In this part of the paper we analyze the firms on direct costs and recovery rates. We proceed as follows. First, we give an overview of the summary statistics of the relevant variables (Sect. 4.1). Second, we present our hypothesis and regression results for direct costs (Sect. 4.2), firm recovery rates (Sect. 4.3) and bank recovery rates (Sect. 4.4). We run cross-sectional ordinary least squares regressions. We first measure the influence of the firm characteristics, second the effect of the time it takes to resolve bankruptcy, third, process characteristics and fourth, we run a regression with the statistically significant variables.
Table 2 provides the summary statistics of the variables. The column with the full sample contains the averages and medians per variable. In the column for the bank loan sample we provide these statistics for the 114 observations with bank debt. The average direct costs are 16.0% of the realized proceeds of the bankruptcy procedure.Footnote 7 The median value of 11.2% and the values in the bank loan sub sample (average of 13.7% and median of 9.4%) are relatively close to this percentage. The average firm recovery rate is 37.2% of total debts. This implies that creditors lose on average 62.8% of the nominal outstanding debt, i.e. the original debt investments minus prior amortizations. The bank recovery rate is strikingly different. On average 80.0% of the value of bank loans is recovered. The median of 99.5% shows that the average is influenced by a small number of low recoveries. In fact, 59.6% of the observations have a recovery above 90%; 22.8% have a recovery between 50% and 90%; and 17.5% have a recovery below 50%.
The average size of the firms in the sample is 10 million Dutch guilders (€4.54 million), while the median is 2.7 million Dutch guilders. The firms have on average 37.3% fixed assets and the average quick ratio is only 0.560. The firms are heavily indebted, as the mean debt ratio is 2.081. In other words, on average the amount of debt is twice the asset value. In the firms with bank debt, this form of debt is 29.0% of total debts. The bank debt is well-secured, because the average size of the collateral is 2.5 times the bank debt. Clearly, on average the banks are extremely careful and require substantial collateral. In our regression analyses, we test whether this conservatism helps in recovering debts. In the book year before the bankruptcy 61.3% of the firms have a positive operating return.
The remaining variables in Table 2 summarize the use of legal bankruptcy procedures in The Netherlands. On average the firms in our sample spend 25 months in the bankruptcy procedures. In 72.3% of the cases assets are sold directly after the procedure has started. On average, it takes 3.4 months to sell the assets. The average number of buyers involved in the sale is 2.291, while the median is one. In 46.7% of the firms, the activities are continued during the procedure. We find conflicting rights on compound bounded assets in 40.9% and procedures started by the trustee in 27.0% of the cases. The average number of disputes is 1.036. In 32.1% of the cases a stay period of 1 or 2 months is allowed, while the average number of months of the stay is 0.401. This implies that 8% of the firms have the extended stay period of 2 months.Footnote 8 The outcome of the procedure is liquidation in 27.7% of the cases. On average, the new management gets 11.9% of the shares. However, this statistic is driven by 19 cases with a non-zero stake.Footnote 9
Our first set of regression tests aims to explain the relative direct costs. Table 3 contains the hypotheses and the regression results.
Table 3 provides the results of four regression models, relating the explanatory variables to direct costs as percentage of total realized proceeds. In model (1) we run regressions of firm characteristics on direct costs. With respect to size, we presume that bankruptcy costs have a large fixed component. This leads to the hypothesis that firm size has a negative effect on relative costs. Because we expect that the marginal influence of size decreases, we apply a log-scaling. The regression coefficient for total assets is negative as hypothesized and the coefficient is significant at the 1% level. To illustrate the effect of size, we calculate the difference between the logarithms of total assets of the 25 and 75th percentile. Asset size changes from 940,000 to 7,740,000 and the difference of the logarithms is 0.916. This implies that the costs decrease by 4.4% points (0.916 times −0.048) when firm size changes from the 25 to the 75th percentile. The fraction of fixed assets, the quick ratio and debt are insignificant. Relative bank debt is significant at the 10% level and has the hypothesized sign. Bank debt is expected to save on costs, because the higher relative bank debt, the less the trustee will have to deal with other creditors than the bank. Economically this effect is also significant, as 1% more bank debt reduces relative costs by 0.126% points. We are glad to notice that the adjusted R
2 of the model is as high as 0.28, which is an excellent fit for a cross-sectional regression.
