Heterogeneous effect of Financial Leverage on Corporate. Yu-Yen Ku 1. Tze-Yu YEN

Transcription

1 Heterogeneous effect of Financial Leverage on Corporate Performance:A quantile regression analysis of Taiwanese companies Yu-Yen Ku 1 Abstract Lecturer, Department of Finance, National Chung Cheng University Ph.D. Candidate, Department of Finance, National Chung Cheng University Tze-Yu YEN Assistant Professor, Department of Finance, National Chung Cheng University This study further expand previous research by assessing the heterogeneous impacts of financial leverage on corporate performance using a sample of 323 non-financial publicly traded companies in Taiwan. In view of the controversial empirical findings in the literature and the restrictions of the least squares regressions, we adopt the method of quantile regression (QR) and report a robust and significant negative relation between leverage and corporate performance among, and only among, the less profitable firms. This new finding, which the conditional mean-focused regressions (OLS) do not capture, suggests that while corporate executive manager still exerts influences on the performance of these firms, the relationship parameter changes across quantiles of the distribution of performance variables. The research provides a more comprehensive understanding of the connection between financial leverage and corporate performance. JEL classification: G34 Keywords: Financial leverage; Corporate performance; Quantile regression 1 Corresponding author. Department of Finance, National Chung Cheng University, 168 University Rd., Min-Hsiung, Chia-Yi, 62102, Taiwan. address: yuyangu_thu@yahoo.com.tw 1

2 1. Introduction Ever since Modigliani and Miller (1958), proposing that capital structure of corporate is irrelevant to firm value, the topic of capital structure has been investigated extensively. Nevertheless, Jensen and Meckling (1976) struggle this point and debate that the effect of financial leverage in firm s capital structure has an influence on managers financial decisions and that these decisions in proper order affect the corporate performance. The topic of the effect of financial leverage on corporate performance has been quite disputed and related in the corporate finance literature. Existing empirical studies have reported ambiguous results on this issue. For example, Jensen and Meckling (1976), Myers (1977), Balakrishnan and Fox (1993), Pushner (1995), Kinsman and Newman (1999), Majumdar and Chhibber (1999), Gleason et al. (2000) and Simerly and Li (2000) find negative correlations between leverage and corporate performance. But Spence (1985), Jensen (1986), Lubatkin and Chatterjee (1994), Nickell et al. (1997), Nickell and Nicolitsas (1999) and Ghosh et al. (2000) find a positive correlation. Moreover, some researchers like Harris and Raviv (1991) and Ghosh (1992) argue that there might be a mixed relation and often contradictory discoveries between leverage and corporate performances. Yet another recent publications (Weill 2008) (González 2013) both report that this relationship varies across countries. A survey of the empirical literature on this debate shows the lack of consensus on the link between leverage and corporate performance. None of these references used quantile regression analysis to account for different impacts for different quantiles of the performance distribution. We want to know if Taiwanese firm s leverage is effective in boosting corporate performance for a sample of large Taiwanese companies and particularly, to determine if this relationship differs across different levels of corporate performance. We show that this ambiguity is largely due to the inappropriate least squares method employed in the literature. Our initial descriptive statistics show that the empirical distribution of the corporate performance measure (ROE) is slightly skewed to the right, meaning that it has a long right tail. This corporate performance measure (ROE) not only in the skewness but also in the kurtosis both indicate that it departure from normality. These findings suggest that the use of least squares to estimate the conditional performance functions are likely to yield coefficient estimates that are not representative of the overall performance distribution (Li et al. 2009). The contribution of this study is to employ the conditional quantile regression estimator method for the investigation of the impact of financial leverage on corporate performance. For the purpose of realizing the heterogeneous effect of financial leverage between firms at different parts of the distribution of performance variables, quantile regression is appropriate as a result of it allows us to check the covariate effects at different points of the performance variables. Instead of focusing on a single measure of the central tendency of the performance distribution, we evaluate the relative importance of explanatory variables at different points of the performance distribution (For example, for more profitable and for less profitable firms separately). We estimate conditional quantile regression models based on balanced panel dataset of 323 Taiwanese publicly traded companies over the period of , this paper examines the effect of leverage on corporate performance among firms with different levels of corporate performance. 2

3 Our main findings about the relation between financial leverage and the corporate performance of publicly traded companies are: (i) ROE as a measure of corporate performance: an insignificant relation among the more profitable firms, and a significantly negative relation among the less profitable firms. Further, the significantly negative relationship exists for firms belong to Chemical & Biotech industry and Textiles industry. (ii) Moreover, the significantly negative relationship is robust to alternative measure of financial leverage. (iii) Financial leverage has a lagged effect on corporate performance. (iv)however, regardless of either ROA or EPS as a measure of corporate performance: a significantly positive relation exists in the median quantile of ROA performance distribution or in the right-hand side quantiles of the EPS performance distribution. The structure of the paper is as follows. Section 2 presents the theoretical background of the relationship between leverage and corporate performance. In Section 3, we present the methodology and variables used in this study. Section 4 describes the data and the collected sample. Section 5 develops the empirical results. In Section 6, we show the robustness tests. Finally, we provide some concluding remarks in Section 7. 3

