affect corporate cost of debt around the word. It is well-known that corporate cost of debt is

Transcription

1 National Culture and cost of debt around the World Abstract This study investigates how Schwartz (1994) cultural dimensions, embeddedness and mastery, affect corporate cost of debt around the word. It is well-known that corporate cost of debt is positively related to bankruptcy risk and agency cost. We argue that firms in countries with stronger embeddedness or mastery tend to have a lower bankruptcy risk. Furthermore, while the agency conflicts are expected to be lower in high embeddedness countries, they are expected to be stronger in high mastery countries. On the other hand, while investors concern less with agency conflicts in countries emphasize on embeddedness, investors worry more about agency conflicts in countries emphasize on mastery. Therefore, we hypothesize that there is a negative relationship between embeddedness and corporate cost of debt across-country. However, the relationship between mastery and corporate cost of debt has to be determined empirically. Using firm-level data from twenty-nine countries over the years from 1993 to 2006, we find a strong and robust negative relationship between embeddedness and cost of debt. The estimated relationship between mastery and corporate cost of debt is insignificant. Our cross-country evidences also indicate that embeddedness is negatively related to bankruptcy risk and the sensitivity to agency activity, while mastery is negatively associated with bankruptcy risk and it is positively related to the sensitivity to agency activity. Keywords: National culture; Embeddedness, Mastery, Bankruptcy, Agency cost, Cost of debt

2 1. Introduction It is well-known that cost of debt is positively related to the default risk of a firm (i.e. the likelihood of a firm to meet its debt obligations). Previous studies on cost of debt indicate that this default risk is positively related to the likelihood of bankruptcy and the asymmetric information problem between shareholders and debt holders (Chen (1978) and Jensen and Meckling (1976)). The bankruptcy argument is intuitive. When a firm goes bankruptcy, it cannot re-pay its liabilities. A firm s bankruptcy risk depends on the corporate policies of this firm. For example, the literature in capital structure generally suggests that there is a positive relationship between debt ratio and bankruptcy risk (Castanias (1983) and Ohlson (1980)). The asymmetric information argument states that self-interested managers would take actions that expropriate the creditors through increasing the default risk and agency cost arises. 1 Hence, there should be a positive relationship between agency cost and cost of debt. Consistent with the asymmetric information argument, previous studies find that firms with more intensive disclosure are associated with lower cost of debt (Sengupta, 1998; Francis, Khurana, and Pereira, 2005). These findings suggest that firms with higher level disclosure tend to have less severe information asymmetry between the management and creditors and hence these firms have lower cost of debt. Furthermore, Bhojraj and Sengupta (2003) find that governance mechanisms help to reduce default risk by mitigating agency costs and hence lead to lower cost of debt. A long the same line, Anderson, Mansi, and Reeb (2003) document that founding family ownership tends to reduce the agency conflicts between the shareholders and the creditors and hence founding family ownership leads to a lower cost of debt. Although studies of cost of debt in the US are abundant, studies on this topic in international setting are rare. Using data from thirty-four countries, Francis, Khurana, and Pereira (2005) document that the level of disclosure is negatively related to cost of debt and this finding is consistent with the asymmetric information argument. The quality of institution is also an important factor that can explain the differences of cost of debt around the world. Bae and Goyal (2009) argue that stronger legal protection promotes contracting efficiency, which in turn results in lower cost of debt. 1 The agency cost defined by Jensen and Meckling (1976) includes contracting cost, transaction cost, moral hazard cost, and information cost. 2

3 Consistent with their hypothesis, they find that both enforceability of contracts and creditor rights are negatively related to loan spreads. In a recent study, Qi, Roth, and Wald (2010) document that political rights also help to reduce bond spreads. Does culture matter in cost of debt? Previous studies have not yet provided an answer to this question. The current study aims to fill this gap by examining how Schwartz cultural dimensions affect cost of debt around the world through bankruptcy risk and agency cost. Chui, Lloyd and Kwok (2002) argue that firms in countries with stronger embeddedness and/or mastery tend to have a higher tendency to avoid bankruptcy. 2 They find that firms in these countries choose to have a lower debt ratio so as to reduce their bankruptcy risk. On the other hand, national cultures also have an influence on the agency conflicts between managers and shareholders. Managers in countries with higher embeddedness emphasis on the positive relationship with the shareholders and they also tend to avoid free cash flow (Chui et al. (2002), Shao, Kwok, and Guedhami (2010)). As a result, the agency activities in high embeddedness countries are expected to be less and debt holders in these countries will be less concern with agency conflicts. On the contrary, strong mastery countries are associated with more severe agency conflicts. The reason is twofold. First, because of greater internal locus-ofcontrol, managers in strong mastery countries are more likely to adopt aggressive financial policies (Chui et al. (2002)). Second, because of the need of flexibility in financial choices, both managers and shareholders in strong mastery countries would prefer the firms to have more free cash flow (Shao et al. (2010)). Therefore, managers in high mastery countries have stronger tendency to participate in agency activities and hence debt holders in high mastery countries will be more concern with agency conflicts. Since both the bankruptcy risk and the sensitivity to agency conflicts are lower in countries with stronger embeddedness, we expect that there is a negative relationship between embeddedness and cost of debt across countries. Although the bankruptcy risk is lower in countries with more mastery, the sensitivity to agency conflicts is higher in these countries. As a result, the relationship between mastery and cost of debt is an empirical question. Using data across twenty-nine countries from 1993 to 2006, we find a strong negative 2 Schwartz (1994) uses the term conservatism, which is changed to embeddedness in his later works. 3

