Corporate Governance and Capital Structure in Brazil: Stock, Bonds and Substitution

Purpose: To study the Brazilian bond and stock markets for testing the stock market development theory of Demirgüc-Kunt and Maksimovic (1996). Originality/gap/relevance/implications: This paper tests the substitution hypothesis of stock market development, from debt to stocks, in a context of improved corporate governance, by analyzing the data with cointegration techniques. The findings show that the rejection of substitution hypothesis, as the bond market has a positive and significant association with stock market improvements. The findings also show that improving the quality of corporate governance could lead equity capital and borrower capital sources to be complementary and not substitutes, suggesting that Brazilian stock market reform has created a virtuous development cycle. Key methodological aspects: Positivist research using quantitative methodology. Data from a sample of 171 firms during 20 years, analyzed with cointegration. The null was a negative association between bond and stock markets. Summary of key results: Null rejection, non-consistent to theoretical framework. The results have shown a positive and significant association between stock and debt in an improved corporate governance context. Key considerations/conclusions: Improving the quality of corporate governance could lead equity capital and borrower capital sources to be complementary, and not substitutes, suggesting that Brazilian stock market reform has created a virtuous development cycle.

• ALEXANDRE RIPAMONTI • EDUARDO KAZUO KAYO • listing segments (Reed, 2002;Mallin & Ow-Yong, 2010). In Brazil, corporate governance reform created the listing segment NovoMercado, with significant improvements in information disclosure and shareholder protection (Rabelo & Vasconcelos, 2002). Consequently, the hypothesis is that the growth of the Brazilian stock market and corporate governance reform would reduce the size of the bond market.
The private and state-owned Brazilian banks possibly have a significant share of companies' long-term loans, whether as direct agents, business intermediaries, or bond issuance advisors or leadership dealers. Through the substitution hypothesis, the profitability of banks would be affected by the reduction of operations. In addition to to this scenario, banks would lobby in order to guarantee profits. By deduction, it is expected that Brazilian banks put up a tough opposition to stock market development and sponsor movements for impeding it.
Studying the consistency of stock market development in Brazil allows us to consider the effects of corporate governance reform, and theories of rational expectations and capital structure. For this reason, this study aims to examine the association between debt and equity in such a context. Using data from 1993 to 2013, 171 Brazilian public companies were analyzed with cointegration techniques (Johansen, 1988;Johansen & Juselius, 1990;Johansen, 1991). The results show that the Brazilian corporate governance reform has stimulated both stock and bond markets in a complimentary movement between debt and equity for companies that moved to or were created in the NovoMercado listing segment. The findings reject the substitution hypothesis in Brazil.
The next section details studies on stock market development. The third section presents the methodology, including sampling and econometric models. In the fourth section, the results are compared with those of other studies and with the suppositions of the capital structure theory. The last section presents some concluding remarks.

THEORETICAL FRAMEWORK
From a macroeconomic perspective, these studies relate stock market development to economic growth, liberalization, corporate governance reform and financial choices of companies. Jensen (1972) revised and discussed the models and theories that tried to explain stock market returns, and found that models with market factors explained returns more accurately.
At the company level, Demirgüc-Kunt and Maksimovic (1996) studied the financial choices and stock market development in 30 developed and developing countries from 1980 to 1991. The results indicated that, in general, a company's leverage is positively associated with bond markets and negatively (although non-significantly) with stock market development. For this reason, the paper could be considered to be a seminal work in stock market development theory. However, when segregating the sample, the authors identified the substitution of bonds by stocks in developed countries and an increasing leverage for the biggest firms in developing countries as the stock market developed.
Otherwise, small companies in developing countries were not affected by the improvement of the stock market. From this observation, the authors hypothesized a substitution of bonds by stocks in developed countries, which became central in the stock market development theory.
