Literature Review and Research Methodology

實證結果

2. Literature Review and Research Methodology

We use quantile regression (Koenker and Bassett 1978; Keonker and Hallock 2001) to examine the economic determinants of the interest rate that a bank charges for its loan (loan rate) and whether the relations between the loan rate and its economic determinants vary across different points on the loan rate distribution. The following is the regression model for our tests:

(1) LRATEt+1,q = b0q + b1qBMHOLDt + b2qBMPLEDGEt + b3qDUALt + b4qINDBt + b5qBDSIZEt + b6qCSCOREt + b7qBigNt + b8qTENUREt

+ b_9qMB_t + b_10qLEV_t + b_11qSIZE_t + b_12qLOSS_t

+ b13qROAt + b14qCAPINTt + b15q IntCovt + b16q PRatet + et

where q indicates a percentile in the conditional distribution of the loan rate. We first estimate equation (1) using the ordinary least square method (OLS) to obtain an average relation between the bank loan interest rate (LRATEt+1) and its economic determinants.

Then, we examine five representative quantiles of the loan rate distribution, 10%, 25%, 50%, 75% and 90%, using the quantile regression method (QReg). That is, we estimate the relation between the loan rate and its economic determinants at these points and examine whether the relation is homogeneous or heterogeneous across these points in the loan rate distribution.

Our dependent variable is the bank loan interest rate (LRATE_t+1). The Taiwan Economic Journal (TEJ) maintains a database for bank loans in Taiwan. For each firm in its database in a year, TEJ compiles loan amounts of all outstanding loans and loan interest rates. For each firm-year, we identify new loans originating in the year. Our bank loan interest rate (LRATE_t+1) is the weighted average interest rate (using loan amount as weight) of the new loans (firm subscript is omitted for ease of exposition for all

variables). Note that the bank loan interest rate is measured in the subsequent year (year t+1) after the economic determinants (explanatory variables in equation (1)) are measured (year t). This is to ensure that a firm’s financial statement information is available to bank loan officers when they assess the risk of the firm (Sengupta 1998; Jiang 2008).

corporate governance variables: (1) board member equity ownership, (2) board member equity pledge, (3) CEO-chairman dual position, (4) board independence, (5) board size, (6) conservatism score, (7) auditor quality, and (8) auditor tenure. We also include eight other variables: (1) market-to-book ratio, (2) financial leverage, (3) firm size, (4) a loss dummy, (5) return on assets, (6) capital intensity, (7) an interest coverage dummy, and (8) the prime interest rate. We explain each of these 16 economic determinants below.

We first discuss eight corporate governance variables. First, board member equity ownership (BMHOLDt) is measured as the percentage of a firm’s outstanding shares held by its board members. Regarding the effect of board member equity ownership on the cost of debt, Jensen (1993) argues that the board with greater ownership in the firm is more likely to monitor management diligently, which reduces agency conflicts between management and outside stakeholders such as debt holders. Empirically,

Ashbaugh-Skaife et al. (2006) find that credit ratings (which are negatively related to the cost of debt) are positively related to board ownership, consistent with Jensen (1993).⁶ Based on these studies, we expect a negative relation between the loan rate and board member equity ownership (BMHOLDt).

Second, board member equity pledge (BMPLEDGE_t) is measured as the percentage of board members’ shareholdings of their firm used as collateral for personal loans from outside financial institutions. In Taiwan, board members sometimes use their

shareholdings in the firm where they are board members as collateral to obtain personal loans from outside financial institutions, a practice we termed board member equity pledge (BMPLEDGEt).⁷ Board members often use the proceeds to purchase additional shares of the firm in order to increase their voting rights (control over the firm). Board member equity pledge can thus cause those board members’ voting rights to exceed their net ownership in the firm (net cash flow rights), and can create a strong incentive on the part of these board members to maintain high share prices. Fan and Wong (2002) show that the separation of voting rights from cash flow rights, which is commonly achieved through pyramid and cross-holding structures in East Asia, provides both means and incentives for controlling shareholders to benefit at the expense of outside or minority shareholders. They find that earnings informativeness decreases in the degree of

