At any point in time, most firms are not in financial distress. This implies that they must suffer value losses unrelated to their leverage-economic shocks-before becoming financially distressed. We show that if estimates of ex-ante financial distress costs are not filtered from the effects of future economic shocks, they are significantly biased upward, as far as an order of magnitude. Filtered from economic shocks, pure ex-ante distress costs average less than 1% of current firm value. We

more »

... estimate sensitivities of ex-ante distress costs to leverage that are generally far too small to offset the expected tax benefits. Extending our analysis to the cross section and time series, we confirm that ex-ante distress costs are highest: i) when the risk premium in debt markets is high, and ii) among firms with high systematic risk. Overall, our results suggest that most firms use debt too conservatively, but we characterize conditions under which they do not. multiplying these figures gives a present value of financial distress costs of 2.3%. However, this calculation overstates ex-ante financial distress costs, because it does not properly filter out economic shocks that would lead the firm to become distressed. 4 For example, if the firm can suffer a 70-percent economic decline before becoming distressed (i.e., it will begin incurring financial distress costs at $30), then the estimated ex-ante financial distress costs are too high by over a factor of three (2.3% vs. (10%)(23%)($100 -$70)=$.69, or .69% of current firm value). Sensitivities to leverage will be similarly inflated. As this example illustrates, accounting for economic shocks in ex-ante financial distress calculations can make a substantial difference. This is also true in aggregate data. Frank and Goyal (2007) find that the average ratio of total debt to market value of assets from 1950-2003 is around 0.29, such that, at any point in time, a firm would need to lose over 70% of its value before defaulting. But, even this may overstate the economic losses required for firms to experience financial distress. Davydenko (2007) examines the market values of firms at default, and finds that, on average, they constitute some 60% of nominal debt obligations. Such a typical case would imply that the required shock to trigger default is about 82% (100% − 30% · 60%). This value, some 18% of the firm's value when the ex-ante distress calculation is made, corresponds to the benchmark used by ex-post studies such as Andrade and Kaplan (1998) , and allows for a "pure" calculation of financial distress costs. Calculating ex-ante distress costs firm by firm thus requires a model for values at or near default, so that future economic shocks can be filtered out. A natural approach is to use a structural model of default that provides both risk-adjusted default probabilities and values at default within a unified framework. 5 In this paper, we implement the model of Leland and Toft (1996) , hereafter LT. Not only is LT tractable and easily estimated, but it also fits Davydenko's (2007) empirically estimated average values at default quite well. This benefit notwithstanding, neither our main point nor results are sensitive to the default rule specified by the LT model. In addition to performing a number of robustness checks within the LT model, we reestimate everything using an entirely different default specification (Merton (1974) ). Neither alters our main conclusions. This is unsurprising given that the differences in default boundaries across specifications are inconsequential, relative to not modeling default values at all. The basic mechanics of the LT model are standard. The firm has two sources of value: 1) unlevered assets whose value can be described by a random process, and 2) debt that generates a tax shield against the firm's 4 Note that this is not an issue in empirical work relying on a model with endogenous default (e.g. Hennessy and Whited (2007)) or papers that estimate the ex-ante value effects of debt directly (e.g., Korteweg (2008) ). 5 A necessarily partial list of these includes Merton (