Demirgüc-Kunt and  found a strong correlation between stock market development and the banking market in most of the 44 developed and developing countries of their sample, from 1986 to 1993. The authors also highlighted the relevance of corporate governance on liquidity and volatility in the stock markets. Thus, the stock market development theory was improved with the inclusion of effects of corporate governance mechanisms and the behavior of developing stock markets. Levine and Zervos (1996) observed a positive association between stock market development and long-term economic growth in a sample of 41 countries from 1976 to 1993, using the stock market development measures of Demirgüc-Kunt and . In this study, institutional and macroeconomic characteristics also emerged as relevant variables in the behavior of bond and stock markets.
For Demetriades and Hussein (1996), the findings about an association between economic growth and stock market development are not consistent when analyzed with different analysis techniques. The economic growth and stock market development were found to be positively related, or at least the former would have a positive effect on the latter. The result was obtained through cointegration and causality techniques for a sample of 16 countries. Authors argued that inappropriate techniques were the main weaknesses of other studies. Although correct in challenging other findings, such a study nonetheless adheres to stock market development theory because it doesn't reject the inverse relationship between bonds and stocks.
De Santis and Imrohoroglu (1997) didn't observe any association between market liberalization, which includes stock market development, and stock price volatility in the developing stock markets of Europe, the Middle East, Asia, and Latin America from 1988 to 1996.
However, contrary evidence was gathered by Levine and Zervos (1998), who observed that stocks tended to increase in value while gaining liquidity, • ALEXANDRE RIPAMONTI • EDUARDO KAZUO KAYO • volatility, and integration after liberalization. Beyond that, the authors found that increasing international integration is experienced by countries with better information disclosure. The results prove the relevance of corporate governance mechanisms, as did Bekaert, Harvey and Lundblad (2005), who found that liberalization of stock markets increases annual economic growth by 1%. Levine and Zervos (1998) later found resistance from Stiglitz (2000), who claimed that liberalization doesn't affect stock markets. Rajan and Zingales (1998) assumed that a country's financial development would reduce the cost of capital, allowing financially constrained companies to grow faster in countries with more developed financial markets. For a sample of 41 countries from 1980 to 1990, authors found their hypotheses to be supported. Such evidence strengthens the idea that stock market development decreases the cost of capital. Wurgler (2000) pointed out the positive association between the legal protection of minority shareholders and the efficient allocation of capital in companies. For Bekaert, Harvey and Lumsdaine (2002), however, the new rules only produced changes for stock market development if they allowed the entrance of foreign capital. Rousseau and Wachtel (2000) examined stock market development and income of countries per capita, concluding stock market development requires such internal resources. Beck, Levine and Loayza (2000) found a positive and significant association between financial intermediary development and economic growth, productivity, tangibility, and private savings rates. It is relevant to the analysis, considering that stock market development could be related to the increase in volume and liquidity of shares of financial intermediaries or banks, as in Kominek (2004). Perotti and Van Oijen (2001) argue stock market development is determined by political risk that arose from selling state owned enterprises. The sample gathered 22 countries from 1988 to 1995. Institutional characteristics were identified as relevant variables for controlling the association between stocks and bonds. Edison, Levine, Ricci and Slok (2002) weren't able to reject the null hypothesis that economic integration doesn't increase the speed of economic growth, analyzing data in a dynamic panel of 57 countries. Conversely, Beck and Levine (2004) reported that economic growth was influenced by the stock and bond market development, as did Tachiwou (2010) when analyzing the countries of the West African monetary union. Chinn and Ito (2006) argue that increasing liberalization leading to increasing stock market development would only hold if legal development was also achieved. Therefore, the institutional variable is highlighted once again between stocks and bonds.  Mitton (2006) found higher stock return, investment, profitability, efficiency, and lower leverage rates in countries with foreign direct investments.
In Belgium, Van Nieuwerburgh, Buelens and Cuyvers (2006) found stock market development determined economic growth and such a relationship varies over time due to institutional changes, consistent with the findings of other studies.
