A. , Liberti J., and Paravisini D.. 2010 . Information and incentives inside the firm: Evidence from loan officer rotation . The Journal of Finance 65 : 795 – 828 . Google Scholar Crossref Search ADS WorldCat Hertzberg, A. , Liberti J., and Paravisini D.. 2011 . Public information and coordination: Evidence from a credit registry expansion . The Journal of Finance 66 : 379 – 412 . Google Scholar Crossref Search ADS WorldCat Iyer, R. , Khwaja A. I., Luttmer E. F., and Shue K.. 2015 . Screening peers softly: Inferring the quality of small borrowers . Management Science 62 : 1554 – 577 . Google Scholar Crossref Search ADS WorldCat Jackson, S. 2001 . Better credit data, better credit decisions . Equipment Leasing Today 44 – 49 . Google Scholar OpenURL Placeholder Text WorldCat Jappelli, T. , and Pagano M.. 2006 . The role and effects of credit information sharing. in The Economics of Consumer Credit: European Experience and Lessons from the US , ed. Bertola Giuseppe, Disney Richard and Grant Charles. Cambridge, US : MIT Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Kalemli-Ozcan, S. , Laeven L., and Moreno D.. 2018 . Debt overhang, rollover risk, and corporate investment: Evidence from the European crisis . Working paper , University of Maryland . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Kang, J. K. , Loumioti M., and Wittenberg-Moerman R.. 2019 . Lifting the bank veil: Credit standards’ harmonization through lending transparency . Working paper , University of Southern California . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Khwaja, A. I. , and Mian A.. 2008 . Tracing the impact of bank liquidity shocks: Evidence from an emerging market . American Economic Review 98 : 1413 – 42 . Google Scholar Crossref Search ADS WorldCat Kovbasyuk, S. , and Spagnolo G.. 2018 . Memory and markets . Working paper , New Economic School . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Kurlat, P. , and Stroebel J.. 2015 . Testing for information asymmetries in real estate markets . The Review of Financial Studies 28 : 2429 – 61 . Google Scholar Crossref Search ADS WorldCat Kurlat, P. , and Veldkamp L.. 2015 . Should we regulate financial information? Journal of Economic Theory 158 : 697 – 720 . Google Scholar Crossref Search ADS WorldCat Leary, M. T. , and Roberts M. R.. 2014 . Do peer firms affect corporate financial policy? The Journal of Finance 69 : 139 – 78 . Google Scholar Crossref Search ADS WorldCat Li, L. 1985 . Cournot oligopoly with information sharing . The RAND Journal of Economics 16 : 521 – 36 . Google Scholar Crossref Search ADS WorldCat Liberman, A. , Neilson C., Opazo L., and Zimmerman S.. 2018 . The equilibrium effects of asymmetric information: Evidence from consumer credit markets . Working paper , New York University . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Liberti, J. M. 2018 . Initiative, incentives, and soft information . Management Science 64 : 3714 – 34 . Google Scholar Crossref Search ADS WorldCat Liberti, J. , Seru A., and Vig V.. 2016 . Information, credit, and organization . Working paper , Northwestern University . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Liberti, J. , Sturgess J., and Sutherland A.. 2018 . Economics of voluntary information sharing . Working paper , Northwestern University . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Liberti, J. , Sturgess J., and Sutherland A.. 2020 . How voluntary information sharing systems form: Evidence from a us commercial credit bureau . Working paper , Brigham Young University . