Credit information sharing and cost of debt: evidence from the introduction of credit bureaus in developing countries

We investigate the effect of credit information sharing on cost of debt, with particular focus on the introduction of credit bureaus in developing countries. Using a large dataset of firms from 28 developing countries over the period 2004-2019, we find that firms’ average cost of debt significantly declines following the introduction of credit bureaus. This finding is robust to an alternative measure of cost of debt, several firm-and country-level controls and to firm-and year-fixed effects. The reduction in cost of debt is more pronounced for less transparent firms and for firms domiciled in countries with weak institutional framework.


INTRODUCTION
In the past two decades, motivated by the need to improve the availability of credit, several countries have implemented credit information-sharing schemes.Between 2004 and 2018, more than 75 developing countries instituted credit information-sharing schemes, and this arguably (partly) explains the burgeoning research on the economic consequences of credit informationsharing schemes (Ayyagari et al., 2021;De Haas et al, 2021;Martinez-Peria and Singh, 2014).For instance, Ayyagari et al. (2021) investigate the effect of credit information sharing on job growth in developing countries, while De Haas et al. (2021) study how information sharing affects the microcredit market in Bosnia and Herzegovina.We contribute to this strand of literature by examining the impact of credit information sharing on corporate cost of debt in developing countries, with particular focus on the introduction of credit bureaus. 1 We further investigate the moderating role of national institutional quality and firm opacity (i.e., lack of transparency).Due to the institutional voids that characterize most developing countries such as weak legal structures and contract enforcement mechanisms (Rajan and Zingales, 1998;Agyei-Boapeah and Machokoto, 2018;Tunyi et al., 2019), borrowers in such countries face a higher cost of credit and may, therefore, benefit from credit information-sharing schemes.
In theory, by sharing credit information about borrowers, lenders can expect to mitigate several lending problems including adverse selection, borrower hold-up, and agency costs which results in lower default rates in credit markets (e.g., Klein, 1992;Pagano and Jappelli, 1993;Vercammen, 1995;Padilla and Pagano, 2000).Credit information sharing gives banks easy access to information about the credit worthiness of new and potential borrowers to make safe lending decisions and to effectively monitor borrower behavior.Existing evidence, rarely based on developing countries, supports this view by documenting that countries with functional credit information-sharing schemes experience improved access to credit and reduced default rates (e.g., Brown, Jappelli, and Pagano, 2009;Houston, Lin, Lin, and Ma, 2010;Fosu, Danso, Agyei-Boapeah, Ntim, and Adegbite, 2020a).
Other studies extend this literature by exploring the relationship between credit information sharing on the one hand and credit constraints (e.g., Love and Mylenko, 2003;Brown et al., 2009;Martinez-Peria and Singh, 2014) or credit intermediation cost (Fosu et al., 2020b) on the other hand.Despite the incremental contributions made by prior studies to this burgeoning literature, they have often focused on either bank-level outcomes or firms' perceived credit constraints.Thus far, there has been limited focus on firm-level observable outcomes.We seek to add to the literature by linking credit information sharing to the firm's actual cost of debt, measured at both real and nominal rates.
Linking credit information directly to the firm's actual cost of debt is important for at least two reasons.First, most firms in developing countries depend on bank loans (Gwatidzo and Ojah, 2014), so the cost of debt can be a critical constraint that affects firm growth and economic growth.
Therefore, both managers and policymakers tend to be keenly interested in the drivers of the corporate cost of debt which makes the relationship between information sharing and cost of debt a policy-relevant issue.Second, the firm's actual cost of debt can easily be analyzed to permit a more robust and objective analysis that is devoid of the subjective perceptions of firms' constraints in Brown et al. (2009) and Martinez-Peria and Singh's (2014) survey measures.Besides directly linking information sharing to firms' cost of debt, we also draw from the institutional and information asymmetry theoretical perspectives to contend that our primary hypothesis could be moderated by country-level institutional quality and firm-level transparency.
Our work relates to but also differs from the work of Martinez-Peria and Singh (2014).First, the authors of the prior study examine how credit information-sharing systems impact firms' access to finance and cover the interest rates of firms' most recent loans in their analysis.
Therefore, their analysis considers the interest rates on specific loans taken by firms and fails to capture the average cost of debt for firms that access multiple loans over a period from the same or different financiers.We overcome this limitation by examining the average cost of debt of firms2 , defined to include loans from all sources (banks as well as private and public lenders) that the firm has accessed.Next, we cover a more recent sample period, 2004-2019 (compared to 2002-2014 in the prior study), which enables us to capture the impact of information-sharing schemes introduced after 2009.Thus, our analysis includes recent schemes introduced in countries such as Ghana, Indonesia, Jamaica, Jordan, Malawi, Nigeria, and Vietnam.Further, our reliance on accounting data published in the annual accounts (vis-a-vis survey data) provides us with a larger panel dataset for robust statistical analysis.For instance, due to the survey data limitation, Martinez-Peria and Singh (2014) could not control for firm-fixed effects for most their firms, which heightens omitted variable concerns in their empirical analysis.Our panel dataset helps us to mitigate such econometric concerns.Lastly, we extend the prior study by investigating how the relationship between information-sharing systems and corporate cost of debt might vary with (1) country-level institutional quality; and (2) firm-level information asymmetry.Overall, we provide a more focused, richer, and statistically robust analysis of the relationship between informationsharing schemes and corporate cost of debt.
Examining the relationship between information-sharing schemes and corporate cost of debt is complicated, not least because of endogeneity concerns.For example, there are potentially many factors that could affect both credit information sharing and firms' cost of debt.We mitigate this identification concern by employing the introduction of a credit bureau as an exogenous shock to the credit market.We follow an approach similar to Ayyagari et al. (2021) and Martinez-Peria and Singh ( 2014) by utilizing a difference-in-difference (DID) technique to estimate the impact of credit bureau introduction on the corporate cost of debt, by comparing countries that introduced credit bureaus, and those which did not, and the years pre-and post-introduction.We focus on credit bureaus because these are private commercial institutions that facilitate the exchange of credit information and provide associated services such as credit scores for lending purposes (Jappelli and Pagano, 2002).On the other hand, credit registries are publicly owned institutions and usually focus on regulatory and supervision requirements and typically cover loans above a specified threshold, which may limit the number of borrowers covered (Jappelli and Pagano, 2002).This focus is also motivated by empirical findings that suggest the introduction of credit registries has no significant effect on firm financing (Martinez-Peria and Singh, 2014).
Using the DID approach and a large dataset of over 7,200 firms from 28 developing countries over the 2004-2019 period, we find a significant reduction in firms' cost of debt by an average of 3.2 to 3.8 percentage points following the introduction of credit information-sharing systems (specifically, credit bureaus).This result is robust to a battery of robustness tests, including controlling for firm-, industry-, country-and year-fixed effects and using alternative measures of cost of debt.Further, we find that the introduction of a credit bureau has an asymmetric effect on firms located in countries that differ in terms of institutional quality.Specifically, introducing a credit bureau appears to be associated with a greater reduction in the cost of debt for firms that operate in countries with poorer institutional quality (i.e., low rule of law, high corruption, and poor regulatory quality).Our results imply that effective credit information-sharing schemes could, to some extent, substitute for the roles played by developed governance institutions.In other words, where national institutions are weak and underdeveloped, credit information-sharing schemes can deliver the greatest benefits to firms by way of significantly reducing their cost of debt.Finally, our analyses reveal that (1) less transparent firms tend to enjoy greater reductions in their cost of capital following credit bureau introductions; and (2) introducing credit bureaus tend to significantly increase firms' debt maturity.
We make the following contributions to the literature.First, we are one of the early studies to provide evidence of a direct and robust link between the introduction of a credit bureau and firms' actual cost of debt for developing countries.Thus, our paper is aligned with the work of Brown et al. (2009), Martinez-Peria and Singh (2014) and Fosu et al. (2020b) who show that credit information sharing alleviates credit constraints.Based on survey data collected from firms in transition countries, Brown et al. (2009) investigate the effect of credit information sharing on how firms perceive cost of credit to be a major business obstacle.We add to Brown et al.'s (2009) work by examining how credit information sharing impacts a firm's actual (observable) cost of debt measured from information contained in the firm's financial statements.Another related study is Fosu et al. (2020b), which examines the relationship between credit information sharing and banks' credit intermediation cost (i.e., the bank's loan spread).While Fosu et al. (2020b) consider the issue from the lender bank's perspective, we take the perspective of the borrower firm by focusing on corporate cost of debt.Overall, our empirical measure of cost of debt gives us the ability to explore how credit information-sharing schemes directly impact a specific attribute of the borrower firm -the cost of debt -a crucial variable for managerial decisions.
An additional contribution that we make to the literature is the new insights we offer by showing how country-level institutional quality (and its dimensions such as control of corruption, rule of law, regulatory quality, etc.) shape the relationship between credit information sharing and cost of debt.This aspect of our study has the potential to guide national policy formulation and evaluation.Further, our paper adds to the work of Francis et al. (2005) who examine the relationship between information asymmetry and risk pricing.The authors find that information uncertainty as measured by accounting information quality is related to firms' cost of debt.Our finding corroborates their results by suggesting that improving the credit information environment at the macro-level through information-sharing schemes helps to reduce the cost of debt for firms.
Finally, our paper covers a larger sample of firms from 28 developing countries, a context that is not only under-researched but has firms that are faced with exorbitant costs of credit, resulting in acute financing constraints.Thus, our study offers guidance to policymakers in developing countries with responsibility for managing credit constraints and the information environment.
The rest of the paper is organized as follows: Section 2 provides a review of relevant literature and proceeds to develop the hypotheses.Section 3 describes the data and empirical methods.We present and discuss the empirical results in Section 4, and Section 5 presents our conclusions.