4 2. Theoretical background Whether debt financing boosts or hurts corporate performance is an empirical issue. Overall, there is no generally agreed relationship between leverage and corporate performance remaining an unresolved puzzle. The researches on the link between financial leverage and corporate performance could be categorized in three classifications. The first one to refer to mainly according to information asymmetries and signalling. Debt is a conceivable signal of the quality of firms and good quality firms are more tending to issue debt (Ross 1977). Therefore, this theory suggests that the highest performing firms, those having the more profitable investments, acquire more debt: Consequently, a positive relationship should exist between financial leverage and corporate performance. The second one to note that debt financing increases the burden on managers to act, as a result of it lower the moral hazard behavior by decreasing free cash flow at the disposal of managers (Jensen 1986). Accordingly, the firms with the higher leverage may better their performance (i.e., a positive relationship should occur between financial leverage and corporate performance). The third one to watch out for a higher leverage implies higher agency costs 2 owing to the disparate interests between shareholders and debt holders which enhance the total cost of the company, so that leverage may be negatively linked to performance (Jensen and Meckling 1976) (Myers 1977). Therefore, previous documents provide contrary contentions on the relationship between financial leverage and corporate performance. 2 (1)Agency costs result from the conflicts of interest between shareholders and managers. Consequently there should exist a positive influence of leverage on corporate performance. (2)Agency costs arise because of the conflicts of interest between shareholders and debtholders suggest that a higher leverage is correlated with a lower corporate performance. For a more detailed analysis of agency costs see Weill(2008). 4

5 3. Research Method Most of the empirical literature to date has focused on the conditional mean of the dependent variable, corporate performance. The general popular method, the traditional ordinary least squares (OLS) regression only enables researchers to approximate the conditional mean at the center of the distribution. Such regression can only give an incomplete description of a conditional distribution(mosteller and Tukey 1977). Koenker and Bassett (1978) proposed quantile regression (QR) that provides estimates of the linear relationship between the independent variables (regressors) and a specified quantile of the dependent variable. QR method estimates the effect of explanatory variables on the dependent variables at different points of the dependent variable s conditional distribution, and thus is capable of detecting the variation in the sensitivity of a firm s performance to the financial leverage across the major quantiles of its performance distribution. Furthermore, two additional features of qunatile regression fit our data better than traditional OLS estimator. First, the classical properties of efficiency and minimum variance of the OLS estimator are obtained under the restrictive assumption of independently, identically and normally distributed error terms. When the distribution of errors deviates from normality, the quantile regression estimator may be more efficient than the OLS(Buchinsky 1998). Second, because the quantile regression estimator is derived by minimizing a weighted sum of absolute deviations, the parameter estimates are less sensitive to outliers and long tails in the data distribution. This makes the quantile regression estimator relatively robust to heteroskedasticity of the residuals. For a more detailed analysis of quantile regression see Koenker and Hallock (2001). This paper employs the QR approach to examine the effect of leverage on corporate performance. The OLS results are also reported for comparison purposes. In terms of Buchinsky (1998), the quantile regression model is: Let ( y i, x i ), i = 1L,, n be a sample from some population where yi is the dependent variable, xiis a (K 1) vector of regressors, βθ is an unknown (K 1) vector of regression parameters to be estimated and is associated with theθ th percentile, and u θi is an unknown error term. Assuming that theθ th quantile of the conditional distribution of y i is linear in x i, we can write the conditional quantile regression model as follows: y i x β + u i θ θi = ' θ ( y ) ' i xi xiβθ Quant = Quant θ ( u ) = 0 θi x i (1) 5

6 Theθ th conditional quantile of y i given i x is denoted by ( ) ' Quant θ y i x i, which is equal to x iβθ. By varying the value of θ from 0 to 1, we trace out the entire distribution of y conditional on x. The estimator of regression parameter vector β θ can be obtained by minimizing a weighted sum of the absolute errors: min βθ ' ' ( 1 θ ) y β + θ β i xi yi xi ' i yi xiβ ' i: yi < xiβ : (2) The estimator of vector of regression parameters βˆ θ does not have an explicit form, but the resulting minimization problem can be solved by linear programming techniques(koenker and Bassett 1978). Since theory and previous empirical investigations recommend either a positive or a negative relationship between leverage and performance, we are specially interested in studying the relation across the distribution of the dependent variable. The quantile regression method allows us to recognize the effects of the covariates at different locations in the conditional distribution of the dependent variable. Therefore we use the quantile regression method (Koenker and Hallock 2001), with the following panel data model specification: Quant ' ' ( y x ) = α + α x γ z, θ it it 0 θ it + t (3) yit θ Where is the dependent variable at quantile. with i = 1L,,323 and t = 1L,, 5 In this paper, we employ the bootstrap method interpreted in Buchinsky (1995) to obtain estimates of the standard errors for the coefficients in quantile regression. This method is robust to relatively small samples and more importantly, it is valid under many forms of heterogeneity (Buchinsky 1995) To control for the relationship between leverage and corporate performance, we also add other variables, such as the the log of Total Sales(SIZE), the ratio of operating expenses for Research and Development to Net Sales(RD), the ratio between the Fixed Asset and Total Assets(AT), the ratio of Total Inventories to Total Assets(INVENTORY) and the Real Gross Profit Growth Rate(RGPGR). We also add annual dummies for the sake of accounting for changes in the macroeconomics environment with influence in the corporate performance within Taiwanese firms. Table 2 summarizes the definitions of the dependent and independent variables in this study. In the following tables we present the main description of variables. First, we present descriptive statistics (Table 3), and then we present correlations among variables (Table 4). It is worth keeping an eye on that correlations between the dependent and independent variables are not high. There are also no high correlations within explanatory variables, which, according to Aivazian et al (2005), does not indicate problems of multicollinearity. 6