4 relationship between embededness and cost of debt. This relationship is robust even after taking into account the well-known determinants of cost of debt such as leverage ratio, return on assets, firm size, macroeconomic variables and variables of the institutional quality. The relationship between mastery and cost of debt, however, is insignificant. Using the ratio between the number of commercial bankruptcy filing and the total number of firms as a proxy for bankruptcy risk, we document that both embededness and mastery are negatively related to bankruptcy risk. We also document that while investors in countries with high scores on embeddedness have lower sensitivity to agency activities, investors in countries with high scores on mastery have higher sensitivity to agency activities. Although these relationships among cultural values and bankruptcy risk/agency conflicts are usually assumed in the previous literature, these relationships have not yet been tested. The rest of this paper is organized as follows. In Section 2, we will develop our hypotheses. In Section 3, we describe the data used in the paper. In Section 4, we investigate the relationship between embeddedness/mastery and bankruptcy risk/sensitivity to agency activities. In Section 5, we report our main results on the relationship between embeddeness/master and cost of debt. Finally, Section 6 concludes the paper. 2. The Development of Hypothesis It is well-known that cost of debt is increasing in bankruptcy risk and agency cost. In this Section, we argue how Schwartz s cultural dimensions, embededness and mastery, affect cost of debt through bankruptcy risk and agency cost. We focus on embededness and mastery because these two dimensions summarize the six values types that Schwartz (1994) used to classify cultures (Chui et al. (2002)). 2.1 Embededness and Cost of Debt Schwartz (1994) argues that individuals are viewed as entities embedded in collective society in embeddedness culture. Since people in embeddedness countries emphasize preserving the public image and security, managers in these countries tend to choose financial policies that can minimize the bankruptcy risk of their firms. The reason is intuitive. When bankruptcy risk increases, security of the firms drops. In addition, bankruptcy also puts the public images of both managers and owners 4

5 in risk. Consistent with this argument, Chui et al. (2002) find that the corporate debt ratio is lower in countries with high embeddedness. On the other hand, people in embeddedness countries also stress on the maintenance of harmonious working relationship. Because of this concern, managers in embeddedness countries care about the benefits of the firms stakeholders, such as employees and creditors (Chui et al. (2002)). Bankruptcy of a firm definitely will jeopardize the livelihood of its employees and disrupting their lives. Hence, managers in embeddedness countries would like to reduce bankruptcy risk. The emphasis on harmonious relationship also affects the agency cost. Jensen and Meckling (1976) show that managers tend to invest in risky projects that benefit the shareholders in the expense of the creditors through increasing the default risk of the firms. Jensen (1986) argues that managers may spend the free cash flow on their own welfares and on unprofitable projects. Since these activities hurt the benefits of both the shareholders and creditors, these activities are threats to harmonious relationship. Therefore, the stress on the harmonious relationship with the stakeholders limits expropriation by managers, which in turn leads to a greater avoidance to agency activities. Furthermore, self-discipline is also an important value in embeddedness countries. To reduce the incentive to participate in agency behavior, Shao et al. (2010) argue that more selfdisciplined managers in strong embeddedness countries tend to avoid free cash voluntarily. Since managers in strong embeddedness countries tend to avoid agency activities, investors in these countries will worry less about agency activities which results in a lower agency cost. In summary, it is expected that bankruptcy risk and investors concern about agency activities are lower in countries with higher scores on embeddedness, which should lead to a lower cost of debt in these countries (path (1) in Figure 1). Our hypothesis is therefore formulated as follows: Hypothesis: The cost of corporate debt of a country is negatively related to the country s level of embeddedness. 2.2 Mastery and Cost of Debt In high mastery society, people have greater emphasis on individual success, internal locus of control, and independence. 3 Since bankruptcy is regarded as failure, managers in countries with 3 Locus of control is a one of the most important concepts in social psychology. When a person is said to have 5