Information sharing and cost of debt: theory and core hypothesis
A bank's role as a lender and delegated monitor poses some challenges.In discharging their roles, banks determine which borrowers are "safe" to lend to and which borrowers pose a risk.Additionally, they decide the level of monitoring required after a loan is extended to a borrower.The borrowers must prove their creditworthiness or face a higher borrowing cost.In fact, these challenges are exacerbated by information asymmetry in credit markets which encourages moral hazards and adverse selection (e.g., Jensen and Meckling, 1976;Myers, 1977).
Against this backdrop, the theoretical implication is that any improvement in the information environment such as a credit information exchange between banks will help to improve lending decisions by mitigating information asymmetry and moral hazard problems.Given lower levels of information asymmetry and moral hazard problems in the credit market, borrower firms could enjoy a lower cost of debt.
Existing theoretical perspectives suggest three major channels through which lower cost of debt could occur when banks share credit information.First, credit information sharing is hypothesized to reduce adverse selection problems and to help banks engage in "safe" lending (Pagano and Jappelli, 1993).According to this view, lending banks tend to only have relevant information about the pool of their local borrowers, which makes it difficult for them to safely lend to non-local borrowers.Consequently, the banks may decline to lend to non-local applicants or lend to them at a higher interest rate.However, charging a higher cost of credit ends up pricing out safe borrowers from the market (e.g., Stiglitz and Weiss, 1981).Pagano and Jappelli (1993) posit that credit information sharing can help banks obtain access to the credit history of non-local borrowers to make sound lending decisions.Accordingly, this perspective predicts that under credit information-sharing regimes, there would be more benefits to borrowers with good credit history, while borrowers with poor credit history will enjoy little or no benefits.
Second, credit information sharing implicitly constitutes an effective monitoring mechanism which reduces moral hazards and increases borrower effort to repay loans (Klein, 1992;Vercammen, 1995;Padilla and Pagano, 2000).Typically, borrowers, after securing credit, engage in riskier activities due to the challenges associated with delegated monitoring in high information asymmetry environments.Credit information sharing improves the information environment and enables borrower behavior to be observed by several banks, thus encouraging borrower repayment efforts.This is because borrowers would not want to be classified as "poor borrowers" and suffer reputational damage.In the end, good borrower behavior should reflect in a lower cost of credit for borrowers.
Finally, and related to the earlier channels, credit information sharing reduces opportunistic borrowing in multiple bank relationships.When credit information is not shared, borrowers have incentives to over-borrow from multiple banks as their overall level of indebtedness and their true credit risk cannot be accurately determined (Bennardo et al., 2015).This situation leads to credit rationing and higher costs of lending.However, credit information sharing has the potential to reduce this agency problem and reduce the cost of debt.
Overall, the theoretical channels and the related empirical evidence suggest that credit information sharing improves lending decisions, reduces loan default, and expands access to credit.
To the extent that the cost of debt is a good gauge of a firm's access to credit, we expect credit information sharing to significantly reduce a firm's cost of debt.Hence, we hypothesize as follows: Hypothesis 1: Credit information sharing reduces the cost of debt for firms.

