The Brazilian Bankruptcy Law Experiment

The Brazilian Bankruptcy Law Experiment Aloisio Araújo, Bruno Funchal and Rafael Ferreira Resumo In early 2005 a new bankruptcy law was approved by the Brazilian Congress, taking effect a few months later. The new legislation improved creditor protection and the bankruptcy system s efficiency. This paper tries to shed some light on the empirical consequences of a bankruptcy reform on a poorly developed credit market. Using data from Brazilian and non- Brazilian firms, we estimated, using three alternative models, the effect of the bankruptcy reform on contractual and non-contractual debt variables. In general, all the models yielded similar results. Concerning contractual debt variables, we found a significant increase in the total amount and longterm debt and a reduction in the cost of debt. For non-contractual debt variables, we found a reduction in the bank debt to public debt ratio, an increase in the number of domestic loan contracts and no impact on the number of foreign loan contracts. Keywords: Financing Policy; Bankruptcy; Law. JEL Codes: G32; G33; K2. 1 Introduction This paper examines the effects of changes in an important institution that regulates the corporate environment: the bankruptcy law. We analyze how changes in bankruptcy law design impact firms debt financing policy, focusing on the Brazilian experience with bankruptcy reform in 2005. Historically, the Brazilian credit market was not an easy source of firm financing. The supply of private credit was notoriously tight compared with developed or even other developing countries. Information from the World Bank indicates that the Brazilian ratio of private credit to GDP was low (0.35) compared not only to the OECD countries (1.02), but also compared to the average figure for Latin 1

American and Caribbean countries (0.44). 1 Moreover, an important part of the available credit came and still comes from the state-owned National Bank for Economic and Social Development, 2 which finances a large share of non-housing investments at subsidized interest rates. The interest rate spread confirms this situation: the Brazilian spread (49%) is more than four times larger than the average spread in Latin American countries (11%) and more than twelve times larger than the average for OECD countries (3.87%). One possible explanation for such a scenario is the obsolete procedures under the former Brazilian bankruptcy law. In Brazil, creditors historically had a low recovery rate from insolvent debtors, thus inhibiting the supply of credit and raising the interest rate spread. Until the new legislation came into effect, the recovery rate of Brazilian creditors in case of bankruptcy was only 0.2%, while the averages of Latin American and OECD countries were 26% and 72% respectively, according to Djankov et al. (2008). But how does one explain such low recovery rate in Brazil? The former Brazilian bankruptcy law allowed for both liquidation and reorganization, but it failed to protect creditors. It also did little to prevent the deterioration of the firm s value, mainly due to the so-called succession problem in liquidation, where liabilities (tax, labor and off balance-sheet liabilities) were transferred to the buyer of a liquidated asset. This feature increased the risk of a potential buyer and depreciated the value of the firm s assets. This deterioration of firms value combined with the low priority given to secured and unsecured creditors relative to tax and labor claims meant that both creditors and debtors used to get basically nothing at the end of the process. Also, the reorganization procedure did not intend to restructure economically viable firms. Its only concern was debt renegotiation, by postponing the payment and setting different terms. It encouraged debtors to use the reorganization procedure as a bargaining mechanism against creditors. Nonetheless, since debtors actually had little hope of getting anything at the end of the liquidation procedure, it also inhibited any potential strategic default. These features acted to eliminate creditors incentives to participate actively in both liquidation and reorganization, resulting in a low supply of credit and in extremely high interest rates. The reform of the Brazilian bankruptcy law was intended to change this scenario. 3 One of the most significant changes was to enhance the protection given to creditors, which was implemented through two channels: first, the new law 1 All figures refer to the 1997-2002 period. Source: World Development Indicator database 2004. 2 Banco Nacional de Desenvolvimento Econômico e Social (BNDES) in Portuguese 3 Gine and Love (2007) and Gamboa-Cavazos and Schneider (2007) studied the effects of bankruptcy reform in Colombia and Mexico, respectively. 2

increased their priority order to receive proceeds; and second, it allowed them to actively participate in the reorganization procedure. These changes are also expected to increase the efficiency of the bankruptcy system by reducing the cost and time of both reorganization and liquidation, leading to an increase in the value of distressed firms. Therefore, the new law not only makes creditors more likely to receive some of their credits, but it also probably increases the residual value of the bankrupt firm, elevating the creditors recovery rates. Both of these effects tend to have a positive impact on the credit markets. Several works have formalized theories on private credit. Townsend (1979), Aghion and Bolton (1992), and Hart and Moore (1994, 1998) showed that when lenders can more easily force repayment they are more willing to extend credit. Scott Jr (1977) theoretically addressed the relationship between capital structure and bankruptcy. When firms take on secured debt, they are not only giving a promise of future repayment, but also the right to the lender to be first in order of priority in case of bankruptcy. Thus, the priority order defined by bankruptcy legislation has a significant value since it reduces the chance that creditors will not be repaid. This value affects the supply of credit and the cost of debt capital and as a consequence firms financing policies. Diamond (2004) showed that in economic environments where lenders have passive behavior, i.e., where lenders do not seek the bankruptcy courts to enforce their rights after a borrower s defaults (like in Brazil), short-term debt can be an effective way to solve the lender passivity problem. The author shows that the existence of externalities across lenders in their refinancing decisions can lead them to a collective action against the firm s assets. 4 In cases like this, short-term debt disciplines borrowers, providing a commitment without which borrowers could only borrow much smaller amounts. 5 This result suggests that changes in the procedures that encourage lenders participation may lead to a decrease in the proportion of short-term debt relative to total debt. Bolton and Schaferstein (1996) analyzed what determines the number of creditors from which a firm borrows. They argued that the optimal choice of debt structure should simultaneously try to meet two potentially conflicting objectives: first, discourage management to default strategically; and second, to avoid too much loss of the firm s liquidation value. Although borrowing from many creditors serves the first objective, it also decreases the liquidation value if the firm defaults for liquidity reasons, resulting in a less efficient outcome ex ante. Also, they found that it is optimal for firms with low credit quality to maximize their liquidation values, by borrowing from fewer creditors. In an applied way, they interpreted the number of creditors as a trade-off between bank debt and public debt, since firms 4 See Jackson (1986) and Webb (1991). 5 In Brazil, the average portion of short-term debt in the capital structure of publicly traded firms was 85% in the pre-reform period, which is an evidence of precisely this effect. 3

