Debiasing the Measurement of Conditional Conservatism

Abstract

Basu’s [1997] measurement of conditional conservatism as the asymmetric timeliness of earnings underlies hundreds of studies. However, many subsequent studies cast doubt on the extent to which Basu’s measure captures conditional conservatism versus statistical biases or alternative constructs (collectively, “biases”), thereby questioning the validity of the inferences that empirical researchers draw from analyses using the measure. We modify Basu’s measure in four simple ways to remove these biases. Our key modification is the inclusion of interactive controls for return variance, a volatility proxy that captures Patatoukas and Thomas’ [2011] return variance effect and various sources of economic optionality and adjustment costs. This inclusion captures volatility-related effects on both the level of earnings and the sensitivity of earnings to returns, and it allows the magnitudes of these effects to vary with the sign of returns. We conduct validation analyses using placebo dependent variables, synthetic returns, and non-conditionally conservative earnings components that show our modified Basu measure is largely free of known biases. We further show that our measure is associated with contracting and other economic variables as predicted by theory. Our findings suggest that researchers can rely on our modified Basu measure to identify the determinants and effects of conditional conservatism.

Keywords: Conditional conservatism, Asymmetric timeliness

JEL Classification: C23, D21, G32, M4