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Loutskina, E. , and Strahan P. E.. 2011 . Informed and uninformed investment in housing: The downside of diversification . The Review of Financial Studies 24 : 1447 – 80 . Google Scholar Crossref Search ADS WorldCat Lucas, R. E. 1976 . Econometric policy evaluation: A critique . Carnegie-Rochester conference series on public policy 1 : 19 – 46 . Google Scholar Crossref Search ADS WorldCat Manski, C. F. 1993 . Identification of endogenous social effects: The reflection problem . The Review of Economic Studies 60 : 531 – 42 . Google Scholar Crossref Search ADS WorldCat Mian, S. L. , and Smith Jr C. W.. 1992 . Accounts receivable management policy: Theory and evidence . The Journal of Finance 47 : 169 – 200 . Google Scholar Crossref Search ADS WorldCat Milbradt, K. , and Oehmke M.. 2015 . Maturity rationing and collective short-termism . Journal of Financial Economics 118 : 553 – 70 . Google Scholar Crossref Search ADS WorldCat Morris, S. , and Shin H. S.. 1998 . Unique equilibrium in a model of self-fulfilling currency attacks . American Economic Review 88 : 587 – 97 . Google Scholar OpenURL Placeholder Text WorldCat Morris, S. , and Shin H. S.. 2002 . Social value of public information . American Economic Review 92 : 1521 – 34 . Google Scholar Crossref Search ADS WorldCat Murfin, J. , and Pratt R.. 2018 . Comparables pricing . The Review of Financial Studies 32 : 688 – 737 . Google Scholar Crossref Search ADS WorldCat Murfin, J. , and Pratt R.. 2019 . Who finances durable goods and why it matters: Captive finance and the coase conjecture . The Journal of Finance 74 : 755 – 93 . Google Scholar Crossref Search ADS WorldCat Pagano, M. , and Jappelli T.. 1993 . Information sharing in credit markets . The Journal of Finance 48 : 1693 – 718 . Google Scholar Crossref Search ADS WorldCat Rajan, U. , Seru A., and Vig V.. 2015 . The failure of models that predict failure: Distance, incentives, and defaults . Journal of Financial Economics 115 : 237 – 60 . Google Scholar Crossref Search ADS WorldCat Rice, T. , and Strahan P. E.. 2010 . Does credit competition affect small-firm finance? The Journal of Finance 65 : 861 – 89 . Google Scholar Crossref Search ADS WorldCat Rostek, M. , and Weretka M.. 2015 . Information and strategic behavior . Journal of Economic Theory 158 : 536 – 57 . Google Scholar Crossref Search ADS WorldCat Schalheim, J. , and Zhang X.. 2017 . An examination of yields on small business equipment leases . Working paper , University of Utah . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Shaked, A. , and Sutton J.. 1982 . Relaxing price competition through product differentiation . The Review of Economic Studies 49 : 3 – 13 . Google Scholar Crossref Search ADS WorldCat Sockin, M. , and Xiong W.. 2015 . Informational frictions and commodity markets . The Journal of Finance 70 : 2063 – 98 . Google Scholar Crossref Search ADS WorldCat Stroebel, J. 2016 . Asymmetric information about collateral values . The Journal of Finance 71 : 1071 – 112 . Google Scholar Crossref Search ADS WorldCat Sutherland, A. 2018 . Does credit reporting lead to a decline in relationship lending? Evidence from information sharing technology . Journal of Accounting and Economics 66 : 123 – 41 . Google Scholar Crossref Search ADS WorldCat Tirole, J. ( 2010 ). The theory of corporate nance. Princeton , NJ : Princeton University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Veldkamp, L. 2006 . Information markets and the comovement of asset prices . The Review of Economic Studies 73 : 823 – 45 . Google Scholar Crossref Search ADS WorldCat Vives, X. 2006 . Information sharing: Economics and antitrust . in The Pros and Cons of Information Sharing, Swedish Competition Authority p.83-100 , Kalmar : Lenanders Grafiska AB . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Vives, X. . 2011 . Strategic supply function competition with private information . Econometrica 79 : 1919 – 66 . Google Scholar Crossref Search ADS WorldCat © The Author(s) 2020. Published by Oxford University Press on behalf of The Society for Financial Studies. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
TI - Learning about Competitors: Evidence from SME Lending
JF - The Review of Financial Studies
DO - 10.1093/rfs/hhaa109
DA - 2020-09-10
UR - https://www.deepdyve.com/lp/oxford-university-press/learning-about-competitors-evidence-from-sme-lending-2gujqX2xLQ
SP - 1
EP - 1
VL - Advance Article
IS - 
DP - DeepDyve
ER -