An Analytical and Empirical Measure of the Degree of Conditional Conservatism

Abstract

There is a profound gap between models of accounting conservatism and the proxies for conditional conservatism currently used by the empirical literature. Not one of the proxies employed by the empirical literature to date obtains from a rigorous definition of conditional conservatism. In contrast, this study defines conditional conservatism in terms of truncated distributions and derives analytically a nonlinear relation between revisions to returns and earnings news for the conservative firm. This nonlinear relation is shown to be mathematically equivalent to two linear relations conditioned on the firm’s degree of conservatism (DCON). From these relations, we derive a model-based proxy of the DCON at the firm-year level, which is a function of the determinants of conditional conservatism. To account for the endogeneity of the firm’s DCON and mitigate sample selection bias, the model is implemented empirically using a switching regression approach in which the switch point, namely, the DCON, is unobservable and endogenously determined. Consistent estimates of the parameters of the switching regression, including the endogenous determinants of conservatism posited by Watts (2003a, 2003b), are obtained by simultaneous maximum likelihood estimation. The results indicate that the DCON is a positive function of contractual information asymmetry and litigation risk but a negative function of taxes.