6 strong mastery tend to avoid bankruptcy, which in turn leads to a lower bankruptcy risk and a lower cost of debt. When a person believes that events can be controlled through one s own efforts, this person is said to process an internal locus of control. Chui et al. (2002) argues that in countries with strong mastery, managers tend to adopt aggressive strategies so as to demonstrate their ability to control external environment. 4 In addition, mastery managers prefer independence and hence they desire to have more financial resources under their control in order to implement their aggressive strategies whenever opportunities arise (Shao et al. (2009)). As a result, it is expected that managers are more likely to participate in agency activities in countries with stronger mastery. Therefore, investors in these countries will worry more about agency activities which leads to a higher agency cost. Since cost of debt is increasing in bankruptcy risk and agency cost, the relationship between mastery and cost of debt is ambiguous (path (2) in Figure 1). 3. Data This study involves quite many variables and Appendix A shows the detailed definitions as well as their sources. This Section briefly discusses those variables that are related to cost of debt. Scores on Schwartz s cultural dimensions, embeddedness (Embd) and mastery (Mast), on fifty countries are collected from Lichi, Goldschmidt, and Schwartz (2007). The financial data are from Compustat Global. This database represents over 90% of the world's market capitalization. 5 To be included in our sample, each firm should have sufficient data to compute its cost of debt. Following Francis et al. (2005), we measure cost of debt of a firm in year t as the ratio between its interest expenses and average total debt. 6 The average total debt of a firm in year t is computed from its total debts in year t-1 and year t. Since firms in the financial and utility industries are heavily regulated, the cost of debt for these firms is largely influenced by government policies. Therefore, as in previous studies in cost high internal locus of control, this person has a strong belief that he can control the events surrounding him. 4 Apart from embeddedness, Chui et al. (2002) also suggest that the preference for internal locus of control make the managers in mastery countries to adopt a lower debt ratio, which leads to a lower bankruptcy risk. They find evidence indicates that corporate debt ratio is lower in countries with higher score on mastery. 5 Compustat Global database covers 90% of the Asian market capitalization, 90% of the Indian market capitalization, 95% of the Taiwanese market capitalization and 95% of the European market capitalization. 6 To alleviate the effect of outliers, any of the accounting ratios that is larger than (less than) its respective 99 th percentile (1 st percentile) in each year is assigned with the value of its 99 th percentile (1 st percentile) in that year. 6

7 of debt, financial and utility firms are excluded from our sample. We measure the national cost of debt of a country j in year t (Cod jt ) as the cross sectional average of individual cost of debt in country j in that year. To ensure that the national cost of debt of a country can represent the average cost of debt in that country, each country is required to have at least fifteen firms in each year over our sample period from 1993 to Furthermore, each country in our sample should have scores on Embd and Mast. We also require each sample country to have valid observations on our control variables that are possible determinants of cost of debt across countries. As a result, our final sample consists of data on about 110,000 firms across twenty-nine countries. However, because of data availability, seven countries in our sample have their sample periods less then fourteen years and, among them, only Argentina has its sample period less than ten years. Table 1 shows the median number of firms per year and the average national cost of debt. As expected, about 47% the firms in our sample come from Japan and the United States. To avoid our findings being driven by observations from Japan and the United States, our analyses are mainly based on country-year observations rather than firm-year observations. (Insert Table 1 here) In a recent study on international cost of debt, Francis et al. (2005) suggest that national cost of debt is related to firm-level variables such as leverage ratio, profitability, and firm size, as well as country-level variables on disclosure, investor protection, and macroeconomics conditions. Using data from twenty-nine countries, Francis et al. (2005) find that while cost of debt is negatively related to firm size and disclosure level, it is positively related to interest rate. In this study, we compute the leverage ratio, profitability, and size of firm i in year t, respectively, as total liabilities dividing by this firm s total assets in year t (Lev it ), net income dividing by this firm s total assets in year t (Roa it ), and the natural logarithm of this firm s total sales in year t (Size it ). Our profitability measure is commonly known as the return on assets. The national leverage, profitability, and firm size of country j in year t are, respectively, the cross sectional average of individual Lev it, Roa it, and Size it of this country in that year. Following prior studies on disclosure (e.g. Sengupta, 1998, Francis et al., 2005), we use 7

8 disclosure index developed by CIFAR to proxy for the country disclosure level (Disc). 7 In our study, we use CIFAR 2005 that reports the disclosure data in We use a private property rights index (ProR jt ) as a proxy for investor protection. The score on this index in a given month is the sum of the indexes on corruption and investment profile. The investment profile index is the sum of three subindexes:- contract viability/expropriation, profit repatriation, and payment delays. These monthly indexes are collected from International Country Risk Guide. ProR jt is the average of the monthly private property rights index in year t in country j. To construct this annual index on private property rights, each country is required to have no missing observations on its monthly private property rights index in each year. 8 A higher score on this private property rights index indicates a higher degree of investor protection. Using data on loans to borrowers in forty-eight countries, Bae and Goyal (2009) document that interest rate spread on loans is decreasing in a similar private property right index. We use the annual lending rate that applies to the short- and medium-term financing for private sector in country j in year t as our measure of this country s interest rate (Int jt ). The interest rate data is from the International Financial Statistics of the International Monetary Fund. We use the ratio between private credit by deposit money banks and other financial institutions and GDP of country j in year t (Bank jt ) as a measure for a country s banking sector development. Furthermore, national cost of debt may be related to the degree of openness of a country s financial market to foreign investors. We use an index on control of capital flows restrictions (Control j ) and the average language dummy variable (Lang j ) as proxies for financial market openness. The latter variable measures how foreign investors are familiar to a country s financial markets. It is expected that firms can access lower cost of capital when there is less control of capital flows in a financial market and/or when foreign investors are more familiar to this market. The data on these two variables are obtained from Chan, Covrig, and 7 CIFAR generate disclosure score by rating the annual reports of list firms in a country inclusion or omission of 90 items. The score fall into seven main categories, including general information, balance sheet, income statement, fund flow statement, accounting standard, stock data and specific items. CIFAR disclosure score includes both mandated and voluntary disclosures elements. 8 The construction of our private property rights index is similar to that of the good government index used in Morck, Yeung, and Yu (2000). The good government index is constructed from the ICRG indexes on corruption, risk of expropriation of private investment, and risk of repudiation. However, the latter two indexes are not available from ICRG after Bae and Goyal (2009) construct their private property rights index based on the above three ICRG indexes and they keep the index scores on risk of expropriation of private investment and risk of repudiation constant after We find that the correlation coefficient between our private property rights index and the good government index used in Morck et al. (2000) is

