An Application of Causal Forest in Corporate Finance: How Does Financing Affect Investment?

Abstract

Answering a causal question with results extendable outside of a narrow sample is challenging. Regression discontinuity design (RDD) provides causal results with strong internal, but weak external, validity. We introduce a novel machine learning technique, causal forest, into the corporate finance literature because it has several benefits over RDD, including greater precision and extendibility of results. We revisit a well-known RDD setting in corporate finance, debt covenant violations. We show that this RDD framework recovers an overly-large treatment effect of default on investment because firms can manipulate the accounting ratios lenders use to define default thresholds. In contrast to previous literature that shows an economically significant decline in investment following a technical default, we find, on average, no change in investment. Intuitively, this result makes sense because firms that can manipulate their default status will only choose to default if there are no consequences to default. For these firms, which comprise the majority of our sample, the treatment effect of default on investment will be zero. We show that causal forest's ability to recover heterogeneous treatment effects is key to reconciling disparate findings in the debt covenant literature. Causal forest can be a broadly useful empirical technique for answering causal questions.