divergence between cash flow rights and voting rights in the seven East Asian economies they examine. Similarly, Kao and Chiou (2002) find that the informativeness of earnings decreases in board member equity pledge in Taiwan. Furthermore, Chiou et al. (2002)

find that the probability of financial distress is positively related to board member equity pledge in Taiwan. Finally, using U.S. data, Francis et al. (2005) examine the effect of dual class ownership structure on earnings and dividend informativeness. The dual class ownership grants one class of shares more voting rights than the other class of shares, resulting in a divergence between voting rights and cash flow rights. They find that earnings are generally less informative, and dividends are at least as (if not more) informative, for dual class firms. To sum up, the above studies all suggest that the separation of voting rights from cash flow rights creates agency conflicts between controlling shareholders and outside/minority shareholders, an effect which, we believe, is likely to increase the loan interest rate. We therefore expect a positive relation between the loan rate and board member pledge (BMPLEDGEt).

Third, CEO-chairman dual position (DUAL_t) is a dummy variable, which is set to one if the chief executive officer also serves as the chairman of the board and zero otherwise. Patton and Baker (1987) and Booth et al. (2002) find the effectiveness of board monitoring is reduced when a firm’s CEO also serves as the chairman of the board.

In addition, Dechow et al. (1996) and Carcello and Nagy (2004) both find that the probability of financial fraud increases in firms where CEOs are also the chairman of the board. In Taiwan, Chen and Yeh (2002) find a positive relation between earnings

management and CEO-chairman dual positions. Ashbaugh-Skaife et al. (2006) find that credit ratings are negatively related to CEO power.⁸ Based on these studies, we expect a positive relation between the loan rate and CEO-chairman dual position (DUALt).

Fourth, board independence (INDBt) is measured as the ratio between the number of independent board members and board size. Myers et al. (1997) find that independent board members curtail managerial perquisite consumption. Prevost et al. (2002) find a positive relation between firm performance and the percentage of independent directors on the board. Moreover, Chen et al. (2004) point out that management uses the

appointment of independent directors to signal to the market that it does not seek private benefits. Lee et al. (2003) find that corporate illegal acts are negatively related to the percentage of independent directors on the board in Taiwan. Finally, Ashbaugh-Skaife et al. (2006) find a positive relation between credit ratings and board independence. Based on these studies, we expect a negative relation between the loan rate and board

independence (INDB_t).

Fifth, board size (BDSIZEt) is measured by the number of board members on the board. There are two opposite views about the effect of board size on monitoring

effectiveness. Lipton and Lorsch (1992) and Jensen (1993) argue that large boards tend to suffer from social loafing and require higher coordination costs. Yermack (1996) find

dominated by management, both of which tend to enhance the effectiveness of board monitoring (e.g., Herman 1981; Zahra and Stanton 1989). Recently, Coles, Daniel and Naveen (2008) find that the relation between Tobin’s Q and board size is U-shaped with Tobin’s Q increasing (decreasing) in board size for complex (simple) firms. Due to the lack of consensus in the literature regarding the effect of board size on corporate

governance, we make no prediction for the relation between the loan rate and board size.

Sixth, we include a measure of accounting conservatism as a non-traditional governance variable. Zhang (2008) argues that conservative financial reporting on the part of borrowers benefits lenders ex post through the timely signaling of default risk (i.e., accelerated covenant violations) and thus can benefit borrowers ex ante through lower interest rates. We measure accounting conservatism (CSCOREt) following Khan and Watts (2007). First, we estimate the following equation with annual cross-sectional regressions:

(2) EARN_t = _1t + _2tNEG_t + RET_t(_1t + _2tMCAP_t + _3tMB_t + _4tMLEV_t) + NEGtRETt(1t + 2tMCAPt + 3tMBt + 4tMLEVt) + t

where:

EARNt= net income before extraordinary items, scaled by beginning-of-year market value of equity.