7 4. Sample selection The academic community is arrested to Taiwan s stock market due to its particular characteristics. Using a sample of Taiwanese publicly traded companies in the Taiwan Economic Journal (TEJ) database from 2008 to 2012, this paper examines the effects of leverage on corporate performance among firms with different levels of corporate performance. Using this database, we study a panel data of firms from 2008 to To avoid selection issues, we studied a balanced panel of 323 firms that has no missing data over the five consecutive years. The final sample consists of 323 firms over the sample period, that is, a total of 1615 firm-year observations of annual financial data. Financial firms are excluded from the study because liabilities and capital structures of firms in the financial sector are fundamentally different from those of non-financial firms. Table 1 shows the classification of 19 industries in the sample. Following the usual practices in the literature on corporate finance, we delete financial insurance firms as they run specific businesses. The textiles electric machinery and chemical and biotech industries together account for about one-third of the sample, while the remaining industries each makes up less than 9%. Our initial descriptive statistics (Table 3) and graphics (Fig. 1 (a-b)) 3 show that the empirical distribution of the key corporate performance measure (ROE) is slightly skewed to the right, meaning that it has a long right tail. Such departure from normality is also highly apparent in the kurtosis. These findings suggest that the use of least squares to estimate the conditional performance functions are likely to give coefficient estimates that are not representative of the overall performance distribution. Fig. 2 depicts the total liabilities to total assets ratio (proxy for financial leverage; LEV1) over the period at various quantiles of the distribution. As can be seen from this figure, the mean total liabilities to total assets ratio decreased sharply prior to the 2010; however, the decrease in the mean has been driven mainly by the upper quantiles of the distribution where firms experienced very steep decreases in their total liabilities to total assets ratios, especially in As a matter of fact, Fig. 2 shows that very little decrease in the total liabilities to total assets ratio occurred at the lower quantiles of the distribution. For example, for firms in the 1th quantile, the total liabilities to total assets ratio in 2010 was slightly at the same level as that of By comparison, the total liabilities to total assets ratio for the 99th quantile decreased from 92.83% in 2008 to 85.89% in The median of the total liabilities to total assets ratio is consistently above the mean, indicating that the total liabilities to total assets ratio distribution is left skewed. The Return on equity (proxy for corporate performance; ROE, Fig. 3) shows a dissimilar pattern. Although its mean increased, especially during the period , this rise was mainly attributable to the steep increases in the ROE for the upper quantiles of the distribution. 3 Fig. 1 (c-f) exhibit histograms and normal probability plots of the ROA and EPS. Similarly, our alternative proxies to measure corporate performance, ROA and EPS, still departure from normality. 7

8 5. Empirical Results Here, we present our results. First we present the simple relationship between financial leverage and corporate performance across distribution. Then we present the complete specification and finally, we present figures that show the evolution of this relationship across the different quantiles of corporate performance distribution. In Table 5 and Table 6, we compare with a mean regression (ordinary least squares (OLS)) to find the hidden error in using this approach instead of ours. The first thing to note is that the mean coefficient is negative significantly in evaluating this relationship but a negative significantly relationship only occurs at the left hand side (from the 0.1 to the 0.5 quantiles) of the corporate performance distribution. Overall, the relationship between leverage and corporate performance in the multiple regression does not change when compared to the simple regression, which tends to confirm the analysis robustness. To take one step ahead, Fig. 4 and Fig. 5 both display the effect of leverage (LEV1) on corporate performance (ROE) and QR estimates with 95% confidence intervals versus OLS estimate. They seem to show the similar pattern. Leverage is likely to decrease corporate performance for lower performances (from the 0.1 to the 0.5 quantiles), which implies the merit of the agency argument in this application to Taiwan. Moreover, the quantile regression approach reveals interesting features concerning the relationship between firms financial characteristics and corporate performance with changes in sign and significant quantitative marginal effects across quantiles. The following five figures (Fig. 6) emphasize this point. Note that the effect of firm size (SIZE), R&D (RD) and Fixed Assets (AT) proportion do not increase corporate performance. Nevertheless, the real gross profit growth rate (RGPGR) seem to have a positive effect on corporate performance, which increases in the right hand side distribution (from the 0.5 to the 0.9 quantiles) of corporate performance in the case of growth rate. More unexpected was the effect of inventories (INVENTORY) that suggests that the share of inventories has not significant effect on corporate performance. For more discussions about leverage effect, Table 7 show the effect of the debt ratio (i.e., financial leverage) on corporate performance (i.e., ROE) across quantiles (from the 0.05 to the 0.95 quantiles with an increment of 0.05 per quantile). The corresponding estimates with 95% confidence level are graphed in Figure 5. First, the OLS estimate of the debt ratio is significantly negative at the 1% level. This result indicates that corporate performance will worse when debt ratio increases. However, as is shown in Table 7, the relationship between the debt ratio and ROE derived from the QR indicates that the debt ratio has a significantly negative effect for firms in the left-hand side and middle-range ROE quantiles (from the 0.05 to the 0.6 quantiles). Moreover, this relationship becomes insignificant for firms in the right-hand side ROE quantiles (from the 0.65 to the 0.95 quantiles). As presented in the two right-hand columns of Table 7, F-tests show the significant difference between slope estimates at theθ against ( 1 θ ) quantiles across various quantiles at least the 5% level, except the one middle regions (from the 0.45 to the 0.55 quantiles). The empirical results reported in Table 7 and graphed in Figure 5 show that the negative debt ratio-to-corporate performance relationship slowly increases as the ROE quantiles decrease. Based on the QR estimates of debt ratio from the 0.65 to the 0.95 quantiles, there is an insignificantly negative association between the debt ratio and a firm s ROE. However, this relationship becomes significantly negative from the 8