usually have many public debt holders, but few banks. 6 Findings on credit market development stress the important role of legal protection of creditors and the efficiency of debt enforcement in supporting these markets (e.g., La Porta et al., 1997; La Porta et al., 1998; Djankov et al., 2007 and Djankov et al., 2008). Considering the effect of information on credit markets, Jappelli and Pagano (2000, 2002), Pagano and Jappelli (1993) and Sapienza (2002) showed the relevance of this factor in determining credit availability. The current empirical literature stresses that creditor protection through the legal system is associated with a broader credit market in a monotone way: the higher the protection to creditors, the more developed the credit market is (e.g., La Porta et al.,1997 and Djankov et al., 2007). In addition to creditor protection arguments, Djankov et al. (2008) showed that the efficiency of debt enforcement measured by its cost, time and the asset disposition predicts debt market development. Qian and Strahan (2007) and Bae and Goyal (2009) went deeper in analyzing the effects of creditors rights on debt. The latter authors analyzed how legal protection affects interest rates, size and loan maturities. They found that creditors rights have no impact on size and price, only on maturity. Qian and Strahan (2007) showed that for economies with stronger creditor s protection, loans have more concentrated ownership, maturities and lower interest rates. Like these authors, our goal is to analyze how changes in creditors rights and bankruptcy efficiency affect contractual debt variables such as size, maturity and price and non-contractual debt variables like number of loans in both domestic and foreign currency and the ratio of bank/public debt. All previous works have looked for cross-country correlations between creditors rights and financial policies. This approach tends to produce results that are sometimes hard to interpret, mostly due to the existence of several omitted factors in cross-country studies. In contrast, our paper focuses on a natural experiment that affected creditors rights and occured in only one country (Brazil). This narrower scope can work to our advantage, since the analysis of a specific reform makes it easier for us to identify of the causal effect of creditors protection on financial policies. Therefore, by comparing the same country in two different periods of time each period with a different law regulating bankruptcy procedures we can better control these omitted variables that pose a problem to cross-country studies. Also, we address the challenge of actually measuring creditor rights. We follow a quasi-experimental approach, taking advantage of the bankruptcy reform in Brazil as an experiment. We compare the behavior of debt related variables of Brazilian firms (our treatment group) with the behavior of the same 6 In Brazil, the average ratio Bank debt/public debt in the pre-reform period was approximately 7. 4

variables of firms from Argentina, Chile and Mexico (our control group). This approach helps to control our analysis for shocks in the credit market common to these countries in the period when the new Brazilian bankruptcy law was being implemented. We start our analysis by looking solely at Brazilian firms. From a panel data set with firms as the cross-sectional units, we estimate the impact that changes in creditors rights had on contractual debt characteristics (size, maturity and price) and non-contractual debt characteristics (bank vs. public debt, number of loans in domestic and foreign currency). The referred changes were brought about by the new Brazilian bankruptcy law and the model we first used to estimate its impacts was a panel model with fixed effects. To see if the estimated effects were specific to Brazilian firms, we ran the same model on a sample of non-brazilian firms. However, any change in macroeconomic conditions affecting Brazil but not the other countries with firms in our sample would invalidate our identification strategy and confound the estimated effect. Hence, to check whether the results were driven by changes in macroeconomic conditions in Brazil instead of the reform, we perform a battery of falsification and placebo tests that replicate our estimation, under slightly different conditions. The falsification tests are based on data from before the reform, which we use to replicate our empirical exercise in the counterfactual situations where the reform was implemented in 2001 or 2003 instead of 2005. We also run placebo tests that try to capture possible pre-reform trends through the inclusion of binary explanatory variables for the years before the reform came into effect. In addition to that, we also consider alternative model specifications. These different models provide checks for the logic of our original empirical design and help minimize concerns about the results. The first alternative method is a diff-indiff model, in which Brazilian firms form the treatment group and non-brazilian firms compose the control group. It would be best if both groups displayed similar behavior in credit variables prior to the bankruptcy reform. However, this is not a straightforward property to attain. And since the identification of average treatment effects using differencein-difference estimation relies on the assumption that treatment and control units experience common trends (as emphasized by Angrist and Pischke (2009) and Blundell and Dias (2002)), the use of a traditional diff-in-diff method may not consistently estimate the average treatment effect on the treated. Because our sample contains firms from four different countries, assuming that all firms are subject to the same macro trend could be exceedingly unrealistic. In fact, differential trends might arise in the evaluation of the bankruptcy reform effect if treated and controls operate in different financial markets, which is exactly our case. Because of this, we use as a third approach a modification of the basic diff- 5