RET_t= annual returns over the 12 months from eight months before the fiscal year-end to four months after the fiscal year-end.

NEGt= one if RET  0, and zero otherwise.

MCAP_t= natural log of the market value of equity.

MB_t= market-to-book ratio.

MLEVt= long-term and short-term debt, scaled by beginning-of-year market value of equity.

Then, we collect the yearly -coefficients (i.e., 1t ~ 4t) and calculate the firm-year measure of conservatism, CSCOREt, using the equation (3) below:

(3) CSCOREt = 1t + 2tMCAPt + 3tMBi,t + 4tMLEVt

The higher the CSCOREt, the more conservative the firm is in its financial reporting in year t. We thus expect that the loan rate is negatively related to conservatism (CSCOREt).

Seventh, we include two more non-traditional governance variables: auditor quality and auditor tenure. Auditor quality (BigN_t) is a dummy variable set to one if a firm is audited by a Big 4 (or previously 5, 6, or 8) auditor.⁹ Auditor tenure (TENUREt) is measured by the number of years in the auditor-client relationship. Ghosh and Moon (2005) and Mansi, Maxwell and Miller (2004) provide evidence that audit quality or perceived audit quality is higher for firms audited by Big N auditors. Since high quality audit mitigates conflicts of interest between management and outside stakeholders, we expect that the loan rate is negatively related to both auditor quality (BigNt) and auditor tenure (TENURE_t).

We now turn to discussion of eight non-governance determinants of the bank loan interest rates. First, the market-to-book ratio (MBt) is often used in prior literature as a proxy for growth. Growth firms are potentially riskier. We thus expect a positive relation between the loan rate and the market-to-book ratio. Second, following prior literature (e.g., Ederington et al. 1987; Ziebart and Reiter 1992; Pittman and Fortin 2004;

Ashbaugh-Skaife et al. 2006), we include financial leverage (LEVt), measured as the ration between total liabilities and total assets, a loss dummy (LOSSt), which is set to one if income before extraordinary items is negative in the current and prior years and zero otherwise, and prime interest rate (PRatet), measured as the average interest rate on a one-month certificate of deposit from five major Taiwan banks.¹⁰ We expect these three variables to be positively related to the loan rate. Third, we include firm size (SIZE_t), measured as the natural logarithm of total assets, return on assets (ROA_t), and a dummy for interest coverage ratio (IntCovt), which is set to one if a firm’s interest coverage ratio (income before interest expense and taxes divided by interest expense) is above the median interest coverage ratio in a year and zero otherwise.¹¹ We expect these three variables to be negatively related to the loan rate. Finally, we include capital intensity (CAPINTt) to examine whether capital structure affects the loan rate. On one hand, capital intensive firms are likely to have greater volatility in earnings due to higher operating leverage (e.g., Baginski et al. 1999; Lev 1983). This view would suggest that the loan rate is positively related to capital intensity. On the other hand, capital intensive firms have more tangible assets. In the event of liquidation, more tangible assets are available to pay creditors. This would suggest a negative relation between the loan rate and capital intensity. We do not predict the sign for capital intensity.

missing values for variables used in this study. We then align the loan interest rate in the next year with economic determinants in the current year. This reduces our sample period to 1996-2005 since we lose year 2006 due to missing next year loan rate (LRATEt+1). To reduce the undue influence of extreme values, we winsorize all variables at 1% and 99%

of their respective distributions. Through the above selection process, we obtain a final sample of 2,599 firm-year observations spanning 1996 to 2005.