De la Torre, Gozzi and Schmukler (2007) analyzed the impact of corporate governance reforms on stock market development in six countries and found the reforms also stimulated internationalization and increased share values and trading, but increased the risk of contagion from international financial crises. Klein and Olivei (2008) found a significant and economically relevant association among liberalization, financial development, and economic growth from 1976 to 1995 in 95 countries. Hasan, Wachtel and Zhou (2009) found that financial companies stimulated Chinese provinces' economic growth, showing a lack of other funding sources. Croci and Petmezas (2010) found that stock market development is associated with the target minority shareholders' return in mergers and acquisitions operations for scale purposes.
Finally, the substitution hypothesis has been extended to developing countries in Demirgüc-Kunt, Feyen and Levine (2013). They observed 72 countries from 1980 to 2008 and found that stock and bond markets develop slowly with economic growth, but in different ways. The bond market shrinks whilst the stock market increases. The authors applied a quantilic regression for analyzing the data.

METHODOLOGY
In general, finance research could be classified as positivist or post-positivist, according to Creswell (2007, p. 36) because they […] use surveys or experiments, with methods of numerical data analysis, testing and checking theories or explanations, identifying variables, which are reported in hypothesis, using standards of validity and reliability, through numerical measurement of information, with the application of unbiased methods and statistical procedures.
This study is similarly focused on validity and reliability, as well as replication and generalization, when testing the stock market development theory and the respective substitution hypothesis.
The main purpose of this research is to examine the relationship between the stock market value and the debt of Brazilian companies. It takes into account the Brazilian corporate governance reform (Rabelo & Vasconcelos, 2002), and the possibility that stock market development has shrunk the bond market, as mentioned in other studies (Demirgüc-Kunt & Maksimovic, 1996;Booth et al., 2001;Demirgüc-Kunt, Feyen, & Levine, 2013).
The hypothesis is that there is complementarity of funding sources between stock and bond markets and not substitution when considering corporate governance reform in Brazil.
The sample is made out of 171 Brazilian public companies from the 2 nd quarter of 1994 to 1 st quarter of 2013. The analysis was performed for the full sample and two subsamples, one of them for NovoMercado and another one for the other listing segments. The data was obtained from a local financial database named Economática and the models were estimated using Stata.
Since the Brazilian corporate governance reform in 2000 (Rabelo & Vasconcelos, 2002), there are four listing segments, with NovoMercado being the one that requires companies to provide full information disclosure and shareholder protection. The sample contains 91 companies from the traditional listing segment, 18 from the N1, 7 from the N2 and 55 from the NovoMercado. The companies combine 20 different activities, with utilities (28) and transportation (27) having the biggest part of the sample.
The sample excluded banks and financial institutions, as well as companies with less than 50 quarters of data or without debt data.
Data on debt, stock market value, debt on equity, return, size, tangibility, and growth of companies was gathered, as shown in Chart 1. From the sum of firm-level variables, full data was obtained for applying cointegration techniques.
The association between debt and stock market value was analyzed as was the relationship between leverage and return. In both cases, the coefficients were controlled in terms of sizing, tangibility, and growth, as they are traditional determinants of capital structure (Titman & Wessels, 1988 Market value of enterprise on total assets Economatica Control variable (Titman & Wessels, 1988) Source: Elaborated by the authors, upon the premises of Demirgüc- Kunt and Maksimovic (1996) and Titman and Wessels (1988).
The mean of debt on equity in the full sample is 19.29 per cent, with 47.31 per cent for NovoMercado and 19.34 per cent for the rest of listing segments, as shown in Table 1. The aggregate data presents a more accurate relationship between debt and stock market value, as shown in Table 2, and allows the observation that leverage on the NovoMercado is significantly higher than other listing segments, as Chart 1 shows.
