Modeling Skewness Determinants in Accounting Research

Abstract

Skewness-based proxies are widely used in accounting and finance research. To study how the skewness of a dependent variable Y varies with conditioning variables X, researchers typically compute firm-specific skewness measures over a short rolling window and regress them on X. However, we show that this standard approach can cause severe biases and produce false findings of both conditional skewness on average and systematic variation in conditional skewness. These biases generalize beyond rolling-window skewness. We develop alternative methods that address these biases by directly modeling the conditional skewness of Y for each observation as a function of X. Simulations confirm that our methods have good type-I errors and test power even in scenarios in which the standard method is severely biased. Our methods are transparent, robust, and can be implemented in a few lines of code. Use of our methods changes a major prior finding.