RDP 2013-13: Inventory Investment in Australia and the Global Financial Crisis 6. Results

6.1 Benchmark Model

The results indicate that the treated companies reduced inventories by nearly one-quarter during the crisis, while the matched control companies lowered inventories by about 10 per cent, on average (Table 2). The difference-in-differences estimate, which is the difference between these two trends, indicates that the treated firms reduced inventories by 13.4 per cent relative to the matched control firms. The Abadie-Imbens matching estimator for the ‘average treatment effect’ is the same as the difference-in-differences estimate, except that it adjusts for the bias caused by matching on continuous variables. This estimate indicates that the treated firms reduced inventories by about 18.9 per cent relative to the control firms. The effect is statistically significant at the 10 per cent level.

To gauge the economic significance of the estimate I note that 34 out of the 276 companies in the sample are treated and that companies in the treatment and control groups are observably very similar in size due to the matching process. Therefore, based on the firm-level data, tighter credit conditions reduced aggregate inventory investment by 2.3 percentage points (−0.189*34/276 = −0.023). Under the maintained assumption that debt maturity is a proxy for financial constraints, the results are therefore consistent with the hypothesis that an adverse credit supply shock had a notable impact on inventory investment during 2008/09.

6.2 Varying Treatment Intensity

To examine the robustness of the main results, I make two changes to the model. First, I allow the key explanatory variable – the ratio of short-term debt to total debt – to take on any value between zero and one, rather than restricting it to be a binary indicator. This effectively measures the ‘intensity’ of treatment in the model. So, for instance, firms with very high ratios of short-term debt receive very high treatments while firms with very low ratios of short-term debt receive very low treatments. Second, I estimate Equation (2) by OLS to provide a simple point of comparison to the matching estimates. This requires an assumption that the share of short-term debt owed by firms prior to the crisis is not correlated with the error term (i.e. corr(STDEBT_i, Δε_ij) = 0). That is, the identification strategy assumes that firms did not adjust the maturity structure of their debt holdings in anticipation of the global financial crisis.^[14]

The results are shown in Table 3. In the first column, I present the results of estimating the equation with the dummy variable for whether a firm is in the treatment group or not. In the second column, I present the results with the continuous short-term debt ratio variable.

The negative coefficient estimates for the short-term debt ratio in both columns confirm that firms that had a relatively high share of debt falling due in 2008/09 significantly reduced their inventories. The OLS estimate in column 1 indicates that the treated firms reduced inventories by 23.1 per cent relative to the untreated firms. This is larger than the estimated effect obtained from the matching approach.

The estimate in column 2 suggests that the effect of debt maturity on inventories increases with the intensity of the treatment. The estimate implies that a 1 percentage point increase in the share of debt that is due within one year is associated with a 25 per cent decline in inventories, on average. At the firm-level, during the crisis period, the standard deviation of the share of short-term debt was about 30 percentage points (with a mean of 38 per cent), so a one standard deviation increase to the share of short-term debt is estimated to lower inventories by about 7.5 per cent (=−0.250*0.30*100). This suggests that a one standard deviation increase to the share of short-term debt has a relatively large negative impact on inventories.

The coefficient estimates on the control variables are generally signed as expected and statistically significant. For instance, firms with higher sales growth, lower levels of leverage, and lower stock levels generally experienced higher levels of inventory investment during the crisis, on average. The level of inventory investment is negatively related to the ratio of trade credit to assets, which suggests that trade credit did not act as an alternative source of financing for firms constrained by intermediated credit. In contrast, firm size and liquidity (as measured by the cash flow-to-assets ratio) are insignificant determinants.

6.3 The Endogeneity of Debt Maturity Choice

The preceding analysis assumes that firms' choices about their debt in 2007/08 were exogenous to investment and financing decisions in 2008/09. Moreover, the analysis assumes that having a large share of debt fall due in 2008/09 caused firms to become more financially constrained and this, in turn, caused inventory investment to fall. But there are a couple of reasons why the debt maturity choice may not be truly exogenous.

