Análisis de 5 por qué?

After examining the importance of debt capacity on cash holdings and assessing their relationship for constrained and unconstrained firms, a reverse causality test is performed in this Chapter using a series of debt capacity models (1) to confirm the substitution effect (negative relationship) between cash and debt capacity reported in Campello et al. (2011), (2) to confirm the positive effect of cash flow on debt capacity reported in Sufi (2009), and finally (3) to assess the importance of cash flow sensitivity of cash in explaining debt capacity. To support the reverse causality relationship between cash and debt capacity, a Granger causality test is performed prior to the actual debt capacity models using one lag of change in cash ratio and two lags of change in debt capacity ratio.34 The bivariate Granger test shows that there is a significant feedback relationship between cash and debt capacity where change in cash Granger cause change in debt capacity, and vice versa.35

Further continuing the internal flexibility analysis, interactions between cash, cash flow and debt capacity are examined using adapted models from Sufi (2009) and Campello et al. (2011) as follow.

34

Taking the change in debt capacity and cash ratio from one period to the previous period eliminates the time series effect and converts the ratios into stationary variables. One lag of change in cash ratio and two lags of change in debt capacity ratio are used in the Granger causality test because the respective lags are found to yield the lowest Akaike Information Criterion (AIC) for the regressions individually.

35

Regressions performed in the Granger causality test reports large t-values and p-value of 0.000 for all respective regressions. At a p-value of 0.000, null hypothesis of no relationship between the change in cash and the change in debt capacity is rejected.

124

DC/(DC + Cash)i,t = DC/IFi,t= c + 1CFei,t+ 4Ln(AT)i,t + 2Crediti,t+ 3Qi,t+ i (i)

DC/Assets i,t= c + 1CFei,t+ 2Cashi,t+ 3Ln(AT)i,t + 4Crediti,t+ 5Qi,t+ i (j)

DC/Assetsi,t= c + 1CFei,t+ 2Cashi,t+ 3(Cash flow*Cash)i,t+ 4Ln(AT)i,t

+ 5Crediti,t+ 6Qi,t+ i (k)

where internal flexibility (IF) is computed as the sum of debt capacity and cash; independent variables are firm characteristics that control for firm size, credit rating, and investment growth prospect following Campello et al. (2011).36 The Campello‟s modified model is next adapted, where the effect of cash holdings on credit is explicitly modelled, allowing for nonlinearities by using a different scaling factor in the dependent variable and including cash and its interaction as independent variables. The interaction term helps assess the effect of cash flow sensitivity of cash on debt capacity. Inclusion of the interaction term makes cash and cash flow conditional upon each other which, in practice, is true to a certain extent in the cash model regressions. Interpretation of estimated coefficients for cash and cash flow become different; and model (k) will be interpreted in a different manner compared to model (i) and (j). Model (k) investigates the use of cash and credit lines when cash flow are assumed to give firms greater access to credit facilities, i.e. positive relationship between credit lines and cash flow documented in Sufi (2009). The final debt capacity model includes industry cash flow volatility and additional control variables from Bates et al. (2009) and Opler (1999). Cash flow is found to add positively to firms‟ borrowing capacity (Campello et al., 2011); as such it is expected that cash flow risk has a negative impact on firms‟ debt

36 _{Total internal flexibility (liquidity) is computed as the sum of lines of credit (LC) and cash in Sufi (2009). In this}

paper, debt capacity is used to replace lines of credit in the computation of total internal flexibility. Firm characteristics are defined as such –Large is a dummy variable for firm size taking value of one if firm‟s sale revenue is equal or more than $1billion; credit is a dummy variable equals to one if firm has rating BBB- or higher; and investment growth prospect is measured using Q ratio.

125

capacity, where firms with volatile cash flow have lower debt capacity, or else being equal. Negative correlation between volatility and debt capacity is found in the univariate tests. The actual impact of volatility on debt capacity is derived from the following regression model.

DC/Assetsi,t= c + 1CFei,t+ 2Cashi,t+ 3(Cash flow*Cash)i,t+ 4INDSTDCFi,t

+ 5Ln(AT)i,t + 6Crediti,t+ 7Qi,t+ i (l)

The Granger causality test performed prior the debt capacity models indicated a feedback relationship between the two variables. The relationship between cash

holdings and debt capacity is subjected to a “reverse-causality story” (Campello et al., 2011). Greater cash holdings may be due to firms utilizing their credit lines in the same period, resulting in higher cash holdings and lower levels of credit lines. The authors test this reverse-causality using the level of drawdowns made by firms because when firms have higher cash holdings and use less of their credit lines, the consequent lower level of drawdown will not lead to greater cash holdings.