In model (2), we include variables for the length of the procedure and we omit insignificant variables in model (1).Footnote 10 Obviously, we hypothesize that the time in months has a positive effect on costs. Again, as we expect that the marginal influence of time decreases, we apply a log-scaling. Our presumption regarding the dummy variable DirSellAssets, indicating whether assets are sold directly, is that this lowers costs. However, if this sale takes more time (TimeSellAssets) the trustee needs to spend more effort selling these assets, and thus costs are expected to rise. It is interesting to find that the coefficient of TimeProc is insignificant. The implication is that, although Dutch bankruptcies may take many months, costs are independent of the length of the procedure, after controling for firm size, bank debt and the time it takes to sell assets. Thus, as we find that the variable TimeSellAssets is significant on 5% level and has the predicted sign, we conclude that direct costs are dependent on the effort and time it takes to sell the assets of the firm and not on the period of time that the bankruptcy procedure is running. The most plausible explanation for this finding is that the total procedure time includes substantial periods of non-activity, while this does not hold for the periods during which assets are sold.
In model (3) we add additional variables for the legal procedures. Filing by the debtor, continuation in bankruptcy, conflicting creditor rights and procedures started by the trustee all turn out to have insignificant impacts on costs. This implies that in the Dutch legal setting these procedures do not lead to measurable inefficiencies. The number of disputes has a positive effect on costs and the coefficient is significant on the 5% level. More disputes by creditors give rise to additional costs. The automatic stay variables both are insignificant, implying that these do not give rise to additional costs. The effect of the piecemeal liquidation dummy is insignificant, which indicates that no costs savings are to be expected in a liquidation situation. It is interesting to notice that the adjusted R
2 changes from 0.280 in model (1) to 0.324 in model (3). This result stresses that the firm characteristics in model (1) have superior explanatory power, in comparison with the legal procedure variables that are added in model (3). In model (4) we include the significant variables from previous models and the dummy for direct sales, because this variable is related to the time to sell assets. The results show that the effects are robust and that the adjusted R
2 of the model improves to 0.328.
In summary, our results for the determinants of direct bankruptcy costs show that firm size and bank debt have a negative effect on the costs. On the other hand, a longer time to sell assets and a larger number of disputes lead to higher costs. These results are in line with the hypotheses. Thorburn (2000) and Sundgren (1998) also document a similar effect of firm size. Thorburn finds that the length of the procedure is a significant determinant of costs, while we document that the period needed to sell assets matters more.
Firm recovery rates
In this section we report regression analyses of firm recovery rates on firm characteristics and procedural characteristics in order to measure which elements in a liquidation-based system are likely to positively influence recovery rates. Table 4 describes the hypotheses and the regression results.
In Table 4, model (1) includes the influence of firm characteristics and shows that the value of total assets is negatively related to the firm recovery rate, but the coefficient is insignificant. Both asset structure variables, fraction of fixed assets and quick ratio, have significantly positive coefficients. The fraction of fixed assets is likely to have a positive sign, because it is a proxy for saleable assets and also inversely related to intangible assets. For the quick ratio we also hypothesize a positive effect on recovery, because companies with more liquid assets have a higher recovery potential. Obviously, for leverage (Debt) we predict a negative coefficient, as more indebted firms have simply more debt to recover. However, we hypothesize a positive effect of bank debt, because banks will put in more effort in the recovery process in case their part of the total liabilities is larger. The estimates corroborate our hypothesized effects. Particularly, the coefficient of bank debt of 0.313 is high. In case a firm has the median amount of bank debt of 21.4%, the recovery rate is 6.7% point (21.4 times 0.313) higher, in comparison with a firm without bank debt. The adjusted R
2 of the model is 0.250.
In model (2) we drop insignificant variables and include time-related variables. None of these variables obtain significance. Clearly, both the length of the bankruptcy process and the time to sell assets do not influence the recovery rates. In model (3), we introduce the procedural characteristics. Continuation in bankruptcy is hypothesized to be positive for the recovery rate, because it indicates that the firm has valuable activities, which may yield higher asset prices. The liquidation dummy is expected to yield a negative coefficient, because realized values are normally lower in piecemeal liquidations, compared to going concern asset sales. We find that the variables continuation in bankruptcy and piecemeal liquidation are significant, respectively at 5% and 10% level. Both coefficients also have the hypothesized sign. The levels of the coefficients imply that the recovery rate increases by 8.1% point when the trustee continues the operations and decreases by 6.1% point in case of liquidation. Given that the average recovery rate is 37.2%, these two variables have a major impact on the creditors’ proceeds. Conflicts, procedures and disputes do not affect recovery rates, nor does the involvement of management. The adjusted R
2 in model (3) is 0.281, which again indicates a minor improvement relative to the firm characteristics in model (1). In model (4) we include only the significant variables in previous models and find that the results are robust.