9 0.05 to the 0.60 quantiles. In brief, the negative effect of the financial leverage on corporate performance is especially apparent when firms confront low ROE conditions (for less profitable firms). Finally, we have evaluated the differences of leverage effect among the industries as defined by the Taiwan Economic Journal classification. As is shown in Table 14, we can see that the effect of leverage on corporate performance is significantly negative for most part of performance distribution when we refer to the Chemical & Biotech industry and Textiles industry. 9

10 6. Robustness tests In order to examine the stability of the results outlined above, a number of additional tests have been carried out. In this section, we first discuss the lagged effect of leverage on corporate performance and later re-examine the quantile-varying relationship between financial leverage and corporate performance using alternative measures in order to assess whether the results reported in Table 7 are driven by model misspecifications or potential correlations with omitted variables. A Lagged Effect of Financial Leverage As an additional test of robustness, we wondered if financial leverage could have a lagged effect in corporate performance. For the purpose of bettering corporate performance, firms can use debt financing as a tool, even so this tool can only be brought some influence in the future periods. We show figures for the conditional effects and confidence intervals for leverage lagged 1-3 periods (three panels, Fig. 7(a), Fig. 7(b) and Fig. 7(c)). Then, we show a comparison of conditional effects for different lags (fourth panel, Fig. 7(e)). Tables 8, Table 9 and Table 10 display leverage lagged effect for one, two and three period lag separately. The similar pattern for the effect of leverage in corporate performance is imposing. As a matter of fact, leverage seems to have a significantly negative impact for the left-hand side of the distribution of corporate performance. Model Specification With Alternative Proxy for Financial Leverage As is shown in Table 7, we assess the financial leverage variable by taking total liabilities divided by total assets (LEV1). In this section, we employ one alternative proxy to measure this variable and re-check its influence on corporate performance. Following Fattouc et al. (2005), we divide long-term liabilities by total assets and name this proxy LLTA. To make use of this proxy to measure financial leverage also gives comparable statistics that are accordant what has been finished in the literature to examine research questions similar to those recounted in this study. The effect of LLTA on corporate performance (ROE) across quantiles (from the 0.05 to the 0.95 quantiles with an increment of 0.05 per quantile) is offered in Table 11. As reported in Table 11, the LLTA has a significantly negative coefficient for lower ROE quantile (the 0.25 quantile) and higher ROE quantile (the 0.85 quantile), whereas this variable becomes insignificant for the extremely left-tail ROE quantiles (from the 0.05 to the 0.20 quantiles), the mid-range ROE quantiles (from the 0.30 to the 0.80 quantiles) and the extremely right-tail ROE quantile (from the 0.90 to the 0.95 quantiles). The results of F-tests reported in the two right-hand columns of Table 11 show that differences across ROE quantiles are not significant at the 10% level. In conclusion, the link between financial leverage and corporate performance across various quantiles reported in this study is robust because the relationship is not driven by the alternative measure of financial leverage in the regression model. 10

11 Model Specification With Alternative Proxy for Corporate Performance As is shown in Table 6, we assess the corporate performance variable by taking ordinary income divided by total equity (ROE). In this section, we employ two alternative proxies to measure this variable and re-check the effect of leverage on corporate performance. Following Mashayekhi and Bazaz (2008), we employ Return on asset (ROA) and Earnings per share (EPS) separately to measure corporate performance. The effect of leverage on ROA and EPS across quantiles are presented in Table 12 and Table 13. As shown in Table 12, the effect of leverage on ROA has a significantly positive coefficient for median quantile(the 0.5 quantile) of the ROA distribution. Similarly, as shown in Table 13, the effect of leverage on EPS has a significantly positive coefficient for right-hand side of the EPS distribution(from the 0.5 to the 0.90 quantiles). Fig. 8 and Fig. 9 both provide the impact of leverage on corporate performance across quantiles with QR estimates with 95% confidence intervals versus OLS estimate. Model Specification With Adding Control Variables So as to appraise if the estimate parameters of using QR model might be biased as a result of a missing-variables issue, we now afford results obtained by covering several variables in our quantile regression model. Following Murphy et al. (1996) and Li and Hwang (2011), we count four liquidity ratios: current ratio (current assets divided by current liabilities), accounts receivable turnover ratio (net sales divided by average accounts receivable), inventory turnover ratio (cost of goods sold divided by average inventory) and total asset turnover ratio (net sales divided by total assets). We cover these liquidity-ratio variables to treat as control variables in our model. We also accompany Nissim and Penman (2003) and Pae et al.(2005) by including the price-to-book (P/B) ratio in the model. Our quantile regression model has, in the aggregate, six explanatory variables and these five control variables. Using this model, we re-inspect the effect of financial leverage on corporate performance across quantiles. We use several figures to interpret the empirical results of this inspection. Figure 10 displays the quantile-varying estimates with 95% confidence levels for the model with eleven variables and year-dummies variables. Several contrasts have been made: Chart a of Figure 6 with Chart a of Figure 10, Figure 5 with Chart b of Figure 10, Chart b of Figure 6 with Chart c of Figure 10, Chart c of Figure 6 with Chart d of Figure 10, Chart d of Figure 6 with Chart e of Figure 10, Chart e of Figure 6 with Chart f of Figure 10. In these contrasts, we find that the shapes of quantile-varying estimates for the six explanatory variables are robust with regard to the inclusion of five control variables. While adding these five control variables in the quantile regression specification slightly changes the parameter estimates, that is to say, it does not influence the construction of the quantile-varying link between firm size, financial leverage, R&D, asset tangibility, inventory, real gross profit growth rate and corporate performance showed in Figure 6 (a), Figure 5 and Figure 6 (b-e) separately. Lastly, the quantile-varying effects of the five control variables on corporate performance are insignificant. Charts g, h, i, j and k of Figure 10 show that no regular form subsist for the quantile-varying estimates of the five control variables. 11