Information sharing and cost of debt: potential heterogenous effects
Our primary hypothesis of credit information sharing leading to reduced cost of debt may be an oversimplification.Institutional quality (e.g., rule of law, corruption, regulation, etc.) could be a crucial moderator as it determines the monitoring efforts required by banks and the lender confidence in applicable rules and regulations (e.g., La Porta et al., 2002;Qian and Strahan, 2007;Demirgüç-Kunt, and Maksimovic, 1998;Rajan and Zingales, 1998;Raddatz, 2006;Kroszner, Laeven and Klingebiel, 2007).Thus, the institutional quality could impact the severity of the levels of a firm's financial constraints.We extend the literature to incorporate possible heterogeneity from institutional quality on our primary hypothesis.
It is plausible for the benefit of credit information sharing to be limited in countries with better governance institutions.This is because stronger institutions mitigate agency problems, and improve firm valuation (e.g., Shleifer and Vishny, 1997;La Porta et al., 2002), which leads to better access to finance (see La Porta et al., 1997Porta et al., , 1998;;Qian and Strahan, 2007;Driss et al., 2023).Qian and Strahan (2007) show that firms in countries with strong creditor protection are much more able to access loans with longer maturity and lower cost.The access to long-maturity loans under environments with better governance institutions suggests less need for costly monitoring (Diamond, 1984).Other studies suggest that institutional quality and financial development are interlinked.For example, for a sample of 49 countries, La Porta et al. (1997) show that countries with weaker institutions have a relatively less developed financial system to support corporate financing.In addition, Acemoglu et al. (2003) show that better institutions could serve as a substitute for financial development.They support their conclusion by demonstrating that although financial development is associated with value creation, this relationship disappears after controlling for the quality of institutions.
Based on the foregoing discussions, we expect the benefit of credit information sharing to be less pronounced in environments where the institutions provide better protection for creditors.Our expectation is consistent with the finding in Djankov et al. (2007) that shows that while credit information sharing is associated with increased credit availability, the effect is stronger in countries with weaker protection for creditors.Accordingly, we formulate our second hypothesis as follows: Hypothesis 2: The negative relationship between credit information sharing and cost of debt is moderated by institutional quality.

DATA AND METHODOLOGY
We gather data from several sources including World Bank (2021a) Doing Business, World Bank (2021b) World Development Indicators, World Bank (2021c) World Governance Indicators and COMPUSTAT Global to test the relationship between credit information sharing and cost of debt.In this section, we discuss the data sources and the empirical implementation of the hypotheses.

Data and measurement of key variables
We obtain firm-level financial data covering all developing countries from the COMPUSTAT Global database over the period 2004 to 2019.We exclude all firms with standard industry classification (SIC) codes ranging from 6000 to 6999 (financial firms) and from 4900 to 4999 (utility firms) as they are heavily regulated.We retain all firms with non-missing values on key variables.We merge the resulting sample with data on credit information sharing from World Bank (2021a) Doing Business database, macroeconomic data from World Bank (2021b) World Development Indicators, and governance data from World Bank (2021c) Worldwide Governance Indicators.This leaves us with an initial sample of 8,635 unique firms form 37 countries.We require all firms to have at least two consecutive observations on the key variables and each country to have at least 15 firm-year observations.This leaves a final sample of 7,269 unique firms from 28 developing countries over the period 2004 to 2019.Our sample begins from 2004 because the data on credit information sharing starts in 2004.