Table 1 reports descriptive statistics for variables in equation (1). The mean loan interest rate (LRatet+1) is 4.118% whereas the median 3.558%. The distribution of LRate_t+1, thus, is right skewed. The first and third quartiles for BMHOLD_t are 13.290%

and 30.470%, respectively. Although the mean BMPLEDGEt is 20.406%, the median is 9.440%. Moreover, about 23.4% of CEOs serve also as the chairman of the board (DUALt) and about 3.00% of board members are independent board (INDBt) with average board size equal to 9.908 members.¹²

Regarding other variables, the mean conservatism (CSCOREt) is 0.182; Big N auditors audit 78.4% of the sample; the mean auditor tenure (TENUREt) is 11.179 years.

The mean MBt (1.418) is higher than the median (1.150); the mean LEVt (0.423) is very close to its median (0.421). Loss firms account for 24.0% of the sample and the mean ROAt and CAPINTt are 0.036 and 0.469, respectively. Finally, the prime rate (PRatet) ranges from the minimum of 1.050% to the maximum of 5.210%.

Table 2 presents the Pearson (above the diagonal) and Spearman (below the diagonal) correlations among key variables. Since the Spearman correlations are

qualitatively similar to the Pearson correlations, we only discuss the Pearson correlations.

We find that the loan interest rate is significantly positively correlated with BMPLEDGEt

(0.274), SCORE_t (0.092), MB_t (0.162), LEV_t (0.079), LOSS_t (0.120), and PRate_t (0.757).

On the other hand, the loan interest rate is significantly negatively correlated with BMHOLDt (–0.100), INDBt (–0.227), BigNt (–0.033), TENUREt (–0.182), ROAt

(–0.147), and IntCovt (–0.120). These univariate correlations are all consistent with our expectations based on prior literature except that we find CSCORE_t is significantly positively correlated with the loan interest rate. We will discuss this more when presenting findings using quantile regression.

10 Table 1: Summary Statistics

Variable Mean Std. Dev. Min Q1 Median Q3 Max

LRATEt+1(%) 4.118 2.099 1.303 2.263 3.558 5.982 9.830 BMHOLDt(%) 23.139 13.007 5.420 13.290 20.400 30.470 82.930 BMPLEDGEt(%) 20.406 25.290 0.000 0.000 9.440 33.640 91.080

DUALt 0.234 0.423 0.000 0.000 0.000 0.000 1.000

INDBt 0.030 0.089 0.000 0.000 0.000 0.000 0.455

BDSIZEt 9.908 3.907 4.000 7.000 9.000 11.000 25.000

CSCOREt 0.182 0.450 –1.075 –0.029 0.141 0.351 2.027

BigNt 0.784 0.411 0.000 1.000 1.000 1.000 1.000

TENUREt 11.179 5.200 1.000 8.000 11.000 15.000 21.000

MBt 1.418 1.011 0.210 0.730 1.150 1.802 8.907

LEVt(%) 0.423 0.145 0.036 0.325 0.421 0.511 0.919

SIZEt 15.727 1.095 12.656 14.937 15.598 16.339 18.242

LOSSt 0.240 0.427 0.000 0.000 0.000 0.000 1.000

ROAt 0.036 0.083 –0.358 0.000 0.033 0.074 0.581

CAPINTt 0.469 0.289 0.008 0.236 0.431 0.675 1.207

IntCovt 0.434 0.496 0.000 0.000 0.000 1.000 1.000

PRatet(%) 2.768 1.621 1.050 1.480 2.130 4.420 5.210

The sample consists of 2,599 firm-year observations during 1996-2005 taken from the Taiwan Economic Journal database.

Variable Definition:

LRATEt+1= weighted average interest rate of new loans in the next year.

BMHOLDt= percentage of a firm’s outstanding shares held by board members.

BMPLEDGEt= percentage of board members’ stockholdings used as pledge for personal loans.

DUALt= one if the chairman of the board is also the chief executive officer, and zero otherwise.

INDBt= number of independent board members divided by board size.

BDSIZEt= number of board members on the board.