12 7. Conclusions In order to re-examine the relationship between financial leverage and corporate performance, this paper employs quantile regression method and a balanced panel dataset of 323 Taiwanese publicly traded companies over the period of The findings show that Ordinary Least Square(OLS) estimates are potentially misleading because they fail to capture how financial leverage shape corporate performance when focusing on the entire performance distribution. Like OLS estimates, the quantile approach shows that financial leverage in general has statistically significant negative impact on corporate performance. Yet the quantile approach adds to the story by revealing that such an effect is conditional on the position the firm occupies in the performance distribution. Using quantile regression, this research finds that the negative effect of financial leverage on corporate performance is especially evident when considering the extreme lower quantile. For middle and upper performance firms, an additional financial leverage does not affect significantly the performance of the firms. The findings could have policy implications that debt financing is most effective for firms with the lowest corporate performance (the less profitable firms). In addition to debt financing, firm size, firm s R&D expenditure and fixed assets of firms all do not better corporate performance. However, firm s executive and manager might enhance real gross profit growth rate to improve corporate performance when firms are successful(in right-hand side ROE distribution). Furthermore, the corporate executive management might not change the inventory shares of firms to affect corporate performance. This paper shows that the robust quantile regression method is quite useful for a better understanding of the impact of financial leverage on corporate performance in the context of Taiwanese publicly traded companies. The conditional quantile regression estimator extends the classical least squares estimation of the conditional mean to a collection of models running for different quantile functions. Accordingly, it permits the effect of a regressor to differ at different points of the conditional dependent-variable distribution, allowing us to portray and test the relations between financial leverage and corporate performance for better and worse performers separately. 12

13 References Aivazian, V. A., Y. Ge, and J. Qiu The impact of leverage on firm investment: Canadian evidence. Journal of Corporate Finance 11 (1 2): Balakrishnan, S., & Fox, I Asset specificity, firm heterogeneity and capital structure. Strategic Management Journal, 14(1): Buchinsky, M Estimating the asymptotic covariance matrix for quantile regression models a Monte Carlo study. Journal of Econometrics 68 (2): Recent advances in quantile regression models: a practical guideline for empirical research. Journal of Human Resources: Fattouh, B., P. Scaramozzino, and L. Harris Capital structure in South Korea: a quantile regression approach. Journal of Development Economics 76 (1): Ghosh, C., R. Nag, and C. Sirmans The pricing of seasoned equity offerings: evidence from REITs. Real Estate Economics 28 (3): Ghosh, D. K Optimum Capital Structure Redefined. Financial Review 27 (3): Gleason, K. C., L. K. Mathur, and I. Mathur The interrelationship between culture, capital structure, and performance: evidence from European retailers. Journal of Business Research 50 (2): González, V. M Leverage and corporate performance: International evidence. International Review of Economics & Finance 25 (0): Harris, M., and A. Raviv The Theory of Capital Structure. the Journal of Finance 46 (1): Jensen, M. C Agency Cost Of Free Cash Flow, Corporate Finance, and Takeovers. American Economic Review 76 (2): Jensen, M. C., and W. H. Meckling Theory of the firm: Managerial behavior, agency costs and ownership structure. Journal of financial economics 3 (4): Kinsman, M. D., and J. A. Newman Debt Level and Firm Performance: An Empirical Evaluation, at Puerto Vallarta, Mexico. Koenker, R., and G. Bassett Regression quantiles. Econometrica: journal of the Econometric Society: Koenker, R., and K. F. Hallock Quantile Regression. Journal of Economic Perspectives 15 (4): Li, M.-Y. L., and N.-C. R. Hwang Effects of Firm Size, Financial Leverage and R&D Expenditures on Firm Earnings: An Analysis Using Quantile Regression Approach. Abacus 47 (2): Li, T., L. Sun, and L. Zou State ownership and corporate performance: A quantile regression analysis of Chinese listed companies. China Economic Review 20 (4): Lubatkin, M., and S. Chatterjee Extending modern portfolio theory into the domain of corporate diversification: does it apply? Academy of Management Journal 37 (1): Majumdar, S. K., and P. Chhibber Capital structure and performance: Evidence from a transition economy on an aspect of corporate governance. Public Choice 98 (3-4):