9 Ng (2005). We use the real GDP per capita (Gdppc jt ) and the average growth rate of real GDP per capita (Gdpgw jt ), computed from years t-4 to t, as measures of national wealth and economic development of country j in year t. Since Fama and French (1989) find that dividend yield can forecast bond returns and is related to business cycle, we also include the dividend yield of country j in year t (Dy jt ) as a control for cost of debt. The data on dividend yield of a country is obtained from Datastream index on this country. Following Chui et al. (2010), we use the change in the exchange rates (Cfx jt ), measured as the average change of the monthly exchange rate calculated from years t-4 to t, as a proxy for the uncertainty in monetary policy and inflation rate in country j in year t. 9 In addition, Bae and Goyal (2009) find that interest rate spread on loans and creditor rights are negatively related across countries. As in Bae and Goyal (2009), we collect the annual creditor rights index (CreR) from Djankov, McLiesh, and Shleifer (2007). This index measures the legal rights of creditors against defaulting borrowers in each country. Since this index is not available after 2003, therefore we keep the scores on this index constant in the post 2003 period. Recently, Qi et al. (2010) find that bond yield spread is negatively related to political rights across thirty-nine countries. As suggested in their study, we collect political rights index (PolR) from La Porta, Lopez-de-Silanes, Schleifer, and Vishny (1999). Furthermore, using a sample of firms in the United States, Anderson et al. (2003) document that founding family ownership decreases cost of debt. As for founding family firms, the major share holders of closely held firms are likely to be long-term investors and their investments tend to be undiversified. Hence, firms with closely held ownership may share similar incentive structure between shareholders and creditors as firms with founding family ownership. Therefore, we expect that closely held ownership may help to explain the variation in cost of debt across countries. We collect the closely held ownership index (Own) from Dahlquist, Pinkowitz, Stulz, and Williamson (2003). On the other hand, Fan, Titman, and Twite (2010) argue that the relative benefit of using debt instead of equity should be lower in either the dividend relief tax system or the dividend imputation tax system. 10 Therefore, it is expected that the cost of debt is higher in these two tax systems. To control for the tax system effect on cost of debt across countries, we 9 The change in the exchange rate of the US dollar is always zero. 10 Dividend payments are not subject to double taxations in the dividend relief tax system and the dividend imputation tax system (Fan et al. (2010)). 9

10 include two dummy variables on tax systems in our analysis. The dummy variable Tax 1 ( Tax 2 ) takes the value of one for a dividend relief tax country (a dividend imputation tax country) and it is zero, otherwise. The classification of tax systems is obtained from Fan et al. (2010). All the variables are updated annually except for Embd, Mast, Control, Lang, Disc, PolR, Own and the tax systems dummy variables. These nine variables are varying across countries but they are constant across time. It is expected that while Lev, Int, Tax 1, and Tax 2 are positively related to cost of debt, Embd, Roa, Size, ProR, CreR, Disc, PolR, and Own are negatively related to cost of debt. The relationships between cost of debt and other variables, however, have to be determined empirically. Table 2 presents the correlation coefficients among these variables and cost of debt. We find that most of the correlation coefficients are significant at the 1% level. We notice that while the correlation coefficient between Embd and cost of debt is significantly positive, the correlation coefficient between Mast and cost of debt is significantly negative. However, these findings are likely the result of the fact the cultural variables are strongly correlated with the firm-specific variables. 11 To explore this possibility, we construct a measure on cost of debt that is adjusted for firm-specific variables. Specifically, we use OLS to estimate the following regression model in each country-year. Cod ijt = α ojt + α 1jt Roa ijt 1 + α 2jt Lev ijt 1 +α 3jt Size ijt 1 + ε ijt, (1) where i,j, and t are indexes for firm, country, and year. Cod, Roa, Lev, and Size are, respectively, cost of debt, return on asset, leverage, and firm size. Adjusted cost of debt of firm i in country j in year t (AdjCod ijt ) is computed as the sum of estimated intercept ( α ojt ) and the estimated residual (εijt ). Accordingly, the adjusted national cost of debt is the estimated intercept ( α ojt ). By construction, adjusted cost of debt is unrelated to the firm-specific variables in each country-year. After controlling for the firm-specific variables, we find that the correlation coefficients between AdjCod and the cultural variables are significantly negative. The findings from the correlation analysis suggest that cultural values and other potential determinants of cost of debt are correlated. Hence, to estimate the 11 For example, firm size is negatively related to both cost of debt and Embd and these negative relationships may lead to a positive relationship between Embd and cost of debt.. 10