The NovoMercado also presented lower tangibility and higher growth than other segments, as shown in Table 1, which suggests that the higher leverage of the segment could be linked to growth opportunities.  The five possible models of Johansen's cointegration (Johansen, 1988;1991) were developed in the sample data. The cointegration is observed among non-stationary variables that relate to one another through residuals in the long run and have a short-run mechanism for error correction.
In this sense, the difference among applied models refers to the existence of an intercept or trend and drift in the long and short-run relationship equations, as well as in the specification of linearity of the trend or even in both (Asteriou & Hall, 2011).
Therefore, the different models allow us to check whether relationship variables stay smooth over time or assume several different shapes, which would be translated into equation coefficients. In Table 2, the five model estimates for all the samples, and the long-run relationship equations, including coefficients and statistical significance are presented.
The simplest model, although improbable, suggests that there is neither an intercept nor a trend in the long-run or short-run relationship with the assumption that there are no deterministic components in the data.
The second model is an intercept only in the long-run relationship. Therefore, there are no linear trends in the data and the intercept is restricted to the long run, with first differences having a zero mean.
The third model has a long-run intercept that is cancelled out by the intercept in the short-run relationship, without linear trends in the level variables, while the specifications are allowed to drift around an intercept.
The fourth includes a trend in the long run with an intercept, which is also allowed in the short run without trends.
The last model allows for linear trends in the short-run model and quadratic trends in the long-run relationship (Asteriou & Hall, 2011).
An estimation of the five alternative models was conducted to determine the best specification and to check whether the changes would have more impact on

DESCRIPTIVE STATISTICS
• ALEXANDRE RIPAMONTI • EDUARDO KAZUO KAYO • the sense and significance of association among stock market values and debt. Relevant differences could indicate bias in the model specification.
Cointegration techniques have been chosen for their many advantages: their ability to measure the correction from disequilibrium of the previous period; the solution they provide to the problem of spurious regression as the error correction mechanisms are formulated in terms of first differences; the ease in econometric modeling and the prevention of increases in long-run relationship errors (Asteriou & Hall, 2011, p. 359). The cointegration techniques are also based on an attempt to validate other findings acquired through time series methods.
The variables' stationarity was tested, along with the appropriate lag length of the relationship among the variables, the best specification among the five possible models, the rank of cointegration, and the long-run and short-run relationship equations. The next section presents the coefficients of long-run relationship, which serves for testing the null hypothesis.
One could consider the omission of the maximum likelihood specification used in the research, and the demonstration of normal distribution of data, as restrictions for this paper. However, such an eventual restriction is likely to be mitigated through the rigorous application of all cointegration techniques' procedures preconized by Johansen (1988;1991), in a way that the identification of the parameters shown in Table 2 were not misstated. This identification required that the debt variable was normalized to 1. Additionally, Table 1 shows the description of all variables.

RESULTS AND DISCUSSION
As presented in Table 1, the mean leverage of NovoMercado was higher (47.31 per cent) than the full sample (19.29 per cent) and the rest of segments (19.34 per cent).
The NovoMercado was marked by higher growth and lower tangibility, as highlighted in Table 1, reinforcing the growth opportunities set as capital structure determinants.
Most specifications of the cointegration models were made using the stationarity checking and information criteria of Akaike (1974) and Schwartz (1978), except for the rank of cointegration, which was determined using the maximum likelihood and trace tests.
The findings were compared with those of the reviewed literature, which was directly related to stock market development theory. In addition, Table 2 shows only the long-run relationship equations, the respective statistical significance, and the maximum likelihood and trace test's results. Time series causality tests weren't developed due to the focus on the long-run and short-run relationship estimates and not on the positioning of variables in a single vector autoregressive. The possibility of spurious regression is solved by the stationarity of the error correction mechanism term, as already noted. The null substitution hypothesis was rejected for the NovoMercado. Table 2 shows that all the models which considered debt presented a positive association between debt and stock market value, with strong significance in most of them and with the tests pointing out the correct specification of the models that were controlled by traditional determinants of capital structure.