We find that the firm recovery rate is higher when firms have more fixed assets, a higher quick ratio, are not liquidated and continue their operations in bankruptcy. These results are in line with expectations. We also find that the effect on recovery is negative for leverage and positive for bank debt. Our results confirm previous tests. Thorburn (2000) documents that recovery rates in Swedish firms are influenced by secured (bank) debt and the outcome of the procedure. Sundgren (1998) finds that indebtedness is a significant determinant of recovery. Bris et al. (2006) report similar results with respect to size (not or only weakly relevant), leverage and secured debt. Gilson et al. (1990) report that distressed exchange offers in the US are more successful when the debt structure is more concentrated, which is in line with our result for bank debt. The results show that even in liquidation-based system it helps to have concentrated bank debt.
Bank debt recovery rates
In this section we focus our attention on a specific type of debt, i.e. bank debt. Banks were involved as creditors in a sub sample of 114 firms. For this sample we test for the determinants of the recovery rate of the bank debt. Table 5 provides the hypotheses and regression results.
Table 5 includes the same variables as present in the analyses on the firm recovery rate. Regression model (1) includes the firm characteristics. Total assets yields a negative coefficient that is significant at the 10% level, indicating that the larger firms in our sample yield a lower bank debt recovery ratio. This finding contrasts with the hypothesis.Footnote 11 The two variables for asset structure have insignificant coefficients. As hypothesized, the debt ratio has a negative impact on the bank’s recovery rates, a result that is significant at the 1% level. The portion of bank debt does not have a significant coefficient. The dummy for a positive return is also insignificant. In model (2) we eliminate the insignificant variables from model (1) and add the secured bank debt and the time variables. The variable for the secured bank debt yields a positive coefficient, which is significant at the 10% level. A plausible explanation for this effect is the absence of the effects for fixed assets and the quick ratio, which were present in the firm recovery regressions. In the discussion of the summary statistics we have noticed that the bank debt is well-secured, as the collateral is on average 2.5 times the bank’s debt. Clearly, the collateral helps the bank in recovering their loans.Footnote 12 The three coefficients for the length of the procedure, including the time to sell assets, are insignificant.
Model (3) takes up the procedural variables. In case the debtor files for bankruptcy the bank’s recovery rate is significantly higher. The dummy for the automatic stay is negative and significant (5% level), while the period of the stay is positively significant (5% level). Because the stay period can be absent, 1 or 2 months, this result should be interpreted as follows. The base situation is no stay, where both variables have a value of zero. In case of a one-month stay, Stay equals one and StayPeriod equals one, leading to an aggregate effect of −0.102 (−0.306 + 0.204). In case of an extended stay of 2 months, StayPeriod becomes two, yielding a joint influence of 0.102 (−0.306 + 2 × 0.204). It should be noted that the summary statistics show that 24.1% of the bankruptcies have a one-month stay period, while only 8.0% has the extended stay. The liquidation dummy is significantly negative at the 5% level. The size of the coefficient implies that liquidated firms have a 16.2% point lower recovery for the bank. In order to assess the added explanatory power from procedural variables, we compare model (2) and model (3). The explanatory power of the model (3) is 0.205, which is higher than the 0.131 in model (2). Apparently, legal procedures have a strong effect on the banks’ recovery rates. Regression model (4) confirms the result of the third model, without any major change in significance of variables.
The results for the bank debt recovery rates show the recovery is lower when firms are larger, more indebted or liquidated. Collateral underlying the bank debt indeed boosts recovery rates. Moreover, when the firm itself files for bankruptcy, the recovery for the bank is higher. The impact of the stay period is ambiguous: the 24.1% of the cases with a one-month stay period are worse off, while the 8.0% with the extended stay have a higher recovery. In comparison with the firm recovery rate, leverage and liquidation have the same effects, i.e. negative. The asset structure influences both recoveries in different ways: fixed assets and quick ratio matter for all creditors, while secured assets matter for the bank.