in-diff model to allow for different firm trends within our treatment and control groups (triple diff). Some important conclusions can be drawn from the results. First, our findings point to a significant reduction in the cost of debt and to an increase in the amount of both total debt and long-term debt. These findings are in line with La Porta et al. (1997, 1998), who argued that better legal protection in bankruptcy motivates creditors to supply loans at better terms. Second, consistently with Scott Jr (1977), priority order has a significant value. Since secured creditors have been beneffited more by the new law than unsecured creditors, the effect is more pronounced on long-term debt, which is known to be correlated with secured debt. In addition to that, we found no statistically significant effect on short-term debt, although the point estimate was negative. This result can be explained by the new law, inasmuch as it encourages lenders to participate more actively in the bankruptcy procedures, mitigating the problem of lender passivity and leading to corporate debt structures with longer maturity (see Diamond (2004)). Finally, in an environment where firms have low credit quality, loan ownership becomes more diffuse as the cost of defaulting for liquidity reasons decreases (see Bolton and Schaferstein (1996)). In line with this, we found a significant increase of public loans relative to bank loans. Our next step was to investigate if the reform had a heterogeneous impact on firms, conditional on tax intensity, level of tangible assets and risk of liquidity default. Tax intensive firms should feel a stronger effect on their long-term debt, since secured creditors now have a priority over tax claims that they did not have prior to the reform. Also, one would expect the effect on firms with more tangible assets to be stronger, since they can use a larger portion of their assets as collateral. Finally, firms with higher risk of liquidity default should benefit more from the reform. We found results consistent to what would be expected for tax payments and liquidity default risk, but not with respect to tangible assets arguments. The remainder of the article is organized as follows: section two discusses the Brazilian bankruptcy reform; section three presents the empirical design; section four discusses the data base; section five reports the results; and section six concludes. 2 The New Brazilian Bankruptcy Law Most of the old legislation regulating the Brazilian bankruptcy procedure was enacted in 1945. Despite prescribing both liquidation and reorganization the latter called concordata (composition with creditors) and intended to prevent or reduce the liquidation of viable enterprises in practice the insolvency process was ineffective in maintaining the value of the firms assets and protecting creditors 6

rights in liquidation. The bankruptcy priority rule was very punitive to creditors. It was specified in the following order: first, labor claims; second, tax claims; third, secured creditors claims; and finally unsecured creditors claims (including trade debts). The process through which the assets were made available to creditors was slow and highly ineffective, mainly because of procedural inefficiency, lack of transparency and the so-called succession problem, whereby tax, labor, and other liabilities, including off-balance sheet liabilities, were transferred to the buyer of an asset sold in liquidation. This liability transfer depressed the market value of an insolvent company s assets. In addition, the priority given to labor and tax claims had the pernicious effect of eliminating any protection to other creditors. The old reorganization procedure, the concordata, basically only postponed debt payment and did not lead to restructuring. The procedure also incentivized an informal use of the system to promote consensual renegotiations, not withstanding an insufficient legislative framework capable of fostering workouts. 7 As a result of the old bankruptcy procedures design, distorted incentives and lack of effective mechanisms to support corporate restructuring resulted in disproportional default rates of potentially viable companies. Figure 1 illustrates one of the dimensions of inefficiency that characterized the former bankruptcy procedures. For a comparative analysis, we use seven groups of countries: the Organization for Economic Cooperation and Development (OECD), Latin America and the Caribbean (LAC), the Middle East and North Africa (MENA), Europe and Central Asia (ECA), East Asia and the Pacific (EAP), South Asia (SAS) and sub-saharan Africa (SSA). 8 Notice that the average time to close a business in Brazil was more than twice the average for the Latin America. This situation eroded the value of assets and thus lowered the amount received by creditors. The creditor recovery rate before the reform illustrates the final effect of an inefficient procedure with poor creditor protection. In Brazil, the recovery rate in case of bankruptcy was a mere 0.2%, while the averages for Latin American and OECD countries were 26% and 72%, respectively. The basic reason for such low recovery was the priority order, since creditors ranked behind labor and tax claims. Thus, the remaining amount from the bankruptcy process used to pay creditors was usually insignificant or even nil. Since creditors knew this ex-ante, they increased the interest rate charged to firms. This was the main reason for the extremely high interest rate spread in Brazil before the new law. 9 7 A workout is an informal renegotiation of loans that takes place outside the judicial courts. 8 The Latin American and Caribbean block is composed of Argentina, Bolivia, Brazil, Chile, Colombia, Costa Rica, the Dominican Republic, Ecuador, El Salvador, Guatemala, Haiti, Honduras, Jamaica, Mexico, Nicaragua, Panama, Paraguay, Peru, Uruguay, and Venezuela. 9 See Araujo and Funchal (2005) 7