Such a finding is contrary to those of Demirgüc-Kunt and Maksimovic (1996), Booth et al. (2001), and Demirgüc- Kunt et al. (2013), which found a retraction of bond markets with the expansion of the stock market. These authors argue that the bigger companies would benefit from stock market development in developing countries and that the economic growth would generate a retraction of bond markets in developed and developing countries.
The rejection of the hypothesis becomes more clear when the coefficients of different samples are matched. The observed association for the full sample and the rest of the segments' samples is negative and significant, except for the restricted model, which should be taken cautiously (Asteriou & Hall, 2011).
Whilst all the models point out complementarity between debt and stock market value for the companies belonging to the NovoMercado, the opposite (substitution) is observed in the segments where there would be less disclosure and shareholder protection.
Analyzing the size effect, the argument that bigger companies would have benefits from stock market development in the presence of developing stock markets (Demirgüc-Kunt & Maksimovic, 1996) was not fully validated, because the size variable presented negative impacts in the NovoMercado compared to most models. The negative impact appears not only to reject the suggestion of a supposed benefit, but also to suggest an advantage of smaller companies within NovoMercado.
The findings are not consistent with the observation of Titman and Wessels (1988) that smaller companies would have higher bond and stock issuing costs and would use mainly short-term debt.
The comparison of the findings for NovoMercado with those of other samples amplifies their relevance. In other samples, the positive and significant association between size is strongly observed, confirming the results of other studies (Demirgüc-Kunt & Maksimovic, 1996;Booth et al., 2001;Demirgüc-Kunt et al., 2013;Titman & Wessels, 1988), which have indicated the importance of size for obtaining benefits from stock market development.
The positive association between bond and stock markets also is consistent with the irrelevance of capital structure theory (Modigliani & Miller, 1958), as the simultaneous issuance of debt and stocks would be irrelevant for the stock market value. On the other hand, the positive association is not consistent with the trade--off (Modigliani & Miller, 1963) and pecking-order (Myers, 1984;Myers & Majluf, 1984) theories, which hold the expansion of bond markets against stock market development, in the presence of tax shields and asymmetric information.
• ALEXANDRE RIPAMONTI • EDUARDO KAZUO KAYO • Asymmetric information appears to be relevant in the findings. The different sense for the variables in the NovoMercado and other samples could indicate that the corporate governance reform allowed better information disclosure and shareholder protection, as observed in studies where the relevance of shareholder protection acts were relevant for stock market development (Demirgüc-Kunt & Levine, 1996;Wurgler, 2000;De la Torre, Gozzi, & Schmukler, 2007;Levine & Zervos, 1998).
The tangibility variable doesn't show any association, for the full sample or the other segments' samples. The argument of Titman and Wessels (1988), that more tangibility leads to more favorable conditions for borrowers, leading companies to issue stocks, was not found to be significant, although a positive and marginally significant direction had been observed in those samples. The opposite, however, was observed for NovoMercado, were less tangible companies were related positively and significantly to stock market development, which confirms their argument (Titman & Wessels, 1988).
The growth variable presented a negative and significant association in most models of NovoMercado, but non-significant for the full sample and other segments. Such a result lends consistency to the argument of Titman and Wessels (1988) that growth would be negatively related to capital structure, even it seems contradictory to the result of the tangibility variable, which could also represent existing growth opportunities.
Additionally, such a result is not consistent with the studies that observed positive association between stock market development and economic growth (Levine & Zervos, 1996;Rousseau & Wachtel, 2000;Beck, Levine & Loayza, 2000;Tachiwou, 2010;Demirgüc-Kunt et al., 2013). However, they are consistent with the findings of Demetriades and Hussein (1996), although they refute the argument that inadequate chosen techniques by the cointegration.