Here, this is tested using firms‟ unused debt capacity and actual debt usage. First,

the level of drawdown is low when firms‟ unused debt capacity is high (i.e. firms preserving their existing capacity), during which low drawdown should not add to cash holdings. If cash holdings are high when unused debt capacity is high, the substitution effect between cash holdings and debt capacity will be supported while eliminating the

possibility that higher cash holdings are due to higher debt use from firms‟ higher debt

capacity. However, using the unused debt capacity ratio denominated by total assets to determine its relationship with cash may be subjected to bias since firms may naturally have higher or lower unused debt capacity depending on its size. Ratio is scaled by debt

126

capacity for models (i) to (l) because it is the unused portion out of total capacity that is the most important and relevant. Second, the reverse-cycle is further tested using actual debt usage out of the total debt capacity. Actual debt usage is measured using (1) long- term debt scaled by debt capacity to measure the proportion of debt used out of total borrowing capacity, (2) net debt issuance scaled by debt capacity to measure actual debt use in each period more accurately predict the actual addition to cash from new debt issuance, (3) long-term debt scaled by total assets, and (4) net debt issuance scaled by total assets, where the latter two are generic measurement that serves as a robustness test for the first two measures scaled by debt capacity, As greater debt usage (due to higher debt capacity) cannot lead to firms having less cash, a negative relationship between debt use and cash holdings implies that firms have greater cash-on-hand not due to greater debt usage from larger borrowing capacity. Rather, cash holdings are carefully

managed to maximise firm‟s internal flexibility according to existing borrowing capacity level. Together, the results from unused debt capacity and actual debt usage will support the trade-off between cash and debt capacity, and the active management of total internal financial flexibility by firms.

127

A7.2 Debt capacity model regression results

Table A27 reports results of the adapted debt capacity model. The previous model (Sufi, 2009) uses lines of credit obtained from the Dealscan database that includes mainly large firms with publicly published details on bank debt. Here, debt capacity is a derived estimate using coefficient estimates from Berger et al. (1996). Although the two measures are not directly comparable, they represent a portion of a

firm‟s borrowing capacity and are expected to have similar trends in the regression

models. Accordingly, all variables are found to enter the debt capacity model with the same sign as Campello et al. (2011).37

First, cash flow positively impacts debt capacity over the entire sample, where increase in cash flow improves the firm‟s borrowing capacity. While higher profits reduce the probability of firms violating the covenants and increase the use of credit lines (Sufi, 2009), increased profitability also contributes positively to firms‟ generic measure of debt capacity. The magnitude of the coefficient estimate of cash flow is, however, much smaller than with the previous model, indicating that cash flow may be a less important variable in explaining total debt capacity compared to committed lines of credit. Supporting this further, cash flows have much lower estimate after the firm fixed effect is taken into account (in Model 6) and after cash holdings is included as additional determinant in Models 3, 7 and 11. The latter implies that cash ratio has variance that partly subsumes the effect of cash flow on debt capacity. Second, cash is found to have a highly significant and negative relationship with debt capacity, supporting the trade-off

37

Campello et al. (2011) perform the credit line regression for financial year 2008 sample clustered by industry. To better compare the results with Campello et al. (2011), Appendix A-XIV reports the debt capacity model (k) performed for year 2008 and the entire sample using industry fixed effects.

128

practice between the two variables. As mentioned, the most important variable (cash flow) for credit line no longer reports high coefficient after cash is taken into account. More importantly, the fit of the model increases more than four times. This suggests that cash is a more important determinant of debt capacity compared to cash flow.

Third, the interaction effect between cash and cash flow is negative and significant in Models 4, 8 and 12. Compared to the previous model, the estimate for the interaction term is small because cash flow enters the debt capacity model with a very small coefficient. From the Fama-MacBeth yearly regression (Model 12), the largest coefficient of interaction term (0.095) is further interpreted as follows. If a firm does not keep cash on hand, the impact of any change in cash flow on debt capacity is 7.6%, e.g. an arbitrary ten percent change in cash flow will increase debt capacity only by less than a percentage point (76 basis points). In practice, change in cash flow to total assets is, on average, 30 basis points from 1980 to 2008, implying that the actual impact of cash flow on debt capacity is minimal and, on average, 2 basis points of total assets. Taking the last two financial years as an example, for a firm holding the average level of cash at 21% of total assets and cash flow to total assets decreasing at average rate 1.1% from 2007 to 2008, the year-to-year decrease in cash flow contributes 0.06 basis point to the total increase in debt capacity to total assets of 80 basis points. This effect is equivalent to nil impact of cash flow on debt capacity from 2007 to 2008. As the year-to-year change in cash flow is comparably smaller than the change in debt capacity, results show that cash flow may not have significant impact on debt capacity. On the other hand, if a firm has median cash flow (6.4% of total assets) in 2008 and a decrease in average cash holdings of 2% from 2007 to 2008, the fall in cash holdings contribute to