CSCOREt= a measure of firm-year specific conservatism estimated using equations (2) and (3) BigNt= one if the observation is audited by a big audit firm, and zero otherwise.

TENUREt= audit firm tenure measured by the number of years in the auditor-client relationship.

MBt= market-to-book ratio.

LEVt= financial leverage measured as the ratio between total liabilities and total assets.

SIZEt= natural logarithm of total assets.

LOSSt= one if the net income before extraordinary items is negative in the current and prior fiscal years, zero otherwise.

ROAt= return on assets.

CAPINTt= gross PPE divided by total assets.

IntCovt= one if a firm’s interest coverage ratio (income before interest expense and taxes divided by interest expense) is larger than the median interest coverage ratio in a year, and zero otherwise.

PRatet= prime interest rate measured as the average interest rate on a one-month certificate of deposit from five major Taiwan banks.

1. LRATEt+1 –0.100^‡ 0.274^‡ 0.003 –0.227^‡ –0.016 0.092^‡ –0.033* –0.182^‡ 0.162^‡ 0.079^‡ –0.013 0.120^‡ –0.147^‡ –0.001 –0.120^‡ 0.757^‡ 2. BMHOLDt –0.116^‡ –0.191^‡ –0.092^‡ 0.024 0.028 –0.012 0.092^‡ –0.068^‡ 0.118^‡ –0.019 –0.101^‡ –0.192^‡ 0.181^‡ 0.098^‡ 0.182^‡ 0.012 3. BMPLEDGEt 0.258^‡ –0.215^‡ –0.003 –0.173^‡ –0.127^‡ 0.121^‡ –0.017 0.099^‡ –0.103^‡ 0.192^‡ 0.149^‡ 0.164^‡ –0.210^‡ –0.007 –0.235^‡ 0.179^‡ 4. DUALt 0.014 –0.071^‡ –0.021 0.004 –0.158^‡ 0.074^‡ 0.055^‡ –0.086^‡ –0.014 0.050^† –0.139^‡ 0.128^‡ –0.095^‡ –0.003 –0.074^‡ –0.061^‡ 5. INDBt –0.244^‡ 0.019 –0.202^‡ 0.016 –0.021 –0.094^‡ 0.056^‡ –0.114^‡ 0.029 0.044^† –0.097^‡ –0.079^‡ 0.113^‡ –0.115^‡ 0.115^‡ –0.305^‡ 6. BDSIZEt –0.077^‡ 0.059^‡ –0.081^‡ –0.154^‡ 0.068^‡ –0.126^‡ 0.071^‡ –0.018 0.017 0.028 0.356^‡ –0.061^‡ 0.031 0.114^‡ 0.019 0.067^‡ 7. CSCOREt 0.113^‡ –0.013 0.040^† 0.076^‡ –0.092^‡ –0.154^‡ –0.114^‡ –0.063^‡ –0.398^‡ 0.344^‡ –0.319^‡ 0.317^‡ –0.416^‡ 0.176^‡ –0.336^‡ –0.001 8. BigNt –0.040^† 0.081^‡ –0.004 0.055^‡ 0.042^† 0.065^‡ –0.125^‡ –0.003 0.107^‡ –0.034* 0.123^‡ –0.040^† 0.060^‡ 0.000 0.039^† –0.050^† 9. TENUREt –0.176^‡ –0.071^‡ 0.156^‡ –0.088^‡ –0.124^‡ –0.039^† –0.059^‡ –0.006 –0.154^‡ –0.061^‡ 0.235^‡ –0.030 –0.006 0.076^‡ –0.065^‡ –0.176^‡ 10. MBt 0.093^‡ 0.180^‡ –0.097^‡ –0.032 0.085^‡ 0.066^‡ –0.580^‡ 0.114^‡ –0.196^‡ –0.146^‡ 0.051^‡ –0.275^‡ 0.519^‡ –0.145^‡ 0.412^‡ 0.250^‡ 11. LEVt 0.110^‡ –0.047^† 0.142^‡ 0.047^‡ 0.042^† 0.049^† 0.298^‡ –0.030 –0.051^‡ –0.155^‡ 0.149^‡ 0.273^‡ –0.376^‡ –0.058^‡ –0.393^‡ –0.133^‡ 12. SIZEt 0.011 –0.183^‡ 0.265^‡ –0.135^‡ –0.099^‡ 0.282^‡ –0.402^‡ 0.102^‡ 0.275^‡ 0.023 0.156^‡ –0.049^† 0.093^‡ 0.036* –0.080^‡ 0.051^‡ 13. LOSSt 0.137^‡ –0.208^‡ 0.137^‡ 0.128^‡ –0.069^‡ –0.071^‡ 0.318^‡ –0.040^† –0.018 –0.360^‡ 0.254^‡ –0.035* –0.662^‡ 0.093^‡ –0.492^‡ –0.040^† 14. ROAt –0.175^‡ 0.233^‡ –0.203^‡ –0.097^‡ 0.139^‡ 0.054^‡ –0.472^‡ 0.078^‡ –0.059^‡ 0.599^‡ –0.355^‡ 0.031 –0.740^‡ –0.160^‡ 0.644^‡ 0.024 15. CAP_INTt 0.025 0.050^† 0.034* –0.019 –0.101^‡ 0.131^‡ 0.155^‡ –0.004 0.086^‡ –0.140^‡ –0.086^‡ 0.016 0.076^‡ –0.146^‡ –0.117^‡ 0.051^‡