14 Mashayekhi, B., and M. S. Bazaz Corporate Governance and Firm Performance in Iran. Journal of Contemporary Accounting & Economics 4 (2): Modigliani, F., and M. H. Miller The cost of capital, corporation finance and the theory of investment. The American economic review 48 (3): Mosteller, F., and J. W. Tukey Data analysis and regression. A second course in statistics. Addison-Wesley Series in Behavioral Science: Quantitative Methods, Reading, Mass.: Addison-Wesley, Murphy, G. B., J. W. Trailer, and R. C. Hill Measuring performance in entrepreneurship research. Journal of Business Research 36 (1): Myers, S. C Determinants of corporate borrowing. Journal of financial economics 5 (2): Nickell, S., and D. Nicolitsas How does financial pressure affect firms? European Economic Review 43 (8): Nickell, S., D. Nicolitsas, and N. Dryden What makes firms perform well? European Economic Review 41 (3 5): Nissim, D., and S. H. Penman Financial statement analysis of leverage and how it informs about profitability and price-to-book ratios. Review of Accounting Studies 8 (4): Pae, J., D. B. Thornton, and M. Welker The Link between Earnings Conservatism and the Price-to-Book Ratio*. Contemporary Accounting Research 22 (3): Pushner, G. M Equity ownership structure, leverage, and productivity: Empirical evidence from Japan. Pacific-Basin Finance Journal 3 (2): Ross, S. A The determination of financial structure: the incentive-signalling approach. The Bell Journal of Economics: Simerly, R. L., and M. Li Environmental dynamism, capital structure and performance: a theoretical integration and an empirical test. Strategic Management Journal 21 (1): Spence, A. M Capital structure and the corporation's product market environment. In Corporate capital structures in the United States: University of Chicago Press, Weill, L Leverage and corporate performance: does institutional environment matter? Small Business Economics 30 (3):

15 Fig. 1 (a-f) give the histograms and the normal probability plots of the three proxies for corporate performance variables. These charts confirm that the data depart from normality and are slightly skewed. Frequency a. Histogram of the ROE Normal F[(ROE-m)/s] Empirical P[i] = i/(n+1) b. Normal probability plot of the ROE Frequency c. Histogram of the ROA Normal F[(ROA-m)/s] Empirical P[i] = i/(n+1) d. Normal probability plot of the ROA Frequency e. Histogram of the EPS Normal F[(EPS-m)/s] Empirical P[i] = i/(n+1) f. Normal probability plot of the EPS Fig. 1 (a-f) Histograms and the Normal probability plots of the three proxies for corporate performance variables 15

16 Fig. 2 Evolution of total liabilities to total assets, % 5% 10% 25% 50% 75% 90% 95% 99% Mean Fig. 3 Evolution of ROE, % 5% 10% 25% 50% 75% 90% 95% 99% Mean 16

17 Table 1. Sample description Industry Number of firms Proportion of firms (%) Cement Foods Plastics Textiles Electric Machinery Electrical and Cable Chemical and Biotech Glass and Ceramics Paper and Pulp Iron and Steel Rubber Automobile Electronics Building and Construction Shipping and Transportation Tourism Trading and Consumer Oil Gas, and Electricity Others Total Note:This table presents the classification of the sample of 323 firms publicly traded companies in Taiwan from 2008 to

18 Table 2. Definitions of variables Variable name Dependent variable Corporate performance(roe) Independent variables Firm size(size) Financial leverage(lev1) R&D (RD) Asset Tangibility (AT) Inventory(INVENTORY) Real gross profit growth rate(rgpgr) Definitions Return on equity=ordinary income/total equity Natural log of total sales=ln(total sales) Total liabilities/total assets Operating expense-r&d/net sales Fixed asset/total assets Total inventories/total assets change of real gross profit/absolute of the previous year s real gross profit 18

19 Table 3. Descriptive statistics Variable Observations Mean Standard Min. Max. Skewness Kurtosis Deviation ROE(%) SIZE LEV1(%) RD(%) AT(%) INVENTORY(%) RGPGR(%) Table 4. Correlation matrix Variable ROE SIZE LEV1 RD AT INVENTORY RGPGR ROE 1 SIZE (0.7506) LEV *** * 1 (0.0000) (0.0837) RD ** *** 1 (0.0425) (0.1063) (0.0011) AT ** *** 1 (0.0156) (0.3609) (0.6238) (0.0002) INVENTORY *** *** (0.1668) (0.0000) (0.0004) (0.1466) (0.2776) RGPGR *** *** (0.0000) (0.9832) (0.6125) (0.7010) (0.0035) (0.3401) Notes:***,**, and * represent correlations that are statistically significant at 1%, 5% and 10% levels,respectively. The numbers in parentheses are P-values. 19