11 relationship between cost of debt and cultural values, we need to control for other possible determinants of cost of debt. (Insert Table 2 here) 4. Culture, Bankruptcy Risk, and Avoidance to Agency Activities Our cultural arguments suggest that bankruptcy risk and avoidance to agency activities across countries are related to both embeddedness and mastery. We investigate these relationships in this Section. Following Claessens and Klapper (2005), we measure bankruptcy risk as the ratio between the total number of bankruptcy filings and the total number of firms. To examine the relationship between cultural values and bankruptcy risk, we estimate the following empirical model. Risk jt Embd 5 1 Gdppc j jt 1 Mast 6 2 Gdpgw j Law Rule jt 1 1 Crisis 7 j 2 jt 1 j, CreR jt 3 jt 1 Int 4 jt 1 (2) where subscripts j and t represent, respectively, country j and year t. Risk, Embd and Mast are the bankruptcy risk, embeddedness, and mastery respectively. To control for the other determinants of bankruptcy risk across countries, we include those variables suggested by Claessens and Klapper (2005). 12 These variables are the rule of law (Rule), creditor right (CreR), the lending interest rate (Int), the natural logarithm of real GDP per capita (Gdppc), the average real GDP growth rate (Gdpgw), and a dummy variable on the systemic banking risk (Crisis). Claessens and Klapper (2005) suggest these variables because of the following reasons. Rule represents the enforcement of legal rules and it is expected that a more strong legal enforcement is associated with more bankruptcy filings. In countries with stronger creditor rights, lenders would be more easily to negotiate debt restructuring with the borrowers. Therefore, bankruptcy filing is negatively related to creditor rights (CreR). Both interest rate (Int) and the systemic banking risk dummy (Crisis) are expected to have positive influences on bankruptcy filings because these variables increase the cost of financing. The variables related to GDP are used to measure the general development of a country s economy and their impacts on bankruptcy filings are uncertain. La Porta et. al. (1998) document that common law 12 We thank Luc Laeven in providing us with the bankruptcy data used in Claessens and Klapper (2005). Claessens and Klapper (2005) collected bankruptcy data from thirty-five countries over the years However, only twenty-four countries out of these thirty-five countries have scores on the cultural values. As a result, the sample of this test consists of twenty-four countries and spans over the years from 1990 to

12 countries provide better investor protection and more efficient judiciary system than the civil law countries. Thus, the transaction cost of bankruptcy in the common law system is expected to be lower than that in the civil law countries. Hence, common law system may generate more bankruptcy filings than the civil law system. We, therefore, include a common law dummy variable (Law) in our estimation. This variable Law takes the value of one for common law countries and it is zero, otherwise. Following Claessens and Klapper (2005), we use OLS with robust standard errors to estimate Equation (1). Since bankruptcy risk is likely to have strong time persistence, we use Petersen (2009) method to compute the robust standard errors of the estimates clustered by country and year. Consistent with the results in Claessens and Klapper (2005), we find that while bankruptcy risk is increasing with systemic banking risk and rule of law, it is decreasing in creditor rights. We also find that the estimated coefficients on interest rate and the common law dummy variable are positive and significant. These findings indicate that higher interest rate and more efficient judiciary system lead to more bankruptcy filings. Our most interesting finding is related to the cultural values. The estimated coefficients on embeddedness and mastery are negative and significant. This finding supports our claim that bankruptcy risk is lower in countries with high scores on embeddedness and mastery. Our finding suggests that when a country s score on embeddedness (mastery) increases from the value at the 10 th percentile to that at the 90 th percentile, bankruptcy risk will decrease by about 2.52% (1.76%) that is about 1.4 times (1 time) of the standard deviation of the average bankruptcy risk across countries. 13 Table 3 shows these findings. (Insert Table 3 here) To examine the relationship between culture and the avoidance to agency activities, we investigate the relationship between propensity to engage in agency activities (PA, the possibility that a manager may re-allocate funds from low risk projects to high risk, sub-optimal projects) and leverage ratio (Lev) across countries. Previous literature on capital structure suggests that this relationship should be negative. Prowse (1990) argues that in a country where investors are more 13 The average bankruptcy risk is ranged from 0.02% to 7.91%. The mean and standard deviation of average bankruptcy risk are 1.97% and 1.77%, respectively. 12