16. IntCovt –0.144^‡ 0.219^‡ –0.224^‡ –0.074^‡ 0.117^‡ 0.055^‡ –0.366^‡ 0.039^† –0.080^‡ 0.469^‡ –0.399^‡ –0.089^‡ –0.492^‡ 0.766^‡ –0.111^‡ 0.032 17. PRatet 0.709^‡ 0.006 0.185^‡ –0.061^‡ –0.320^‡ 0.002 –0.007 –0.051^‡ –0.167^‡ 0.233^‡ –0.125^‡ 0.066^‡ –0.022 0.010 0.073^‡ 0.027

The sample consists of 2,599 firm-year observations during 1996-2005 taken from the Taiwan Economic Journal database. See Table 1 for variable definition.

Pearson (Spearman) correlations are reported above (below) the diagonal. *, †, ‡ indicate significance at 10%, 5%, or 1% respectively.

12 4. Empirical Findings

Table 3 reports our findings from estimating equation (1) using the ordinary least square (OLS) method and the quantile regression (QReg) method at the 10%, 25%, 50%, 75% and 90% quantiles. We report the OLS results to provide a baseline for comparison with the QReg results. We first discuss the OLS results in the “OLS Regression” column. Overall, all explanatory variables are significantly related to the loan rate in expected direction except for DUALt, INDBt, CSCOREt, and BigNt. The explanatory power of our model is high with adjusted R-squared equal to 0.6430. We discuss our results in more detail below.

Among our eight governance variables, we find a significantly negative coefficient on BMHOLDt (–0.013, p-value < 0.0001), BDSIZEt (–0.013, p-value <

0.10), and TENUREt (–0.010, p-value < 0.05), consistent our expectations based on prior literature. These findings suggest that firms with greater board equity ownership, larger boards of directors, or longer auditor tenure are associated with a lower loan interest rate. On the other hand, the coefficient on BMPLEDGEt is significantly positive (0.008, p-value < 0.0001) as expected, suggesting that firms with boards whose members collateralize their stockholdings have a higher cost of debt. This finding is consistent with collateralized shares creating a divergence between voting rights and net cash flow rights and thus agency conflicts between controlling

shareholders and minor/outside shareholders (Chiou et al. 2002; Kao and Chiou 2002).

We, however, do not find that CEO-chairman dual position (DUALt), board

independence (INDBt), and conservatism (CSCOREt) are significantly related to the loan rate in our sample, failing to support our expectations based on prior literature.