Cost of debt
We follow the extant literature (Francis, Khurana and Pereira, 2005;Bliss and Gul, 2012;Chui, Kwok and Zhou, 2016;Fungáĉová, et al., 2017) and measure cost of debt (CoD) as the ratio of contemporaneous interest expense to average total debt, where the average is over the current (year t) and the previous year's (year t-1) debt.Specifically, we have: This measure of cost of debt represents the implicit interest rate paid by firms on their debt.This measure is more appropriate for assessing the overall link between credit information sharing and a firm's borrowing cost due to its ability to capture debt capital obtained from banks as well as from other private and public sources.Studies relying on only loan-level data from banks arguably present a partial view of firms' cost of debt (e.g., Chui et al., 2016).To mitigate the effect of extreme outliers, we follow Wang, Han and Huan, (2020) and winsorize this variable at the top and bottom five percent.
Due to the significant variation in inflation among the sample countries, we follow Chui et al., 2016) and adjust the cost of debt variable (CoD) for inflation () to obtain real cost of debt (RCoD) as follows as: Consequently, we use the real cost debt as our main measure of cost of debt and the nominal measure of cost of debt for robustness checks.

Credit information sharing
We relate our measures of cost of debt to credit information sharing and control for a wide array of firm-and country-level features.The World Bank (2021a) Doing Business database compiles data on the extent and type of credit information on borrowers that financial and nonfinancial institutions collect and share.We adopt the introduction of private credit bureau (credit bureau introduction) as our measure of credit information sharing.In a later analysis for robustness, we also consider the introduction of credit registry as an alternative proxy for improvement in credit information sharing.We do not use other measures, such as the depth of credit information sharing or the coverage of credit bureaus, due to the endogenous nature of these variables, especially when considered together with some general improvements in institutional quality.

National institutional quality and control variables
We also relate our analysis to country-level institutional quality.We follow the existing studies (Tong and Wei, 2011) and measure institutional quality as the simple average of six institutional features of each country, namely, voice and accountability, political stability and absence of violence/terrorism, government effectiveness, regulatory quality, rule of law, and control of corruption.A firm is deemed to have a low level of, or weak, institutional quality when the institutional quality index is below the median country.
Further, we control for other salient firm-level features that could affect cost of debt.Consistent with previous studies (e.g., Bauwhede et al., 2015;Chui et al., 2016;Fungáĉová, et al., 2017;Bardos et al., 2021), we include the size of the firm (Firm Size), measured as the natural logarithm of the firm's assets.Large firms have lower risk of failure (e.g., Berger and Udell, 1995).They may also be older and, thus, have a longer history to meet the disclosure requirement, making them less opaque (Bonaccorsi di Patti, 2004;Bonini et al., 2016).We also control for the proportion of a firm's tangible assets to total assets (Tangibility) because firms with a higher degree of tangible assets are more transparent (Bonaccorsi di Patti, 2004) and more able to meet collateral requirements and, consequently, have better access to external finance (Fungáĉová, et al., 2017).
Next, we include the firm's liquid assets and cash.Firms with a larger cash balance and those with high liquidity are more likely to meet day-to-day financial obligations and are perceived to be more financially stable.Therefore, these firms are expected to have a lower cost of debt.We measure Liquidity as the proportion of the firm's current assets less stock scaled by current liabilities.We measure Cash as the ratio of the balance of cash to total assets.More profitable firms are perceived to be more efficient and are, therefore, expected to have lower cost of debt.We control for profitability, measured as the ratio of pre-tax profit to total assets (Profitability).
Finally, we control for industry concentration based on the Herfindahl-Hirschman index (HHI), the sum of the squared market shares (by assets) at the three-digit industry standard classification level.At the country level, we also include gross domestic product (GDP per capita) and inflation to control for macro-level differences that can influence cost of debt.Appendix A contains detailed definitions of all variables used in the study.

Empirical methods
We investigate the relationship between credit information sharing and cost of debt by estimating a series of panel fixed effect models with several control variables.We take advantage of the time and cross-sectional dimensions of our data to minimize endogeneity problems arising from the omission of time invariant unobservable firm characteristics.We specify the following baseline equation: where    is an indicator variable with a value of one for countries in the years after the introduction of an operational credit bureau (i.e., where the coverage of the credit bureau as a proportion of the adult population is strictly positive), and zero for the prior years and for countries that either never introduced or always had a credit bureau throughout the sample period.Our definition of credit bureau (CB) introduction and empirical approach is in line with Ayyagari et al. (2021).In Eq. ( 3),  is a set of firm-, industry-and country-level control variables discussed in subsection 3.1;  is year fixed-effect;  represents firm fixed-effect;  is an error term assumed to be independent and identically distributed with mean zero and variance   2 .The inclusion of firm fixed-effect controls for potential omitted variable bias while the year fixed-effect controls for time invariant factors at the country level.A negative and statistically significant value of  will be consistent with our primary hypothesis, H1.
While we address potential endogeneity arising from omitted variables through the inclusion of year fixed-effect () and firm fixed-effect (), we take a step further to address the potentially endogenous nature of the decision to introduce a credit bureau.Specifically, we match countries that introduced a credit bureau (taken as the treatment group) with those that either never introduced or always had a credit bureau during the sample period (taken together as a control group) and re-estimate our regressions based on the matched sample.We match these countries on the pre-adoption three-year average measures of GDP per capita, real GDP growth and regulatory quality as in Ayyagari et al. (2021) based on the propensity scores and nearest neighbor one-toone matching with replacement and common support.We then invoke the difference-in-difference estimation previously described based on the matched sample.
We hypothesized earlier that the impact of credit information sharing on cost of debt is likely to be higher for firms in countries with weaker institutions.To test this hypothesis (H2), we modify our baseline equation to include a relevant interaction term, as follows: To support H2, we expect the coefficient on the interaction term between credit bureau introduction and Low institutional quality () to be negative and significant.