20 Table 5. Simple regression Independent variable OLS 10th Q 25th Q 50th Q 75th Q 90th Q Dependent variable:roe LEV *** *** *** *** 1.97e (0.0140) (0.0211) (0.0112) (0.0059) (0.0070) (0.0167) R 2 /Pseudo R N Note:A constant and a full-set of year-dummies are introduced but not shown. Bootstrapped SE (1000 reps) are shown in parentheses. *** Stands for significant coefficient at 1% level; **at 5%; *at 10% QR Estimate Line OLS Estimate Line Leverage(LEV1) on Corporate performance Fig. 4 Effect of Leverage(LEV1) on Corporate Performance (ROE): QR estimates with 95% confidence intervals versus OLS estimate under simple regression (Table 5). 20

21 Table 6. Multiple regression Independent variable OLS 10th Q 25th Q 50th Q 75th Q 90th Q Dependent variable:roe (Return on equity;roe= ordinary income/total equity) SIZE * (0.1346) (0.1353) (0.0722) (0.0534) (0.0594) (0.1213) LEV *** *** *** *** (0.0138) (0.0205) (0.0107) (0.0060) (0.0068) (0.0158) RD *** ** * ** ** (0.0450) (0.1217) (0.1001) (0.0254) (0.0298) (0.0639) AT *** *** *** (0.0130) (0.0132) (0.0077) (0.0054) (0.0069) (0.0100) INVENTORY (0.0140) (0.0115) (0.0068) (0.0057) (0.0061) (0.0164) RGPGR *** ** *** *** (0.0003) (0.0017) (0.0011) (0.0010) (0.0008) (0.0008) R 2 /Pseudo R N Note:A constant and a full-set of year-dummies are introduced but not shown. Bootstrapped SE (1000 reps) are shown in parentheses. *** Stands for significant coefficient at 1% level; **at 5%; *at 10% QR Estimate Line OLS Estimate Line Leverage(LEV1) on Corporate performance Fig. 5 Effect of Leverage(LEV1) on Corporate Performance (ROE): QR estimates with 95% confidence intervals versus OLS estimate under multiple regression (Table 6). 21

22 a. Firm size on corporate performance b. R&D on corporate performance c. Asset tangibility on corporate performance d. Inventory on corporate performance e. Real gross profit growth rate on corporate performance Fig. 6 (a-e) Marginal effects throughout the distribution and confidence intervals (Source: Table 6). 22

23 Table 7. Impact of financial leverage (LEV1) on corporate performance (ROE) across quantiles Estimated results of quantile regression Tests of the equality-of-slope estimates across quantiles Quantile Estimate (p-value) Quantile Estimate (p-value) Quantile F-statistics (p-value) (0.000)*** (0.610) 0.05 vs (0.0000)*** (0.000)*** (0.918) 0.10 vs (0.0000)*** (0.000)*** (0.829) 0.15 vs (0.0000)*** (0.005)*** (0.735) 0.20 vs (0.0000)*** (0.000)*** (0.754) 0.25 vs (0.0000)*** (0.000)*** (0.371) 0.30 vs (0.0000)*** (0.000)*** (0.290) 0.35 vs (0.0001)*** (0.000)*** (0.044)** 0.40 vs (0.0123)** (0.004)*** (0.005)*** 0.45 vs (0.4313) (0.000)*** OLS (0.000)*** Notes: The value in the parenthesis denotes the p-value. ***Stands for significant coefficient at 1% level; ** at 5%; * at 10%. When the QR model is estimated, it employs a multiple regression approach that includes the six explanatory variables (SIZE, LEV1, RD, AT, INVENTORY and RGPGR) and year-dummies simultaneously. We employ the bootstrap method illustrated in Buchinsky(1995) to obtain estimates of the standard errors for the coefficients in QR method. The F-tests of the equality-of-slope parameters across various quantiles and the differences between slope estimates at the θ and (1-θ) quantiles are presented in the two right-hand columns of this table. Source: Table 6 23

24 Table 8. Multiple regression Independent OLS 10th Q 25th Q 50th Q 75th Q 90th Q Variable Dependent variable:roe SIZE (0.1418) (0.1766) (0.0744) (0.0606) (0.0594) (0.1197) LEV1_ ** (0.0147) *** (0.0189) ** (0.0090) (0.0053) (0.0077) * (0.0096) RD * (0.0460) (0.2017) (0.0961) ** (0.0244) ** (0.0306) (0.0618) AT (0.0133) (0.0147) (0.0074) *** (0.0052) *** (0.0070) *** (0.0104) INVENTORY ** (0.0144) *** (0.0113) (0.0064) (0.0064) (0.0061) (0.0160) RGPGR *** (0.0003) (0.0018) (0.0012) ** (0.0010) *** (0.0008) *** (0.0008) R 2 /Pseudo R N Note:A constant and a full-set of year-dummies are introduced but not shown. Bootstrapped SE (1000 reps) are shown in parentheses. *** Stands for significant coefficient at 1% level; **at 5%; *at 10%. LEV1_1 represent the former lag 1 period variable of leverage (LEV1). 24