Surprisingly, we find that BigNt is significantly positively related to the loan rate, contrary to our expectation.¹³

The coefficients on eight non-governance variables are all significant and in expected directions. For example, financial leverage is strongly positively related to the loan rate (1.971, p-value < 0.0001). In addition, as expected, firms with a larger size (SIZEt) or higher profitability (ROAt) are associated with a lower loan interest rate.

13 See Table 1 for variable definitions.

Numbers in brackets are two-tailed p-values of the t-statistics.

Figure 1. Ordinary Least Squares and Quantile Regression Estimates

Panel A: BMHOLD Panel B: BMPLEDGE

Panel C: DUAL Panel D: INDB

Panel E: BDSIZE Panel F: CSCORE

-0.020-0.015-0.010-0.0050.000BMHOLD

0 .2 .4 .6 .8 1

Quantile

Board Member Equity Ownership (BMHOLD)

0.0020.0040.0060.0080.0100.012BMPLEDGE

0 .2 .4 .6 .8 1

Quantile

Board Member Pledge (BMPLEDGE)

-0.300-0.200-0.1000.0000.1000.200DUAL

0 .2 .4 .6 .8 1

Quantile

CEO-Chairman Dual Position (DUAL)

-2.000-1.0000.0001.0002.000INDB

0 .2 .4 .6 .8 1

Quantile

Board Independence (INDB)

-0.040-0.0200.0000.020BDSIZE

0 .2 .4 .6 .8 1

Quantile Board Size (BDSIZE)

-0.400-0.2000.0000.2000.400CSCORE

0 .2 .4 .6 .8 1

Quantile

Conservatism (CSCORE)

Panel I: MB Panel J: LEV

Panel K: SIZE Panel L: LOSS

-0.2000.0000.2000.4000.600BigN

0 .2 .4 .6 .8 1

Quantile Large Audit Firm (BigN)

-0.060-0.040-0.0200.0000.020TENURE

0 .2 .4 .6 .8 1

Quantile Auditor Tenure (TENURE)

-0.200-0.1000.0000.1000.2000.300MB

0 .2 .4 .6 .8 1

Quantile Market-to-Book Ratio (MB)

0.5001.0001.5002.0002.5003.000LEV

0 .2 .4 .6 .8 1

Quantile Leverage (LEV)

-0.300-0.200-0.1000.000SIZE

0 .2 .4 .6 .8 1

Quantile Firm Size (SIZE)

-0.2000.0000.2000.4000.600LOSS

0 .2 .4 .6 .8 1

Quantile LOSS

Figure 1. Ordinary Least Squares and Quantile Regression Estimates—continued Panel M: ROA Panel N: CAPINT

Panel O: IntCov Panel P: PRate

Notes:

The horizontal axis in each panel depicts cost of debt quantiles whereas the vertical axis in a panel reports the coefficients on an explanatory variable.

In each panel, the bold dotted line represents the OLS coefficient estimate for an explanatory variable and fine dotted lines depict the 90% confidence interval for the OLS estimate.

In each panel, the solid curve line reports the QReg coefficient estimates for an explanatory variable corresponding to various cost of debt quantiles and the shaded band represents the 90% confidence interval for the QReg estimates.

-3.000-2.000-1.0000.0001.000ROA

0 .2 .4 .6 .8 1

Quantile ROA

-1.000-0.5000.0000.500CAPINT

0 .2 .4 .6 .8 1

Quantile Capital Intensity (CAPINT)

-0.400-0.2000.0000.2000.4000.600IntCov

0 .2 .4 .6 .8 1

Quantile Interest Coverage (IntCov)

0.4000.6000.8001.0001.200PRate

0 .2 .4 .6 .8 1

Quantile Prime Interest Rate (PRate)

Panel B: Differences in the coefficients on BMPLEDGE across LRATE quantiles