13 concerned with agency problem, the link between leverage ratio and propensity to engage in agency activities should be more negative. Using data from Japan and the United States, Prowse (1990) finds that investors in the U.S. are more sensitive to agency conflicts then the investors in Japan. Since people in embeddedness (mastery) countries worry less (more) with agency conflicts, we expect this relationship to be more profound in mastery countries and it is less significant in embeddedness countries. To test these conjectures, we employ the two-stage regression model suggested by Prowse (1990). In the first stage, we use OLS to estimate the following regression model for each countryyear in our sample. Lev ijt = α jt + β 1jt PA ijt 1 + β 2jt Oprisk ijt 1 + β 3jt Profit ijt 1 +β 4jt Tax ijt k=1 γ kt Ind kjt + ε ijt, (3) where the subscripts i,k, j, t represent respectively firm i, industry k, country j, and year t. Lev, PA, Oprisk, profit, tax, and Ind are respectively the leverage ratio, the propensity to engage in agency activity, the operating risk, the profitability, the effective tax rate, and industry dummy variable. To measure the propensity to agency activities (PA), we use the inverse of asset utilization ratio. The asset utilization ratio, which is total assets divided by total sales, is commonly used as a proxy for agency cost in previous accounting and finance literature (Ang, Cole, and Lin (2000). This ratio indicates how efficient a manager is in generating sales from total assets. Agency problem arises when managers allocate funds on nonprofit-maximizing projects intentionally and this behavior reduces the asset utilization ratio (i.e. increases PA). Since creditors are reluctant to lend money to firms with severe agency problem, PA is negatively associated with leverage ratio (Lev). Previous literature on capital structure also suggests that leverage ratio is decreasing in business risk and profitability (for example Prowse (1990)). Business risk of a firm in year t is calculated as the absolute value of the coefficient of variation of its interest coverage, which is operating income divided by total interest expenses. This coefficient is estimated from years t-4 to t and each firm is required to have no missing observation on interest coverage in each estimation period. We use return on assets, which is the ratio between net income and total asset as a proxy for a firm s profitability. To capture the tax shield of debt, we also include the effective tax rate, which is the ratio 13

14 between income taxes and operating income, in Equation (3). The industry dummy variables in Equation (3) are used to control for the industry effect on debt ratio. Following Chui et. al. (2002), firms are classified into four industries and they are primary industry, manufacturing industry, advanced manufacturing industry, and services industry. Using the estimated coefficient on PA (β 1jt ), we construct a measure of sensitivity to agency conflict (SAC jt ) for country j in year t as β 1jt. A larger value of SAC jt indicates that investors worry more about agency conflict. 14 In the second stage, we regress SAC jt on embeddedness and mastery as well as other possible variables that may affect the sensitivity to agency conflict. Below is our empirical model. SAC jt = α o + α 1 Embd j + α 2 Mast j + α 3 Own j + α 4 CreR jt + α 4 Bank jt + α 5 Mktcap jt + ε jt, (4) where Embd j, Mast j, and Own j are respectively the embeddedness index, the mastery index, and the index on closely held ownership for country j. CreR jt, Bank jt, and Mktcap jt are creditor rights, banking sector development, and stock market development in country j in year t, respectively. We expect that while embeddedness has a negative impact on the sensitivity to agency conflict, mastery has a positive influence on this sensitivity. Anderson et al. (2003) find evidences indicating that closely held ownership may reduce agency conflict between shareholders and creditors. Therefore, we expect that investors of closely held ownership firms concern less with the agency conflict. Since Brockman and Unlu (2009) document that creditor right helps to diminish agency conflict, we expect that investors in countries with strong creditor rights worry less about agency conflict. Khurana, Martin, and Pereira (2006) argue that firms in less developed financial markets tend to have more cash holdings because of the relatively more expensive external financing. We use the development of the banking sector (Bank jt ) and the stock market (Mktcap jt ) to proxy for the development of a financial market. On the one hand, more mature financial markets will have stronger monitoring systems, thus reducing investors concern with agency activities. On the other hand, more developed financial markets tend to be more profit-oriented and managers may have stronger incentive to participate in agency activities. Therefore, investors in more developed financial markets may worry 14 The average estimated coefficients on PA, Oprisk, Profit, and Tax are (-5.93), (-1.03), (-11.23), and (-3.71), respectively. The t-statistics are in the parentheses. To compute these averages, we first calculate the country means of these estimates and then compute the cross-country averages. 14

15 more about agency conflicts. As a result, the effect of the development of financial market on the sensitivity to agency activities is uncertain. Consistent with our expectation, while the estimated coefficient on Embd is significantly negative, the estimated coefficient on Mast is significantly positive. The estimated coefficients on other variables, however, are insignificant. It is worthwhile, moreover, to explore alternative explanations for the variation of sensitivity to agency activities across countries in future research. Our result indicates that while investors in low embeddedness countries worry more about agency conflicts, investor in high mastery countries concern more with agency conflicts. Specifically, when a country s score on embeddedness (mastery) increases from the value at the 10 th percentile to that at the 90 th percentile, sensitivity to agency activities will decrease (increase) by about 83% (34%) of one standard deviation of the average sensitivity to agency activities across countries. 15 Table 4 shows the findings. (Insert Table 4 here) 5. Culture and Cost of Debt To investigate the relationship between cultural values and adjusted cost of debt across countries, we use OLS to estimate the following regression model. AdjCod jt = α o + β 1 Embd j + β 2 Mast j + F j γ 1 + A jt 1 γ 2 + ε jt, (5) where subscripts j and t represent country j and year t. AdjCod is the cost of debt adjusted for firm characteristics. These characteristics are firm size, return on assets, and leverage ratio. Embd, and Mast are, respectively, embeddedness, and mastery. F is a vector of variables that are constant across time. These variables are the disclosure index (Disc, a higher score indicates a better transparency), the political right index (PolR, a higher value indicates a higher right), the index on closely held ownership (Own, a higher values indicates more closely held ownership), the index on capital flow restriction (Control, a higher value indicates more restrictions), average language dummy variable (Lang, a higher value indicates a higher level of familiarity to foreign investors), and two dummy variables for the dividend relief tax system (Tax 1 ) and the dividend imputation tax system (Tax 2 ). A is 15 The average sensitivity to agency activities (SAC) is ranged from to The mean and standard deviation of SAC are, respectively, and