Descriptive statistics
Table 1 displays the descriptive statistics.The average real (nominal) cost of debt for a firm in our sample is 24% (31%).There is a substantial variation in the real (nominal) cost of debt as the standard deviation of 46% (49%) suggests.The credit bureau introduction ranges from zero to one, with an average value of 0.38.The average financial development score is 0.86 and the average institutional quality value is -0.34, which suggests that the average country in our sample is below the world average for institutional quality.The average firm has a size of 8.71, asset tangibility of 0.37, liquidity of 1.06, cash of 0.08 and profitability of 0.02.
Table 2 highlights some key observations across our sample countries.Cost of debt is highest in Ghana (52%), Vietnam (47%), Malawi (43%), Bangladesh (42%), Egypt (42%), Namibia (42%) and Zambia (42%).Vietnam introduced operational credit bureau as late as 2014 while Bangladesh never introduced a credit bureau during the sample period.Further, all these countries, except Namibia, have significantly lower levels of institutional quality.Notably, out of the 28 countries in our sample, 15 introduced credit bureaus during the sample period.We also observe a drop in cost of debt for most of the countries (see, cost of debt differentials in Table 2).

[INSERT TABLE 1 HERE] [INSERT TABLE 2 HERE]
The correlation matrix in Table 3 shows that the correlation among our variables is low, with most coefficients falling below 0.5.However, a few correlation coefficients are high.For example, the correlation between nominal cost of debt and real cost of debt is high, but these are alternative measures of cost of debt that do not enter the models simultaneously.There is also a high correlation between credit bureau introduction and some country-level variables (e.g., institutional quality and financial development).There also seems to be a high correlation among the other country-level variables (e.g., GDP growth, financial development, and institutional quality).To mitigate multicollinearity concerns, we run alternative models with and without these variables, and these alternative models do not change our conclusions.

Baseline results
We report the results for the test of H1 in Table 4.The dependent variable in all the models is real cost of debt, and the independent variable is Credit bureau (CB) introduction.In Model 1, we only control for firm and industry level variables and firm-and year-fixed effects.Models 2-4 sequentially add country-level control variables to assess the impact of multicollinearity since some of these variables are moderately correlated.We present the full model that controls for firm-, industry-, and year-fixed effects in Model 5 of Table 4. Models 6 and 7 of Table 4 replicate Models 4 and 5 of Table 4, respectively, but replace the firm-fixed effect with country-fixed effect.
As we show in Table 4, our proxy for credit information sharing (Credit bureau introduction) is negative and statistically significant at the 5% level or lower throughout the models.The introduction of credit bureau is associated with a reduction in firms' cost of debt by between 3.2 and 3.8 percentage points.Given that the average firm has a cost of debt of 24% in real terms, this effect is economically significant (being as high as a 15.8% reduction for the average firm). 3  Overall, our findings are consistent with H1 and imply that credit information-sharing schemes 3 The reduction in cost of debt is expressed as a percentage of the average cost of debt.
help to reduce information asymmetry and improve access to credit.
The control variables offer some insight into the determinants of cost of credit in developing countries.First, the coefficients of Firm Size and asset Tangibility are negative and statistically significant, which suggests that larger firms and firms with more tangible assets are more likely to have lower cost of debt.Also, firms with more liquid assets (Liquidity), a larger cash balance (Cash), and higher profitability are more likely to have lower cost of debt.Further, firms in more concentrated industries are likely to have a higher cost of debt, possibly due to the view that these firms are less efficient and riskier (e.g., Bolton and Scharfstein, 1990;Chevalier, 1995); however, this effect is statistically insignificant for the models controlling for country-fixed effect.
Concerning the country-level variables, we find that better institutional quality and financial development are associated with a lower cost of debt.

The moderating effect of institutional quality (H2)
We present the results for the tests of our Hypotheses 2, which posits that institutional quality would moderate the negative relationship between cost of debt and credit information sharing, in Table 5.In Model 1 of Table 5, we interact Credit bureau (CB) introduction with an indicator variable for low level of institutional quality (Low Institutional Quality).The coefficients on Credit bureau introduction are negative but statistically insignificant; however, the coefficients on the interaction term between Credit bureau introduction and Low Institutional Quality are negative and statistically significant at the 1% level.This implies that the average firm in a country with a low level of institutional quality will experience a drop of 6.5 percentage points in its cost of debt following the introduction of credit bureau in its country of domicile.This effect falls to 2.5 percentage points for a similar firm in a country with a high level of institutional quality.This finding suggests that information-sharing schemes deliver the greatest benefits in environments of low institutional quality where institutional voids may heighten information asymmetry and moral hazard problems.Thus, better institutional governance and credit information-sharing schemes seem to be close (but not perfect) substitutes.Collectively, our results presented in this subsection are in line with Djankov et al. (2007) who show that while credit information sharing is associated with increased credit availability, the effect is stronger in countries with weaker protection for creditors.
In Models 2-7 of Table 5, we take a closer look at our results in the hope to provide deeper insight into the specific institutional arrangements that may substitute for credit informationsharing schemes.As noted earlier, our measure of institutional quality is a composite measure derived as the simple average of six different institutional variables.The variables are voice and accountability, political stability and absence of violence, government effectiveness, rule of law, regulatory quality, and control of corruption.We present results that replace the indicator variable for low level of the composite institutional quality measure with a similar indicator variable for the individual component variables.Since the individual variables are highly correlated, we do not include all the six variables in a single regression.The results show that the moderating impact of institutional quality that we reported earlier is driven by only three of the component variables - (1) regulatory quality, (2) control of corruption, and (3) government effectiveness.It seems that these three variables can shape information asymmetry and moral hazards to alter lending decisions and can ultimately affect cost of debt even in the absence of credit information-sharing schemes.