First, firms that are particularly reliant on short-term debt may be riskier, on average, and this unobservable credit risk may explain the link between debt maturity and inventory investment. To see this, consider the following example. Suppose the ‘unlucky’ firms had established access to revolving credit facilities, but the prospects of these firms deteriorated by relatively more than other firms as the economy weakened. If these firms became concerned about solvency they may have liquidated their inventories and drawn down their existing credit facilities, causing a rise in short-term debt obligations. In this case, financial constraints contributed to the decline in inventory investment, but the observed debt maturity choice would be an outcome of the financial constraints rather than a cause.

Second, if an adverse shock to fundamentals caused investment opportunities to dry up, the ‘unlucky’ firms may have found it harder to secure long-term financing and lenders may have forced them to increase short-term borrowing instead. Short-term debt would be again a symptom of adverse economic shocks rather than a cause (Benmelech and Dvir 2013).

To examine these alternative explanations for the observed relationship between debt maturity and investment, I adapt the simple OLS regression from the previous section by splitting the key explanatory variable – the share of debt that is short-term – into two components – the share of fixed-term debt and the share of revolving debt falling due within the year. I then examine which component is most correlated with inventory investment.

The logic is that there are at least two reasons why a firm could have a high share of short-term debt. The firm may have issued a fixed-term loan several years ago and it just so happens that a large proportion of this long-term debt falls due within the coming year. Alternatively, the firm may have drawn down on its revolving credit facilities during the year so a large share of debt is outstanding by the end of the year. If the debt obligations were undertaken by the firm several years before the crisis they are unlikely to represent an endogenous response to deteriorating economic conditions. Using the share of maturing long-term debt rather than just the outstanding share of short-term debt may help to gauge whether the effect of debt maturity on investment is causal or not.

To undertake this test, I use an alternative source of company accounts information as Morningstar do not provide the necessary firm-level split between revolving and fixed-term credit. Instead, I collect the firm-level credit data on an annual basis from Compustat. Estimating Equation (2) using the Compustat dataset provides a useful cross-check on the main results based on the Morningstar dataset. More information about the Compustat data is reported in Appendix A.

The results again provide strong evidence of an inverse relationship between the share of short-term debt and inventory investment. This is shown by the negative coefficient estimate on the short-term debt variable in column 1 of Table 4. But the results indicate that this inverse relationship reflects a negative association between short-term revolving debt and inventory investment (column 2). There is also a negative correlation between short-term fixed debt and inventory accumulation but this relationship is not statistically significant.

So this test casts some doubt on a causal interpretation of the link between debt maturity and inventory investment. Instead, the results suggest that the ‘unlucky’ firms were unobservably different to the ‘lucky’ firms in terms of credit risk, and that these unobservable characteristics may be driving the link between debt maturity and investment. More specifically, the evidence is consistent with the least creditworthy firms drawing down their existing revolving credit facilities while also cutting back their stocks in order to free up liquidity.

6.4 Placebo Test

I also conduct a ‘placebo test’ in which I ‘pretend’ the shock to credit conditions occurred at a point in time other than the global financial crisis. The experimental design assumes that the actual shock to credit conditions occurred in 2008/09 and examines how inventory investment was affected that year. But, suppose instead that the experiment is run on the assumption that the shock occurred in 2005/06 and we examine how inventory investment was affected at that time. If we still find evidence of a significant effect of debt maturity (the proxy for credit conditions) on inventory investment, then this would suggest there is some confounding factor driving the correlation, rather than a causal relationship.

To conduct the placebo test I estimate the Abaide-Imbens matching regression, but separately run the model on each of the two years before the global financial crisis as well as the two years after the crisis. If the model is capturing a genuine causal effect of the credit supply shock, then the effect should be largest (and only statistically significant) in the actual year of the crisis. The coefficient estimates from each year are shown in Figure 13.

As expected, the coefficient estimate from the model is largest (and only statistically significant) when the treatment is assumed to occur in 2008/09. This coincides with the ‘actual’ crisis period. So, in this case, the results support the claim that the regression estimates are capturing a true causal relationship between debt maturity and inventory investment.

Footnote

I have also estimated quantile regression models, which show that the relationship between debt maturity and inventory investment holds across the entire cross-section of firms and is not specific to certain points in the distribution. These results are available upon request. [14]