16 a vector of variables that updated annually and we use the values of these variables in year t-1 to predict the adjusted cost of debt in year t. These annual variables are the real per capita GDP (Gdppc), average per capita GDP growth rate (Gdpgw), the dividend yield (Dy), the lending interest rate (int), the bank sector development (Bank), the average change of exchange rate (Cfx), the private property right index (ProR), and the creditor right index (CreR). ε is the error term. Since cost of debt is strongly time persistent, we use the Petersen (2009) procedure to compute the standard errors clustered by country and year. Table 5 reports the findings. (Insert Table 5 here) Consistent with the finding of Qi et al. (2010), we find that the estimated coefficient on PolR is negative and significant. The estimated coefficient on Own is significantly negative. This finding is consistent with Anderson et al. (2003) and suggests that closely held ownership may help to reduce the agency conflict between shareholders and creditors. Consistent with our expectation, the estimated coefficient on Tax 1 is significantly positive. This result indicates that cost of debt is higher in countries using the dividend relief tax system and it is consistent with Fan et al. (2010) which argue that the benefit of debt is lower in this tax system. Consistent with the finding of Francis et al. (2005), the estimated coefficient on Int is positive and significant. We find that the coefficient on Disc is negative but insignificant and this finding is inconsistent with the finding in France et al. (2005). We also find that the estimated coefficients on ProR and CreR are insignificant and these findings are inconsistent with the findings in Bae and Goyal (2009). Among the macroeconomic variables, we find that while the estimated coefficient on real per capita GDP (Gdppc) is significantly negative, that on the average change in exchange rate (Cfx) is significantly positive. These findings suggest that the cost of debt is lower in wealthier countries and it is higher in countries with more uncertainty monetary policy and inflation rate. After controlling for the possible determinants of cost of debt across countries, we find that the estimated coefficient on Embd is significantly negative and the estimated coefficient on Mast is insignificant. The first finding is consistent with our expectation that cost of debt should be lower in embeddedness countries because of lower bankruptcy risk and less agency conflict in these countries. The second finding is also consistent with our expectation that the negative effect of bankruptcy risk 16

17 on cost of debt is offset by the positive impact of sensitivity to agency activities on cost of debt in mastery countries. The 10 th percentile, median, and 90 th percentile scores on embeddedness in our sample are respectively 3.18, 3.55, and Our result indicates that if the country increases its score on embeddedness from the l0th percentile to the median (90 th percentile), this country s adjusted cost of debt will decrease by 3.26% (7.83%). These changes are economically significant. 16 To explore the robustness of our finding, we implement a couple of additional tests. First, we use OLS to estimate Equation 5 in firm-year level and Panel A of Table 6 shows the findings. 17 Consistent with the result in Table 5, the estimated coefficient on Embeddedness (Embd) is significantly negative and that on Mastery (Mast) is negative but insignificant. The findings on the other variables but the change in exchange rate (Cfx) and the private property rights index (ProR) are the same as that reported in Table 5. While the estimated coefficient on Cfx is positive but insignificant, the estimated coefficient on ProR becomes significantly positive. One should be noticed that the results from the above firm-year level regression could be driven by the firms listed in Japan and the United States. To explore this possibility, we re-estimate Equation 5 using firm-year data from countries other than Japan and the United States. In this smaller sample, we still find that the estimated coefficient on Embd is negative and significant. The estimated coefficient on Mast is insignificant. The estimated coefficients on Disc, PolR, and Own continue to be negative and significant and the estimated coefficient on Int continues to be significantly positive. However, the estimated coefficients on Tax 1, Gdppc, Dy, and Cfx become insignificant and the estimated coefficient on Control becomes significantly negative. Panel B of Table 6 reports the findings. (Insert Table 6 here) To investigate if our results are robust to alternative estimation method, we estimate Equation 5 using random effect model. We find that while the estimated coefficient on Embd and PolR are significantly negative, that on Tax 1 and Int are significantly positive. The estimated coefficients on 16 The mean and standard deviation of the average adjusted cost of debt across countries are 20% and 14.6%, respectively. 17 As in previous tests, we use the Peterson (2009) method to compute the standard errors of the estimates clustered by country and year for the robustness test, except for the random effect model. The variancecovariance matrix of the error terms estimated from the random effect model is adjusted for heteroskedasticity and within country serial correlation. 17