129

almost all of the consequent increase in debt capacity of 80 basis points. Finally, except for Model 1 and firm size, most of the control variables enter the regression model with the same sign as in the previous study. The positive relationship between firm size and debt capacity is reported in Models 1, 2 and 10, while the rest of the models report a negative relationship. Theoretically, a positive relationship is expected as larger firms have greater capacity to borrow. However, it is noted that Campello et al. (2011) report a similar negative relationship for firm size in their credit lines regression using a Compustat sample.

BKS identified cash flow volatility as a major determinant of cash holdings, and previous Chapters found significant relationship between volatility and debt capacity; henceforth, the variable is included in the modified debt capacity model to test its impact on debt capacity. Results are reported in Models 5, 9 and 13 in Table A27. Fit of the model improves only slightly when cash flow volatility is included. Industry cash flow volatility is highly significant even with cash holdings present in the model. There is a strong negative relationship between volatility and debt capacity, where firms with increased cash flow risk have lower debt capacity. However, addition of the volatility variable does not change the relationship of existing variables on debt capacity.

To sum up the findings in Table A27, cash flows are found to have little impact on debt capacity when cash and cash flow risk are considered, since the bulk of the variance is explained by the latter two variables. The interaction term of cash flow and cash holdings shows that even when consideration is made to the extent cash flow adds to cash holdings in each period, cash flow exhibits minimal impact on debt capacity. Instead, cash is a significantly more important variable of debt capacity, followed by

130

cash flow risk. Results obtained are not entirely in line with Campello et al. (2011) and Sufi (2009), where cash flow is a significant explanatory variable of credit lines even when cash holdings are accounted for. The difference in results is attributed to (1) difference in dependent variable used – the previous study used direct measure of credit lines, while a generic measure of debt capacity is adapted here; and (2) difference in firm sample.

First, debt capacity and credit lines encompass different components that should be interpreted differently. The estimated debt capacity is a more generic measure of the total borrowing capacity of firms while credit lines should be a subset of debt capacity. Debt capacity ratio is a firm-level measure of expected asset liquidation value from Berger et al. (1996), and used in Almeida and Campello (2007) and Hahn and Lee (2009). Subject to lenders accurately predicting asset liquidation value of firms and extending loans accordingly, the generic measure provides a fair measure of the general level of borrowing capacity in firms. From the previous chapters, debt capacity has important relationship with cash holdings for the entire sample and across firms with varying levels of financial constraints. Debt capacity is an important component in total internal flexibility and a form of substitute for cash. Second, the sample of firms used in the previous study may consist mainly of larger firms with reported information on bank credit commitments. This is supported by results in the fourth and fifth columns in Table A28 Panel A. For the debt capacity model performed on a sample of large firms, cash flow is found to exhibit greater explanatory power, 0.255 compared to 0.069 for the sample of small firms using the Fama MacBeth regression model. The estimates obtained for large firm subsample are comparable to previous studies. This shows that

131

the difference in results may be due to difference in samples of firms used and fundamental difference in dependent variable used.

Moving on, to analyse whether the relationships between cash and debt capacity are consistent across all firms, the modified debt capacity model is utilised over different subsamples of firms sorted according to several financial constraints criteria. Table A28 shows that the impact of cash holdings on debt capacity is similar for constrained and unconstrained firms. Regardless of the level of constraints, cash holdings remain an important driver of debt capacity. The other variables report a similar trend as there is no significant difference in their relationship with debt capacity over different firm subsamples. This means that variables influencing debt capacity do not vary for firms with different levels of financial constraints, which either implies that an important variable is omitted from the model or that debt capacity is a more stable variable that may not change as much as cash holdings (reported in Chapter A5) when financial constraints and other firm variables vary.38

The latter supports our hypothesis that debt capacity is the first-level variable that exhibits less periodic change compared to the second-level variable cash which can be monitored and adjusted in each period. The debt capacity models suggest lines of credit to be dependent on the level of cash flow and amount of cash savings firms have but our hypothesis is slightly different in terms of cause-and-effect. We presume firms‟ cash policies to be first dependent on the level of debt capacities firms have and, depending on the changes in credit capacities, firms adjust their cash savings accordingly in each period. This is because cash policies are generally more flexible and

38

The problem of omission of important variables is reduced as regressions are performed with constant terms and firm and year fixed effects.

