Entropy-balanced accruals

Abstract

This study assesses whether the accrual-generating process is adequately described by a linear model with respect to a range of underlying determinants examined by prior literature. We document substantial departures from linearity across the distributions of accrual determinants, including measures of size, performance, and growth. To incorporate non-linear relations, we employ a recently developed multivariate matching approach (entropy balancing) to adjust for determinants in place of relying on a linear model. Entropy balancing identifies weights for the control sample to equalize the distribution of determinants across treatment and control samples. In simulations drawing random samples from deciles where a linear model displays poor fit, we find that entropy balancing significantly improves accrual model specification by reducing coefficient bias relative to linear and propensity-score matched models. Consistent with entropy balancing retaining sufficient power, we find that its estimates detect seeded accrual manipulations and explain variation in accruals around equity issuances.