The effect of firm-level information asymmetry
While we document that the introduction of credit bureau reduces firms' cost of debt especially for countries with lower institutional quality, we need to consider the potential effect of firm-level heterogeneity.We address this issue by looking at the potential moderating role of firm-level information asymmetry.The role played by firm-level information asymmetry (or opacity) in alleviating financial constraints is well documented in the literature stream.Information asymmetry reflects the amount of effort banks need to exert to assess or verify a borrowing firms' credit worthiness (Boot and Schmeits, 2000;Bonaccorsi di Patti and Dell'Ariccia, 2004).At the same time, information asymmetry exacerbates the conflict of interest between shareholders and creditors (Liao, Chen and Lu, 2009;Fosu et al, 2016).Consequently, opaque (i.e., less transparent) firms are perceived as riskier as their quality and risk choices cannot be easily and readily determined.
Further, monitoring may not work as an effective tool to discipline high information asymmetry (less transparent) firms as the quality of their assets and operations cannot be effectively observed.The empirical literature supports the view that less transparent (opaque) firms are riskier (Beck, Demirgüç-Kunt, and Maksimovic, 2008;Bushman and Williams, 2012;Nier, 2005;Nier and Baumann, 2006;Kroszner, Laeven and Klingebiel, 2007;Fosu et al., 2017).For instance, Kroszner et al. (2007) show that firms with opaque assets experience a larger decline in growth relative to their less-opaque counterparts during a crisis.Consequently, opaque firms tend to have higher financing costs relative to their less-opaque counterparts (Duffie and Lando, 2001;Lu, Chen and Liao, 2010).Hence, we conjecture that the benefit of credit information sharing should be larger for opaque firms.We test this conjecture by interacting Credit bureau introduction with an indicator variable for less-transparent firms (Less Transparent), which takes a value of one where firm's level of transparency is below the median firm's level, and zero otherwise.We measure a firm's level of transparency (as an inverse measure of information asymmetry) using the principal component factor of asset Tangibility, Firm Age and Firm Size.We present these results in Table 6.
In Model 1 of Table 6, the coefficient on the interaction term of Credit bureau introduction with Less Transparent firms dummy variable is negative and significant.The findings support our conjecture that the reductions in cost of debt following the introduction of credit bureau is greater (lower) for less (more) transparent firms.That is, firms with higher (lower) levels of information asymmetry appear to benefit more from credit information-sharing schemes.Further, we explore the moderation effect of the disaggregated components of our measure of firm-level transparency to identify the crucial factors driving the results.Specifically, we interact Credit bureau introduction with Small Firms, Young Firms and Low Tangibility dummy variables (which takes a value of one for firms whose total assets, age and proportion of tangible assets, respectively, are below the median firm's and zero otherwise).The results are presented in Models 2-4 of Table 6.
The coefficients on Credit Bureau introduction and the interaction terms with Small Firms and Young Firms (Models 2 and 3) remain negative, the impact on cost of debt is statistically significant only through the interaction terms.Whilst the coefficient on Credit Bureau introduction remains significantly negative, the coefficient on the interaction term with Low Tangibility is surprisingly significantly positive.Overall, the results essentially indicates that small firm and young firms experience 5.8 and 5.4 percentage point additional reduction in cost of debt respectively following the introduction of credit bureaus, whilst firms with lower level of tangible assets experience relatively higher cost of debt compared with other firms.

Credit information sharing and credit demand
So far, our analysis indicates that firms experience a significant reduction in their cost of debt following an introduction of credit bureau in the countries within which they operate.It is plausible that that the reduction in the cost of debt and/or any associated increase in credit supply (due to information sharing) will lead to an increase in the volume and term structure (i.e., maturity) of loans extended to the firms.We test this conjecture by replacing cost of debt with firm-level financial leverage (the ratio of total debt to total assets) and the debt maturity (the ratio of longterm debt to total debts).We present the results in Table 7. Model 1 (the leverage regression) shows that the coefficient on Credit bureau introduction is positive but statistically insignificant.
In Model 2 (the debt maturity regression) the coefficient on Credit bureau introduction is positive and statistically significant.These results suggest that although firms in developing countries that introduce credit bureau systems do not experience an increase in their leverage ratios, they seem to have a 2.3 percentage point increase in the maturity of their loans (i.e., the proportion of their long-term debt to total debt).