Figure 1 (HERE) On June 9, 2005 the new bankruptcy legislation (Law 11,101/05) came into effect. The new law improved on existing rules by integrating the insolvency system into the country s broader legal and commercial systems, providing in- and out-ofcourt options to reorganize, and striking a reasonable balance between liquidation and reorganization. It also significantly improved the flexibility of the insolvency legal system by allowing the conversion of reorganization proceedings into liquidation, by establishing a period in which debtors can apply for rehabilitation in response to liquidation proceedings filed against them, and by introducing a new out-of-court reorganization system for prepackaged restructuring plans. Also, the new law imposed a new constraint on debtors bankruptcy requests: the value of the ending liabilities must exceed forty times the minimum-monthly wages. All these changes in the bankruptcy system s design have helped to dissuade debtors from misbehaving. The new liquidation procedure introduced six key changes. First, labor credits are now limited to an amount equal to 150 times the minimum monthly wage to each worker. Second, secured credits are now given priority over tax credits. Third, unsecured credits are given priority above some of the tax credits. Fourth, the distressed firm may be sold (preferably as a whole) before the creditors list is constituted, which can speed up the process and increase the value of the bankruptcy state. Fifth, tax, labor, and other liabilities are no longer transferred to the buyer of an asset sold in liquidation. Finally, any new credit extended during the reorganization process is given first priority in the event of liquidation. 10 The first two changes have had a direct impact on secured creditors priority. Since under the former bankruptcy law, secured creditors came after all labor and tax claims, the priority given to secured creditors has increased significantly. The third one has increased unsecured creditors priority. The fourth, fifth and sixth changes, in turn, have increased the value of firms in bankruptcy and as a consequence the amount recovered by creditors. The more creditors expect to receive in an insolvency state, the less they will require firms to pay in the solvency state, thus reducing the cost of capital. Brazil s new reorganization procedure was inspired by U.S. Chapter 11 of the Bankruptcy Code. Whereas the previous law did not permit any renegotiation between the interested parties, and only a few parties were entitled to recover assets, now management makes a sweeping proposal for recuperation that must either be accepted by workers, secured creditors and unsecured creditors (including trade creditors) or the distressed firm will be liquidated. Creditors play a more significant role in the procedure than previously, including negotiating and voting for the reorganization plan. 10 See Araujo and Funchal (2005) 8

This new design of the Brazilian bankruptcy procedures brings new incentives to the interested parties. The incentive for debtors to default strategically is reduced, for two main reasons: first, the conditions under which debtors can file for bankruptcy are limited to those prescribed by law; and second, a reorganization procedure can be converted to liquidation at the creditors discretion, a feature that nearly eliminates the use of reorganization as a bargaining mechanism. Figures 2 and 3 give a clear indication of this change in incentives. They show, respectively, the number of liquidation requests and the number of reorganization requests, both before and after the Brazilian bankruptcy reform. Notice that in both cases the number of requests dropped abruptly after the reform, which is in line to what would be expected of reduced incentives to strategic defaults. Figure 2 (HERE) Figure 3 (HERE) Creditors, on the other hand, now face new incentives to actively participate in the bankruptcy procedure, due to three key changes: first, now creditors play a more significant role in the procedure than they previously did, including negotiating the reorganization plan and then voting on its validity; second, they can file for out-of-court reorganization; and third, their credits priority in case of liquidation is higher now than it was under the previous legislation. Figure 4 illustrates this new reality. Observe that the use of the reorganization procedure increased, mainly in periods of crisis like the mortgage crises period when the liquidity default problems were more evident. In this case, the number of liquidation requests, which previously was more than ten times greater than reorganization requests, dropped to practically the same as the number of reorganization requests. Figure 4 (HERE) Figure 5 (HERE) As a consequence of the bankruptcy reform, in 2006 the creditor recovery rate increased to 12% in Brazil, while the average of Latin American and OECD countries remained stable (29% and 67%, respectively). It is still growing, reaching 17% in 2009. Also, the average time to close a business in Brazil has fallen from ten to four years, which tends to reduce depreciation of assets. 11 All these factors point to a potential effect on the supply side of debt, since creditors conditions have improved with the bankruptcy reform. These changes should have an impact on each of the debt variables we examine in the rest of this paper. 11 Source: Doing Business 2010, World Bank. 9

3 Empirical Design In this section we present the framework used to estimate the effect of the new Brazilian bankruptcy law (henceforth referred to as NBBL) on seven different debt variables divided into two groups: Contractual debt terms: cost of debt, short-term debt, long-term debt and total debt; Non-contractual debt terms: bank and public debt ratio; number of national debt contracts; number of foreign debt contracts. To do this, we compare outcome variables from a group of firms that were affected by this institutional change to a group of firms unaffected by the new legislation. The former is the treatment group and contains only Brazilian firms. The latter is the control group and includes only non-brazilian firms, from Argentina, Chile and Mexico. This quasi-experimental approach allows us to control our analysis for common shocks in the credit market in the period when the new Brazilian bankruptcy law was being implemented. We consider three different methods in our empirical design. First, we analyzed the effect of changes in the bankruptcy law design on debt characteristics (contractual and non-contractual) only for Brazilian firms. We run the same model for non-brazilian firms, our control, and check if the effect we found for the Brazilian firms is indeed a result of the bankruptcy reform or if it comes from general market conditions instead of the bankruptcy reform. Also, we perform several falsification and placebo checks that aim to prove the robustness of the results. Second, we perform a difference-in-difference analysis where in the same econometric model we consider both Brazilian firms and non-brazilian firms, the former being the treatment group and the latter the control group. However, a great part of the literature on program evaluation focuses on establishing the counterfactual capable of producing a viable identification strategy and a good control group. This effort is necessary because one rarely encounters an ideal natural experiment environment, in which cross-sectional units are found in the treatment group by accident. Indeed, our problem would be much simpler if firms were randomly selected to be regulated by the NBBL. In that case, it would suffice to compare the value of credit-related variables for firms in the experimental group with the value of the same variables for firms in the control group. Since compliance with the new law is mandatory for almost all Brazilian firms, 12 the selection rule we would like to be random is in fact deterministic, and we are forced to choose different techniques 12 Law 11,101/05 does not apply to financial institutions, insurance companies, credit unions, state-owned and controlled companies, private pension entities and some others. 10