132

subject to less restriction, while debt capacity generally depends on many factors such as the nature of assets, nature of business and bank relationships. As managers in practice may not have total control over debt capacity, they adjust cash-on-hand periodically to cater for their financial flexibility needs, while both debt capacity and cash add to total financial flexibility, cash holdings can be managed more easily and is in practice varied more often compared to debt capacity.

133 Table A27 Regressions of debt capacity models (i) to (l) for period 1980 to 2008

This table reports the debt capacity models (i) to (l) results using the CRSP/Compustat sample from 1980 to 2008. All firms are included with the exception of financial (SIC=69000-6999), utilities (SIC 4900-4999), and government (SIC=9000-9999) firms. Firms with missing or negative total assets and total sales are excluded. The interaction term between cash flow and cash holdings is computed as the product of cash flow to total assets and cash holdings to total assets for each firm-year observation. Variable definitions are reported in Chapter A2.3. The dependent variable is debt capacity ratio computed as debt capacity to the sum of debt capacity and cash (internal flexibility) for Mod el 1 and debt capacity to total asset for the rest of the models. Large is a dummy variable taking a value of 1 if firm has sales revenue equal or more than US$1billion, and zero otherwise. Credit is a dummy variable with value 1 if firm has a S&P long-term credit rating of BBB- and higher, and zero otherwise. OLS model reports estimates from OLS regressions with heteroskedasticity adjusted standard errors. FE model controls for firm and year fixed effects. FM model is based on Fama-MacBeth regressions using Newey and West (1987) standard errors controlling for autocorrelation. All regressions include a constant term (not reported) . The table reports coefficient estimates where *** represents significance at 1% level, and ** and * represents significance at 5% and 10% respectively. T-values are reported in the parentheses. MODEL1 MODEL2 MODEL3 MODEL4 MODEL5 MODEL6 MODEL7 MODEL8 MODEL9 MODEL10 MODEL11 MODEL12 MODEL13

Model OLS FE FM

Dependent DC_IF DC_AT DC_AT DC_AT DC_AT DC_AT DC_AT DC_AT DC_AT DC_AT DC_AT DC_AT DC_AT CFe 0.186*** 0.113*** 0.043*** 0.067*** 0.062*** 0.001 0.013*** 0.024*** 0.023*** 0.105*** 0.053*** 0.076*** 0.074*** . (54.68) (62.24) (29.71) (30.66) (28.85) (0.45) (9.86) (13.61) (12.79) (15.52) (5.95) (9.11) (8.53) CASH -0.458*** -0.467*** -0.443*** -0.375*** -0.380*** -0.382*** -0.438*** -0.445*** -0.435*** . (-265.04) (-252.82) (-236.70) (-209.32) (-204.07) (-205.73) (-34.25) (-38.36) (-29.62) CASH_CFe -0.091*** -0.107*** -0.050*** -0.049*** -0.095*** -0.095*** . (-14.56) (-17.31) (-9.39) (-9.32) (-7.03) (-7.39) LARGE 0.072*** 0.001 -0.027*** -0.028*** -0.030*** -0.034*** -0.040*** -0.040*** -0.037*** 0.019*** -0.011*** -0.011*** -0.013*** . (27.13) (0.36) (-24.27) (-24.90) (-26.48) (-25.11) (-34.40) (-34.47) (-32.00) (5.82) (-6.64) (-7.24) (-7.61) CREDIT 0.060*** -0.006*** -0.026*** -0.027*** -0.024*** -0.032*** -0.038*** -0.038*** -0.034*** -0.003 -0.018*** -0.018*** -0.018*** . (16.71) (-3.10) (-17.15) (-17.69) (-15.87) (-21.36) (-29.68) (-29.79) (-27.18) (-1.20) (-5.05) (-5.03) (-5.00) Q -0.039*** -0.016*** -0.001*** -0.001*** -0.001*** -0.001*** 0.004*** 0.004*** 0.004*** -0.015*** -0.001 -0.001 -0.000 . (-91.60) (-71.95) (-7.65) (-7.10) (-3.70) (-4.27) (26.29) (26.54) (24.71) (-21.31) (-1.13) (-1.08) (-0.18) INDSTDCF -0.327*** -0.178*** -0.158*** . (-57.84) (-31.30) (-4.89) Adj R-Sq 0.142 0.098 0.435 0.436 0.451 0.093 0.412 0.413 0.418