18 other variables are insignificant. We also notice that in all the regressions using adjusted cost of debt as the dependent variable, the magnitudes of the estimated coefficient on Embd are about the same and they are ranged from to To examine if our findings are sensitive to our two-step regression approach, we estimate the following variant form of Equation 5: Cod jt = α o + β 1 Embd j + β 2 Mast j + F j γ 1 + A jt 1 γ 2 + ε jt, (6) where the dependent variable is cost of debt (Cod) instead of the adjusted cost of debt. To control for the firm characteristics, we include the average leverage ratio (Lev), the average return on assets (Roa), and the average firm size (Size) in the matrix of annual variable, A. The other variables in Equation 6 are the same as that in Equation 5. The results reported in Panel D of Table 6 indicate that while the estimated coefficient on Embd, Own, Gdppc, ProR, and Roa are significantly negative, that on Control and Int are significantly positive. Our result indicates that if the country increases its score on embeddedness (Embd) from the l0th percentile to the median (90 th percentile), this country s cost of debt will decrease by 1.55% (3.74%) that is about 0.4 (1) standard deviation of the average cost of debt across countries. Overall, our findings suggest that there is profound and robust negative relationship between embeddedness and cost of debt across countries. Among the other possible determinants of cost of debt across countries, the most robust finding is the positive relationship between interest rate and cost of debt. 6. Conclusion In this paper, we conduct an international study on whether national culture help to explain the variation of corporate cost of debt around the world. We argue that firms in countries with high scores on embeddedness tend to have lower bankruptcy risk and investors in these countries have less concern with agency activities of the managers. Therefore, comparing to countries with low scores on embeddedness, firms in high embeddedness countries have lower cost of debt. On the other hand, although firms in countries with high scores on mastery tend to have lower bankruptcy risk, investors in these countries worry more about agency activities of the managers. As a result, the relationship between corporate cost of debt and mastery is ambiguous. We find evidences indicate that bankruptcy risk is indeed lower in countries with high scores 18

19 on embeddedness and/or with high scores on mastery. Using asset utilization ratio as an inverse measure on the propensity to engage in agency activities, we find that while investors in high embeddedness countries have less concern about agency conflicts, investors in high mastery countries worry more about agency conflicts. These results provide support to the conjectures usually made in previous studies in culture and finance (for example Chui ei al. (2002) and Shao et al. (2010)). However, apart from the cultural values, it would be interesting in future research to find out other possible determinants of sensitivity to agency activities across countries. Consistent with our hypothesis, we find a strong negative relationship between embeddedness and corporate cost of debt across countries. This result is robust to alternative model specifications and estimation method. We also document that cross-country variations in cost of debt are positively related to the lending interest rates and average change in exchange rate and they are negatively related to political rights index, closely held ownership index, real per capita GDP, and a dummy variable for the dividend relief tax system. Among all the possible determinants of the variations of corporate cost of debt across countries, we find that embeddedness and interest rate have the most robust impact on corporate cost of debt across nations. Our result suggests that if a government in a low embeddedness country would like to lower its interest rate in order to reduce the cost of debt, this government needs to implement a stronger expansionary monetary policy than a government in a high embeddedness. It would be worthwhile to investigate the relationship between culture and the effectiveness of monetary policy in future research. 19

20 References Anderson Ronald C., Sattar A. Mansi, and David M. Reeb, 2003, Founding family ownership and the agency cost of debt, Journal of Financial Economics 68, pp Ang, James S., Rebel A. Cole, and James Wuh Lin Agency costs and ownership structure, Journal of Finance, 55(1): Bae, K.H., & Goyal, V. K Creditor Rights, Enforcement, and Bank Loans, Journal of Finance, 64(2): Beck, Thorsten and Ed Al-Hussainy, 2010, Financial Structure Dataset, The World Bank. Bhojraj, S., & Sengupta, P Effect of corporate governance on bond ratings and yields: the role of institutional investors and the outside directors, The Journal of Business 76(3): Brockman Paul and Emre Unlu Dividend policy, creditor rights, and the agency costs of debt, Journal of Financial Economics 92: Castanias, Richard, 1983, Bankruptcy risk and optimal capital structure, The Journal of Finance 38, Chan, Kalok, Vicentiu Covrig, and Lilian Ng What determines the domestic bias and foreign bias? Evidence from mutual fund equity allocations worldwide. Journal of Finance 60: Chen, A. H Recent developments in the cost of debt capital. Journal of Finance, 33(3): Chui, Andy C.W., Lloyd, A.E., & Kwok, C.C.Y The determination of capital structure: Is national culture a missing piece to the puzzle? Journal of International Business Studies, 33(1): Chui, Andy C.W., Sheridan Titman, and K.C. John Wei Individualism and momentum around the world. Journal of Finance, 65: Claessens, S. & Klapper, L. F Bankruptcy around the World: Explanations of Its Relative Use. American Law and Economics Review, 7(1): Dahlquist Magnus, Lee Pinkowitz, Rene M., Stulz, and Rohan Williamson Corporate governance and the home bias. Journal of Financial and Quantitative Analysis, 38: Djankov Simeon, Caralee McLiesh, and Andrei Shleifer Private Credit in 129 countries. Journal of Financial Economics, 84: Fama, Eugene F., and Kenneth R. French, 1989, Business conditions and expected returns on stocks and bonds, Journal of Financial Economics, 25(1): Fan, Joseph P.H., Garry J. Twite, and Sheridan Titman An international comparison of capital structure and debt maturity choices. Journal of Financial and Quantitative Analysis, forthcoming. Francis, J.R., Khurana, I.K. & Pereira, R Disclosure incentives and effects on cost of capital around the world, The Accounting Review 80 (5):