to estimate the impact of the new law. This suggests a third approach, the triple difference method, also called diff-in-diff with different trends. 3.1 First Difference Model The first empirical specification we adopt is a standard panel data model, with firm fixed effects. We apply to each of the dependent variables the following model: y it = η i + βdbl t + ΓX it + u it, (1) where dbl t is a dummy variable that indicates whether the new bankruptcy law was in effect in year t, which equals one from 2005 on and zero otherwise, X it is the vector of controls, η i and u it are the fixed-effects term and the residual term respectively. Then we estimate the model in equation (1) for our list of dependent variables and repeat the exercise with a different data set. We run the same regressions, but instead of using the Brazilian firms (our treatment group), we use only the non-brazilian firms from our sample. This exercise is useful to investigate if the results we found were indeed particular to Brazilian firms or indicated some global trend unidentified in the specification of our basic model. Nonetheless, even if our results hold only for Brazilian firms, they could still be a product of changes in Brazilian macroeconomic conditions instead of being driven by the reform itself. To check this, we perform a battery of falsification tests. 3.1.1 Falsification Tests These tests aim to verify whether any effect we may find and attribute to the new law is in fact particular to the post-2004 years or is just the expression of a more general trend, already present in the years prior to the new law. To do that, we run regressions for the two following models, which we built to check if some effect can be found by pretending as if the bankruptcy reform was enacted not in 2005, but in 2003 (equation (2)) or in 2001 (equation (3)). In this exercise, no data from 2005 onwards were used. y it = η i + βdbl2003 t + ΓX it + u it, (2) y it = η i + βdbl2001 t + ΓX it + u it, (3) where dbl2003 t and dbl2001 t are dummies analogous to dbl t in equation (1). However, instead of stipulating 2005 as the year in which firms began to feel the effect of the new law, we codify dbl2003 t as one from 2003 forward and dbl2001 t as one from 2001 onwards. 11

In these tests, when we estimate the values of the coefficients associated with the false bankruptcy reforms of 2001 and 2003, we expect to find no effect or at least an effect different from that found when we estimated the impact of the true reform. If the results turn out to be similar to those obtained in the estimation of equation (1), than we probably cannot guarantee that our findings are due to the new law, instead of some broader macroeconomic movement. 3.1.2 Placebo Tests Besides falsification tests, we also run several placebo tests to check the robustness of our results. In these tests, in addition to the bankruptcy law dummy variable, other coefficients are introduced of some possible pre-reform trend. For each outcome variable, we introduce up to three dummies relative to the years immediately before the bankruptcy reform, resulting in the following model specifications: y it = η i + βdbl t + δ 1 d2004 t + ΓX it + u it, (4) y it = η i + βdbl t + δ 1 d2004 t + δ 2 d2003 t + ΓX it + u it, (5) y it = η i + βdbl t + δ 1 d2004 t + δ 2 d2003 t + δ 3 d2002 t + ΓX it + u it. (6) In these placebo tests, it is important to find no evidence, in the years of the reform, of an effect similar to the one possibly found for the years after the reform. Otherwise, it would point to the existence of a trend in debt variables unrelated to the bankruptcy reform. 3.2 Diff-in-Diff Model Without the benefit of a controlled randomized trial, we turn to a difference-indifference approach, which compares the change in the outcome of units in the treatment group before and after the intervention to the change in the outcome of units in the control group, after controlling for non-observables that are time invariant. The difference-in-difference model 13 can be specified as the following firm fixedeffect linear regression model: y it = α + β (dbrazil i dbl t ) + ΓX it + η i + ψ t + u it (7) The right hand side of the equation includes the firm s fixed effects (η i ), to control for specific factors that are fixed over time; the vector of controls, X it ; and the coefficient β, that is, the difference-in-difference estimator of the impact of Brazilian bankruptcy reform on debt variables. We define the diff-in-diff dummy variable as dbrazil i dbl t, where: 13 See Meyer (1995). 12

dbrazil i : is a dummy variable that equals 1 when the cross section unit i belongs to the treatment group (the group of Brazilian firms), and 0 otherwise; and dbl t : is a dummy variable that equals 1 if the time series units is 2005 or higher, and 0 otherwise. By comparing changes, we control not only for time-invariant characteristics that affect debt variables and might confound the effects of the bankruptcy reform, but we also allow for the existence of a general macro trend influencing the outcomes of both groups. The control group is expected to emulate the treatment group s behavior had the law not been in effect, acting as a replacement for the ideal but unobserved counterfactual we needed. This is the key underlying identification assumption of this model. 3.3 Triple Differences Model One basic feature of diff-in-diff models, such as the one presented in the previous subsection, is the assumption that both the treatment and the control groups are subject to the same macro trend that influences their outcomes. Nonetheless, although this is arguably a valid assumption when both groups displayed similar behavior in credit variables prior to the bankruptcy reform, this is not a straightforward property to attain. In fact, differential trends might arise in the evaluation of the bankruptcy reform effect if treated and controls operate in different financial markets, which is exactly our case. Because of this, we use a modification of the basic diff-in-diff model to allow for different firm trends within our treatment and control groups (triple diff). The model specification we adopt is a panel data model, with fixed effects and a firm-specific trend. We apply to each of the dependent variables the following model: y ist (0) = δ i t + λ s + X it Γ + η i + u ist (8) y ist (1) = δ i t + λ s + β + X it Γ + η i + u ist (9) For each variable considered, the index i indicates the cross-section dimension and the index t indicates the time-series dimension, as usual. The subscript s is reserved for the firm s country of origin. The first term in equation (8), y ist (0), is the outcome variable when firm i is not affected by the NBBL in period t. The term Γ, is the k 1 coefficients matrix and X it is a 1 k matrix containing the k covariates we use as controls, to capture some firm heterogeneity concerning the size of assets, profitability, and so forth. 14 14 In our model, these control variables are: (1) total assets; (2) price-to-book value; (3) return 13