Alternative dependent variable and subsample analysis
We execute further tests to ensure robustness of our key findings.The prior analyses are based on the real cost of debt which directly adjusts our nominal cost of debt variable for inflation, and hence, does not include the inflation variable as a regressor.Further, our main analysis has so far included countries that have had credit bureaus throughout the sample period as part of the control group in our empirical set-up.In this section, we conduct additional analysis to assess the robustness of our results by first replacing the dependent variable with the nominal cost of debt and, second, by excluding the countries that had credit bureaus throughout the sample period.We conduct further analysis of the baseline results based on a propensity score matched sample.
Table 8 presents the results of the set of regressions with nominal cost of debt as the dependent variable.Model 1 presents the results for the baseline model, and Models 2 and 3 present the results that include the interaction terms of Credit bureau introduction with dummy variables for low institutional quality and less transparent firms.The coefficient on Credit bureau introduction remains negative and significant in Model 1, which confirms our baseline results that suggest that the introduction of credit bureaus leads to a reduction in the cost of debt.Also, the sign and significance of the coefficient on the interaction terms in Models 1 and 2 are consistent with our earlier results, which confirms our conjecture that firms in countries with low institutional quality and less transparent firms have greater benefits from the introduction of credit bureaus.

[INSERT TABLE 8 HERE]
We report the baseline results that exclude countries that had credit bureau throughout the sample period in Table 9.Therefore, in this analysis, countries that introduced credit bureaus during our sample period were strictly compared with those countries that never had credit bureaus.
While this analysis has a relatively smaller sample, it offers a cleaner test of the effect of credit information sharing on corporate cost of debt, since those countries that had credit bureaus in place before our sample start date were excluded.The results we report in Table 9 are similar to our baseline results in Table 4, except that the coefficients are larger in Table 9.This indicates that while our results are robust, the original analysis biases our estimates downward, hence, our conclusions are based on conservative estimates of the benefits of introducing credit bureaus.

[INSERT TABLE 9 HERE]
Additionally, we conduct an analysis that utilizes control firms in similar countries that did not introduce credit bureaus during our sample period.This analysis involves a difference-indifference estimation based on a (country-level) propensity score matched sample, and we report the findings in Table 10.We matched the sample countries that introduced credit bureaus to control countries based on propensity scores,4 and we use one-to-one nearest neighbor matching with replacement and common support.Overall, the results based on the matched sample are qualitatively similar to the results based on the unmatched (full) sample.The introduction of bureaus is associated with a 4.4 percentage point reduction in the cost of debt, and this effect is significantly higher for firms in countries with weak institutional quality and for less transparent firms.

[INSERT TABLE 10 HERE]
Our focus has so far been on credit bureaus because of their distinctive feature of facilitating the exchange of credit amongst lending institutions in developing countries.Being private sector organizations, the credit bureaus tend to be more efficiently run than the public sector-managed credit registries.This notwithstanding, we extend our analysis to cover the introduction of credit registries for robustness purposes.We present the results in Table 11.While the coefficients on credit registries have a negative sign, they are statistically insignificant in all the firm-fixed effect models and only marginally significant in the models that exclude the firm-fixed effects.Overall, we find mild evidence of the introduction of credit registries resulting in reduced cost of debt in developing countries.This finding is consistent with Martinez-Peria and Singh (2014) who report that the introduction of credit registries has no significant effects on firm financing.

Additional robustness tests
Finally, we explore the causal effect of the introduction of credit bureaus on cost of debt by assessing the parallel trends assumption underlying the difference-in-difference estimates and the implications of the staggered nature of the timing of the introduction of credit bureaus in our sample.The parallel trends assumption suggests that the cost of debt in the countries that introduced a credit bureau (treated group) and in the countries that did not (control groups) would have followed a parallel path in the absence of the introduction of credit bureaus.This helps to rule out the possibility that our results are simply mimicking differential trends in the cost of debt prior to the introduction of credit bureaus in the treated countries (see, Baker, Larcker and Wang, 2022).
Traditionally, the parallel trend assumption has been tested using event-study style analysis (Baker et al., 2022).However, recent advances in the econometric literature suggests that this approach could be biased in the presence of dynamic and heterogenous treatment effect (Sun and Abraham, 2021;Callaway and Sant'Anna 2021;Goodman-Bacon, 2021;Baker, Larcker and Wang, 2022;Borusyak, Jaravel, and Spiess, 2022).In particular, the average causal effect we document in a staggered difference-in-difference setup could be seen as a weighted average of all possible 2 x 2 (i.e., two-group and two-period) difference-in-difference estimate based on comparing (1) the countries that introduced a credit bureau during the sample period to the countries that did not (treated vs. never treated), ( 2) the countries that introduced a credit bureau earlier to the countries that had not yet introduced a credit bureau (early treated vs later controls), and (3) countries that introduced a credit bureau later to the countries that had already introduced credit bureau (later treated vs. earlier controls). 5As noted in Goodman-Bacon ( 2022) and Baker et al., 2022), the third group can result in the "bad comparison" or "forbidden comparison" problem as units treated earlier serve as control units in the difference-in-difference estimation, which can attenuate the average estimated effect when the treatment effect is dynamic.
To address this potential attenuation problem arising from the different weights and dynamic effects associated with staggered timing of the introduction of credit bureau in our sample, we employ the novel interaction-weighted estimator by Sun and Abraham (2021).We estimate a difference-in-difference model accommodating relative time and group indicators and, thus, yielding interaction-weighted cohort-specific average treatment effect (CATT). 6We present the results of this analysis in Figure 1.Overall, our inference remains unchanged.First, the results show that the pre-trend coefficients of the cost of debt are largely not statistically different from zero, with the only significant point noted in the relative time 12 years away from the year of introduction of credit bureau.The insignificant pre-trend coefficients in Figure 1 suggest that the parallel trends assumption holds.Next, the results show that the cost of credit declines sharply after the introduction of the credit bureau, and this decline continues through to the fourth year afterwards and almost flattens out the following year.