The models' analysis of cointegration between leverage and return has relevant aspects, such as the fact that the debt and stock market value, directly measured, seem to better reflect the bond and stock market development.
According to the rank of cointegration that was found through maximum likelihood and trace statistics, neither of the models where leverage and return are have only one cointegration vector, instead of the quantity used in the estimates.
Otherwise, the perfect specifications were obtained for debt and stock market value where the trace and max statistics presented only one rank of cointegration, once the results indicated the existence of a rank (Asteriou & Hall, 2011).
In all samples, the return presented a negative and significant association with the leverage of companies, consistent with Demirgüc- Kunt and Masimovic (1996) and Demirgüc-Kunt et al. (2013). Size and growth presented positive and significant associations for the full sample and other segments' samples, just as tangibility, where only the impact was different for other segments. In the NovoMercado, the control variables presented several impacts and associations. Note: Table 2 presents the cointegration equations corresponding to long-run relationships between the variables of: debt and stock market value, e; debt on equity and return, both of them controlled for size, tangibility and growth. The relationship was estimated through the techniques of Johansen (1988;1991) and Johansen and Juselius (1990), in the models which contain intercept (constant); restrict intercept (rconstant); trend (trend); restrict trend (rtrend), e; without intercept and trend (none). Max and tr refers to the maximum likelihood and trace statistics, for computing the quantity of cointegration ranks, which result in specification of models' analysis (Pantula's principle). All the models were applied to full sample, to the companies of NovoMercado and for the set of other segments. In the models with debt, a lagged period was applied and 1 rank of cointegration, and in the model with debt on equity 4 lagged periods were applied and 1 rank of cointegration. Variables' details are presented in Chart 1. Statistical significance: t p.10; * p.05; ** p.01; *** p.001.
Source: Elaborated by the authors.

CONCLUDING REMARKS
The findings of this research show that corporate governance reform in Brazil simultaneously stimulated the stock bond market, confirming the stock and bond complementarity hypothesis in the Brazilian companies that moved to or were created in the NovoMercado.
Such results are contrary to those observed in other studies, but are consistent with the theoretical framework of capital structure irrelevance for NovoMercado. The bond market size increased in the same way as that of the stock market for this listing segment, making plausible the supposition that a company's stock value does not depend on its capital structure, but only on its projects' risks.
The substitution hypothesis, standard for developed countries, was not validated for the Brazilian stock market. A possible explanation for such an observation is that developing countries have more projects, such as infrastructure, housing, and real estate, to be explored than developed ones.
Even if capital structure determinants were the same in developing and developed countries, theories with assumptions of symmetric information, such as capital structure irrelevance (Modigliani & Miller, 1958) could offer consistency to the findings due to the reduction of asymmetric information that comes from the corporate governance reforms. Such reduction would stimulate the complementarity between stock and debt, which is reinforced by the findings for the NovoMercado. The direct measures of stock market value and debt showed better specification than debt on equity and return for capturing the level of stock and bond market development. Therefore, corporate governance reform could, with institutional characteristics, result in a virtuous cycle between stock and debt, making these sources complementary and not substitutes.
The data was analyzed with cointegration techniques, which could imply an improvement on the traditional tools applied in prior studies about stock market development and a stimulus of the understanding of dynamic econometric models. Several advantages found for techniques that estimate the relationship between variables apparently not associated, show that the subject of capital structure could be explored in ways beyond panel data techniques.
However, making aggregate variables from the weighting of firm-level variables, by theoretical portfolio linked to the liquidity issues could become a restriction. Such a restriction should be addressed with a cointegration panel in future research, which combines the advances in panel data with long-run and short-run relationships.
Future research could also specify models with variables for institutional characteristics and even for industries. The direct measures of asymmetric information could be used in the model specification in order to assess the stock market development theory. Of course, the NovoMercado would represent better information access to investors. Using the asymmetric information measure in this context, the association between stock and debt will be even more accurate.