The last three terms are the firm fixed-effects control terms (η i ), which account for all time-invariant unobservable firm-specific factors; a heterogeneous linear trend (δ i ); a country-specific effect (λ s ); and the idiosyncratic error term (u ist ). This specification has the advantage of allowing for a great deal of heterogeneity among firms of different countries (through the parameter λ s ) or even among firms within the same country (through the incidental parameter η i ). And the fact that it also contemplates different trends for firms within each group implies a specification that is general enough to allow a different trend for each firm. The difference between equations (8) and (9) is the presence of the parameter we are interested in estimating, denoted by β in equation (9). We can use dummy variables to transform the empirical model formed by these two equations into a single-equation regression model: y ist = βd it + X it Γ + η i + δ i t + λ s + u ist (10) In this formulation, D it is the indicator variable, which equals one if firm i is subject to the NBBL in period t and zero otherwise. Since we wish to consistently estimate β, we subtract from equation (10) the analogous equation for t 2. By doing that, we eliminate both η i and λ s, arriving at the differenced form of the equation: y ist = β D it + X it Γ + δ i + ɛ ist (11) In this equation, we have δ i := 2δ i ; and ɛ ist = u ist = u ist u ist 2. Then, we estimate the differenced equation using the within estimator. Notice that in this specification the diff-in-diff parameter β represents the average change in the trend of the dependent variable for the treated firms. If this parameter is significant, it means that the NBBL caused a variation in the path of the financing policy. 3.4 Non-Brazilian firms as a counterfactual To establish a necessary counterfactual to identify the effect of the NBBL, we need a set of firms capable of emulating the behavior of Brazilian firms that were not influenced by the new legislation. Our group of choice is composed of firms from three different Latin American countries: Argentina, Chile and Mexico. Of course, we expect to find both differences and commonalities between firms from different countries. Regarding the differences, our hope is that they can, to a great extent, be explained by the heterogeneity our specification allows. As to the commonalities, we expect the Argentine, Chilean and Mexican credit markets to have significant similarities to the Brazilian credit market. Latin American is a region with a weak history of creditor protection and low credit recovery on assets (ROA); (4) earnings before interest and taxes (EBIT); (5) tangibility; (6) tax intensity; and (7) interest coverage ratio. 14

rates. Also, the four countries have civil law systems, which, as pointed out by Djankov et al. (2008), correlates with inefficient bankruptcy outcomes, higher interest rates and longer bankruptcy procedures, leading to more depreciated assets and lower recovery rates. When controlled by the appropriate covariates and the fixed-effects both in level and trend (when the specification so indicates), we expect these commonalities to work in our favor, satisfying the standard exogeneity assumptions, as stated by Wooldridge (2002), making these non-brazilian firms an adequate control group for our empirical exercise. Regarding different macro trends that possibly affect firms from different countries, we take advantage of the period of relative stability the region experienced during the 12 the years covered by our sample. Latin America is a region with a long history of economic turmoil. During the 1980s and 90s, many Latin American countries faced a variety of severe crises with significant economic consequences. And although some of them in fact originated abroad, most can be traced back to the region s own weaknesses. In the past decade, however, this scenario has changed considerably, partially on account of the global commodities boom and, for some countries, partially because of sound economic policies. 15 15 Since 2002, when Brazil experienced a confidence crisis caused by fears that the newly elected government would favor unorthodox economic policies, the country has entered a stability cycle not seen in recent decades. From 2003 to 2007, the country experienced a period without significant political or economic distress, and even the international financial crisis of 2008 and 2009 had a smaller impact on Brazil than on most developed countries, such as the U.S. and Britain (source: Brazilian Central Bank Report). In turn, Argentina, which has a history of external vulnerability, from the mid-nineties to the beginning of the following decade faced repeated economic shocks, from the crisis in Mexico (1994), Asia (1997), Russia (1998) and Brazil (1999). In 2002, after years of economic turmoils, the country was forced to abandon the one-to-one parity it had established between the local currency and the American dollar, and announced a US$150 billion sovereign debt default. The Argentine economy hit its lowest point. However, since then Argentina has definitely improved in some fronts, helped by the undervalued local currency and the upward trend in international commodities prices. And although the improvements have come along with a somewhat unstable political situation and inflation rates high enough for the local government to be accused of tampering with the official price indexes, the Argentine economy is far from the near-collapse scenario it faced in the past. And the external shocks that periodically shook the country s economy in the nineties are no longer a reality (source: Argentine Central Bank). Chile is a country in which government has a limited role in the management of the economy. Its recent history is an exception to the rule of bad economic policies in Latin America. During most of the 1990 s Chilean economy showed great strength, with high growth rates. The beginning of the subsequent decade, however, came with the worsening of the world economy, specially in the aftermath of September 11th. This economic downturn, led by the economies of the U.S., Europe and Japan, had an important impact in countries like Chile, but soon after the Chilean economy recovered, boosted by the boom in the commodities prices. During this first decade of the century, the country has seen a few general elections, but the major parties seem to agree on merits of the economic policies that have secured the country s economic stability (source: Chilean Central Bank). 15