CONCLUSION
We rely on a large dataset of over 47,000 firm-years from 28 developing countries to examine the impact of credit information-sharing schemes on firms' cost of debt, with a particular focus on the introduction of credit bureaus in developing countries.Due to reduced information asymmetry and lower moral hazard problems that could result from the implementation of credit informationsharing schemes, firms in developing countries (a context that faces acute financial constraints) are likely to benefit from a lower cost of debt.We find robust empirical evidence in support of this hypothesis.In particular, the introduction of credit bureau in a country is associated with a reduction in firms' cost of debt by between 3.2 and 3.8 percentage points.Given that the average firm has a cost of debt of 24% in real terms, this effect is economically significant, that represents a 13.3%-15.8%reduction in the cost of debt for the average firm.
We also find that the benefits of reduced cost of debt resulting from credit information-sharing schemes are significantly higher in countries with lower institutional quality and for less transparent firms.These results provide evidence to suggest that reductions in information asymmetry is potentially a major channel through which credit information sharing impacts corporate cost of debt.Overall, our key analysis leads us to conclude that firms in developing countries indeed enjoy a lower cost of debt when they are domiciled in countries that have functioning credit information-sharing schemes, especially when their countries are characterized by poor institutional quality (particularly, government effectiveness, lower regulatory quality, and high corruption).
Our work has implications for both corporate and national policy formulation.First, firms and countries seeking growth can mitigate financial constraints by instituting schemes that encourage lenders to share information about their borrowers among themselves.That is, to the extent that sharing borrower information reduces information asymmetry problems in credit markets, firms and countries can cut the cost of debt and enhance access to credit.Second, policymakers in countries that have strong governance institutions may not reap huge benefits from credit information-sharing schemes, since these schemes seem to be close (but not perfect) substitutes for institutional quality.
Despite making important contributions to the literature stream, our study is not exempt from the limitations that confront most studies based on archival data.It is possible that our empirical constructs and variables are measured with error.Therefore, our results and conclusions should be interpreted with caution.Future studies can utilize other research approaches including interviews and surveys to validate our findings.presents the sample countries and indicates whether a country introduced a credit bureau.Cost of debt differential is the difference between the latest period and earliest period cost of debt.Detailed definitions of all variables can be found in the Appendix.'No*' indicates that the country never introduced a credit bureau during the sample period.'No' indicates that the country always had a credit bureau during the sample period.'Yes' indicates that the country introduced a credit bureau during the sample period (with the year of the introduction stated in parentheses).

Figure 1 :
Figure 1: Event study of credit information sharing and cost of debt

Table 1
This table presents the summary statistics for the full sample.Detailed definition of each variable can be found in TableA1.

Table 3 :
This table presents the correlation among the variables used in our analysis.All variables are as described in the Appendix." a " represents significance at the percent level or lower.Estimation results for the effect of information sharing on cost of debt.All variables are as described in the Appendix.Standard errors robust to heteroscedasticity and clustering at the firm level are in parentheses.*, **, and *** indicate significance at 10%, 5% and 1%, respectively.

Table 5 :
Credit information sharing and cost of debt -moderating effect of institutional quality.Estimation results for the effect of information sharing on cost of debt.All variables are as described in the Appendix.Standard errors robust to heteroscedasticity and clustering at the firm level are in parentheses.*, **, and *** indicate significance at 10%, 5% and 1%, respectively.

Table 6 :
Credit information and cost of creditfirm level heterogeneity Estimation results for the effect of information sharing on cost of debt.All variables are as described in the Appendix.Standard errors robust to heteroscedasticity and clustering at the firm level are in parentheses.*, **, and *** indicate significance at 10%, 5% and 1%, respectively.

Table 7 :
The effect of credit information sharing on leverage and debt maturity Estimation results for the effect of information sharing on leverage and debt maturity.All variables are as described in the Appendix.Standard errors robust to heteroscedasticity and clustering at the firm level are in parentheses.*, **, and *** indicate significance at 10%, 5% and 1%, respectively.

Table 8 :
Credit information sharing and nominal cost of credit.

Table 9 :
Credit information and cost of credit -excluding countries that always had a credit bureau over our sample period.Estimation results for the effect of information sharing on cost of debt.All variables are as described in the Appendix.Standard errors robust to heteroscedasticity and clustering at the firm level are in parentheses.*, **, and *** indicate significance at 10%, 5% and 1%, respectively.

Table 10 :
Credit information and cost of credit -matched sample Estimation results for the effect of information sharing on cost of debt.All variables are as described in the Appendix.Standard errors robust to heteroscedasticity and clustering at the firm level at the firm level are in parentheses.*, **, and *** indicate significance at 10%, 5% and 1%, respectively.

Table 11 :
Credit information and cost of credit -Credit registry Estimation results for the effect of information sharing on cost of debt.All variables are as described in the Appendix.Standard errors robust to heteroscedasticity and clustering at the firm level are in parentheses.*, **, and *** indicate significance at 10%, 5% and 1%, respectively.