25 Table 9. Multiple regression Independent OLS 10th Q 25th Q 50th Q 75th Q 90th Q Variable Dependent variable:roe SIZE (0.1405) (0.1645) (0.0701) (0.0569) (0.0566) (0.1177) LEV1_ * (0.0146) *** (0.0197) (0.0093) (0.0050) (0.0065) (0.0108) RD * (0.0461) (0.2087) (0.1028) *** (0.0224) ** (0.0306) (0.0570) AT (0.0134) (0.0146) (0.0074) *** (0.0052) *** (0.0070) *** (0.0103) INVENTORY * (0.0144) *** (0.0113) (0.0063) (0.0062) (0.0064) (0.0163) RGPGR *** (0.0003) (0.0017) (0.0011) ** (0.0009) *** (0.0007) *** (0.0008) R 2 /Pseudo R N Note:A constant and a full-set of year-dummies are introduced but not shown. Bootstrapped SE (1000 reps) are shown in parentheses. *** Stands for significant coefficient at 1% level; **at 5%; *at 10%. LEV1_2 represent the former lag 2 period variable of leverage (LEV1). 25

26 Table 10. Multiple regression Independent OLS 10th Q 25th Q 50th Q 75th Q 90th Q Variable Dependent variable:roe SIZE (0.1362) (0.1654) (0.0722) (0.0582) (0.0545) (0.1156) LEV1_ ** (0.0141) *** (0.0180) (0.0095) (0.0048) (0.0061) (0.0108) RD ** (0.0448) (0.1901) (0.1027) *** (0.0223) ** (0.0288) (0.0567) AT (0.0131) (0.0149) (0.0071) *** (0.0052) *** (0.0068) *** (0.0100) INVENTORY * (0.0141) *** (0.0117) (0.0063) (0.0060) (0.0057) (0.0165) RGPGR *** (0.0003) (0.0017) (0.0012) ** (0.0010) *** (0.0008) *** (0.0009) R 2 /Pseudo R N Note:A constant and a full-set of year-dummies are introduced but not shown. Bootstrapped SE (1000 reps) are shown in parentheses. *** Stands for significant coefficient at 1% level; **at 5%; *at 10%. LEV1_3 represent the former lag 3 period variable of leverage (LEV1). 26

27 a. Lag 1 period of leverage (LEV1_1) on corporate performance b. Lag 2 period of leverage (LEV1_2) on corporate performance c. Lag 3 period of leverage (LEV1_3) on corporate performance d. leverage (LEV1) on corporate performance th 25 th 50 th 75 th 90 th LEV1 LEV1_1 LEV1_2 LEV1_ e. Different lags of leverage (LEV1) on corporate performance Fig. 7 (a-d) Marginal effects throughout the distribution of corporate performance (ROE) and confidence intervals for lagged leverage (multiple regression; Table 8, Table 9 and Table 10). Comparison between marginal effects of different lags of leverage (Fig. 7 (e) ). 27

28 Table 11. Robustness test: alternative proxy for financial leverage (LLTA) on corporate performance (ROE) across quantiles Alternative proxy for financial leverage: LLTA (Long-term liabilities/total assets) Estimated results of quantile regression Tests of the equality-of-slope estimates across quantiles Quantile Estimate (p-value) Quantile Estimate (p-value) Quantile F-statistics (p-value) (0.149) (0.160) 0.05 vs (0.4245) (0.342) (0.343) 0.10 vs (0.7507) (0.198) (0.023)** 0.15 vs (0.9069) (0.167) (0.133) 0.20 vs (0.5904) (0.044)** (0.160) 0.25 vs (0.2705) (0.053) (0.145) 0.30 vs (0.2900) (0.322) (0.153) 0.35 vs (0.9982) (0.407) (0.504) 0.40 vs (0.6691) (0.944) (0.827) 0.45 vs (0.8695) (0.834) OLS (0.576) Notes: The value in the parenthesis denotes the p-value. ***Stands for significant coefficient at 1% level; ** at 5%; * at 10%. When the QR model is estimated, it employs a multiple regression approach that includes the six explanatory variables (SIZE, LLTA, RD, AT, INVENTORY and RGPGR) and year-dummies simultaneously. We employ the bootstrap method illustrated in Buchinsky(1995) to obtain estimates of the standard errors for the coefficients in QR method. The F-tests of the equality-of-slope parameters across various quantiles and the differences between slope estimates at the θ and (1-θ) quantiles are presented in the two right-hand columns of this table. Source: Table 6 28

29 Table 12. Multiple regression Independent OLS 10th Q 25th Q 50th Q 75th Q 90th Q variable Dependent variable:roa (Return on asset;roa=net Income Exc Dispo/total assets) SIZE *** *** *** *** *** *** (0.0384) (0.0541) (0.0252) (0.0240) (0.0429) (0.0676) LEV ** (0.0039) (0.0057) (0.0031) (0.0026) (0.0040) (0.0080) RD * (0.0128) (0.0572) (0.0320) (0.0282) (0.0294) (0.0387) AT (0.0037) (0.0044) (0.0035) (0.0026) (0.0038) (0.0058) INVENTORY (0.0040) (0.0061) (0.0029) (0.0025) (0.0036) (0.0078) RGPGR (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) R 2 /Pseudo R N Note:A constant and a full-set of year-dummies are introduced but not shown. Bootstrapped SE (1000 reps) are shown in parentheses. *** Stands for significant coefficient at 1% level; **at 5%; *at 10% OLS Estimate Line QR Estimate Line Fig. 8 Effect of Leverage(LEV1) on Corporate Performance (ROA): QR estimates with 95% confidence intervals versus OLS estimate under multiple regression (Table 12). 29