This relative stability plays in our favor by reducing the variance of the macro shocks. This is true as long as the average macro effects that affect our model s dependent variables remain unchanged in the pretreatment and post-treatment periods. And although macro shocks affecting only some of these countries or affecting some countries more severely than others tend to harm our estimates, there is no reason to expect these effects to be meaningful, on average, given the number of countries we have in our sample and the number of time periods. 4 Data Set and Descriptive Statistics Our sample contains firm-specific accounting fiscal-year-end data on 698 publicly traded firms, from 1999 to 2009, disregarding observations from financial institutions, since their financial policy differs strongly from those of firms in other sectors and they are not subject to the NBBL. About half of these firms (338) are Brazilian and belong to the treatment group; the rest of them belong to the control group, which is divided as follows: 108 firms are Mexican, 82 are Argentine, and the remaining 170 are Chilean firms. The data were obtained from Economatica. We consider as firm debt the sum of balance sheet short-term and long-term debt, plus suppliers account (also called trade credit). We also have information on debt concerning its source (bank or public debt). The cost of debt is calculated, for each firm, as total year s interest expenses, divided by its total debt (except trade debt) over the same period. To act as controls, we also collected fiscal-yearend information on the firms total assets, price-to-book ratio, return on assets and earnings before interest and taxes (EBIT). The reform may affect firms debt variables in a heterogeneous way, depending on some borrower characteristics, such as: 1. Tangibility: tangible assets are easier to collateralize, thus having a direct effect on debt characteristics. The straightforward approach is to expect firms with a higher proportion of tangible assets to benefit the most from the new law, since the reform made secured creditors climb one position in the priority order, placing them right in front of tax claims and second only to labor claims. Nonetheless, we consider that in an economic environment Finally, we turn to the Mexican economy. The geographic proximity and economic ties between the United States and Mexico have contributed to increasing the level of exports, but also made the Latin American country much vulnerable to downturns of the American economy. The crisis of 2000 and 2001 harmed the Mexican economy, that had, like most Latin American economies, to deal with a smaller inflow of foreign capital and a drop in the exports. Mexican GDP fell 0.3 percent that year, although inflation remained low and interest rates drop to a level that for long had not been seen. The years of 2002 and 2003 were a period of moderate growth for the Mexican economy, but it kept recovering in the following years, until the most recent international financial crisis (source: Mexican Central Bank). 16

where creditors rights are weak, collateralized debt plays a significant role in insolvency situations, inasmuch as creditors are allowed to collect the asset placed as collateral as soon as debtors become insolvent. We capture asset tangibility through the ratio between PP& E (property, plant and equipment) and total assets. 2. Tax intensity: firms that pay more taxes should benefit more from the bankruptcy law reform, since the new law improved the priority of secured creditors over tax claims. We tried to capture tax intensity by the ratio between tax expenditures (EBIT minus Net Profit) and revenue. 3. Risk of liquidity bankruptcy: Firms more likely to experience liquidity problems should face a grater risk of bankruptcy. This forces lenders to increase interest rates in order to break even, which in turn encourages this type of firm to default strategically. By increasing creditors recovery rate, the new law should mitigate these problems, making the effect of the reform stronger for firms with a greater risk of experiencing liquidity problems. Following Asquith et al. (1994), Andrade and Kaplan (1998) and Almeida et al. (2011), we employ the twelve-month variation of interest coverage ratios (EBIT divided by financial expenses) as a measure of risk of liquidity bankruptcy. We also have data on the number of loan contracts, divided between local currency and foreign currency contracts. We use the local and foreign currencies as proxies for the origin of loans. This sample was built directly from explanatory notes to the Brazilian firms financial statements. 16 Our data set contains 218 firms from 2001 to 2009. We then merged these data with the Economatica database. Table 1 presents the sample mean values of the explained variables, for both the treatment and the control groups, in the pre- and post-treatment periods. Panel A display descriptive statistics for the whole sample. It shows a decline in the cost of credit for Brazilian firms, while the same cost for the control group remains almost stable. For almost the entire set of the remaining contractual debt variables, there is an increase in value for the Brazilian firms despite the fact that this increase was significant only for long-term debt. For non-brazilian firms, on the other hand, we can observe a drop in the value of these same variables, within this same period. At first glance, these numbers could indicate a possible shift in the supply of credit in Brazil. The improvement in almost all measures of credit quantity and in its price illustrates a possible supply-side shock. 16 Each firm s balance sheet is available at the website of BOVESPA (Brazilian Stock Exchange). We analyze section twelve ( Loans and Financing ), which provides detailed information on the loans. 17

We also investigate the ratio between bank debt and public debt, which we consider to be a proxy of the relationship between single and multiple lenders. Notice that for both Brazilian and non-brazilian firms, there is a reduction in this ratio. There is also a significant increase in the number of both local and foreign contracts. Tabela 1 (HERE) However, although the amount variables did present an economically significant increase in total debt, this increase is not statistically significant. Going deeper in our analysis, we observed that this happens because the total debt/total assets ratio is excessively high for firms in financial distress (negative equity). Therefore, we decided to exclude these firms from our sample and then recalculated the descriptive statistics, even though this class of firms is very important for our study. We report these statistics in Panel B. The results are qualitatively the same: a significant decrease in the cost of debt for Brazilian firms, contrary to what was observed for non-brazilian firms; a drop in the total debt of both the set of Brazilian and non-brazilian firms (with the former experiencing a lesser drop than the latter); an increase in long-term debt for Brazilian firms, as opposed to a reduction for non-brazilian firms; and a decline in short-term and trade debt that is similar in both groups. 5 Results This section presents the results regarding the effect of the new bankruptcy law on firms debt characteristics. We first present the results from the first difference model and its robustness checks, then we move to the findings from the differencein-difference and triple differences model. Also, we show some evidence on aggregated data on the Brazilian private credit market for firms. 5.1 First Difference Results Table 2 presents the estimated effect of the bankruptcy reform on contractual debt terms. Regressions 1 and 4 point to a positive effect on the total amount of debt and long-term debt of Brazilian firms. The coefficients indicate an increase of 20% and 47% for total and long-term debt respectively. There is no evidence of changes in short-term debt and trade credit (regressions 2 and 3). In fact, the short-term debt regression shows a negative point estimate of our parameter of interest. And finally, regression 5 shows that the new bankruptcy law brought a reduction of aproximately 15% in the cost of debt financing. In relative terms, this represents a reduction of almost 30%, since the average cost was 56% during the pre-treatment period (see Table 1). 18

