Accounting data and the prediction of risk in the extremes

Abstract

AbstractIn portfolio formation, the systematic risk of a security is of fundamental importance. In the traditional asset‐pricing model, the concept of beta represents anticipated systematic risk which, to be useful in the real‐world portfolio formation, can only be estimated by currently‐available data.Prior research has revealed variation in the reliability of the prediction of beta across the risk spectrum. In this paper we focus on beta prediction in the extreme risk categories and consider the predictive contribution of accounting information across the risk spectrum. Our empirical model utilizes accounting risk measures, shown to be related to sample‐wide average betas in prior research, and incorporates the intertemporal relationship between successive period betas in predicting next‐period betas.Our results provide evidence that inclusion of accounting risk measures, alone or in combination with market beta, substantially improves beta prediction for high risk securities, but not for low and medium risk securities. These results are consistent using either traditional OLS or analytically‐corrected (Bayesian‐adjusted) beta estimation techniques.