These results are in line with the theoretical literature on credit (Townsend, 1979; Aghion and Bolton, 1992; Hart and Moore,1994, 1998; and Scott Jr., 1978). The new law improved the level of creditor protection and the efficiency of bankruptcy procedures. Creditors now have a higher chance of recovering a larger portion of their loans. And the more they expect to receive in the insolvency state, the less they will require firms to pay in the solvency state and the more they will be willing to lend. Thus, these are signs that the Brazilian bankruptcy reform had a positive effect on the supply of credit. We also have results on debt maturity. Notice that long-term debt increases while short-term debt remains stable, leading to a higher proportion of long-term debt in the average firm s capital structure. This result is somewhat similar to what both Qian and Strahan (2007) and Bae and Goyal (2009) found. The new bankruptcy law design encourages lenders to participate in bankruptcy procedures more actively, eliminating the need to extend only short-term debt as a discipline mechanism and leading, after the reform, to a debt structure with longer maturity (see Diamond, 2004). We also investigated how the effect of bankruptcy reform relates to tax intensity (Taxes/Revenue), likelihood of liquidity default and asset tangibility (PP&E/ Total Assets). Some of these heterogeneous effects are in accordance with the theory, but most of their estimates showed no statistical significance, pointing to a homogenous effect of the law on firms debt financing. We found a weak significance of taxes on long-term debt, with the point estimate being positive, which indicates that more tax intensive firms benefited more from the bankruptcy law than the less tax intensive ones (see regression (4) in Table 2). Tax intensive firms should feel a stronger effect on their long-term debt, since secured creditors now have priority over tax claims they did have before, and secured debt is more strongly correlated with long-term debt than to short-term debt. We also found a heterogeneous effect of tangibility on long-term debt (see regression (4) in Table 2), although the results relative to tangibility are not easy to interpret. One would expect that firms with lower tangibility to have been less benefited by the reform than those with more tangible assets. However, we found that they benefited more from the reform. One possible interpretation of this finding is that firms with more tangible assets can use them to serve as collateral for secured loans. Since creditors can take seize these collateralized assets before the insolvent firm files for bankruptcy, this ability works as a substitute for a bankruptcy procedure. Therefore, firms with lower tangible assets to use as collateral should benefit more from the reform. Finally, we detected one more heterogeneus effect, concerning the likelihood of liquidity default and the cost of debt (see regression (5) in Table 2). Firms with higher odds of experiencing liquidity problems benefitted more from the reform. The interaction between the proxy for the probability of liquidity 19

defaults and the bankruptcy reform dummy shows that firms with more chances of defaulting (negative variation in interest coverage ratios) have lower cost of debt in comparison with firms having a lesser possibility of defaulting. Table 2 (HERE) Having established the positive general effect the new bankruptcy law had on contractual debt variables, we now analyze the effects on the non-contractual terms of debt. Table 3 presents the estimated effect of the NBBL on the number of domestic and foreign contracts, and on the ratio of bank debt to public debt. Notice the number of domestic contracts increased approximately 33% after the bankruptcy reform, while the number of foreign contracts remained relatively stable. This result goes against previous findings, such as Qian and Strahan(2007), according to whom foreign bank loans are particularly sensitive to the legal institutional environment, and the share is positively correlated to creditors protection. Our result appears to be in accordance with Mian (2006), who argued that geographic and cultural distance play a bigger role than that played by legal institutions in determining the amount of loans provided by foreign financial institutions. Esty (2004) presented a positive correlation between foreign loans and creditors rights, but a negative correlation to financial system development. Since the NBBL increased creditors rights and financial development, it is fair to expect an ambiguous effect on foreign loans. As for the remaining results for bank to public debt ratio in Table 3, the reform brought a reduction of approximately 55% in this ratio, pointing to an expansion of public loans relative to bank loans. Such evidence can be explained by an increase in the firms liquidation value after the reform. 17 This increase in value reduced the need for borrowing from fewer creditors, and tended to help maximize firms liquidation value, as stated by Bolton and Schaferstein (1996). It thus prompted firms to issue public debt, inducing a reduction in the bank debt to public debt ratio. Turning to the heterogeneous effects of the new law on non-contractual debt variables, we observe that for all the interactions between the bankruptcy law dummy and tax intensity, tangibility and likelihood of liquidity default, none of them have any statistical significance, except for tangibility on the number of foreign contracts, pointing to a stronger effect of the law on firms with more tangible assets. Table 3 (HERE) 17 A proxy for the firm s liquidation value is the creditors recovery rate. It used to be 0.2 cents on the dollar before the reform, but right after the reform it goes to 17 cents on the dollar. 20