Accounting-Based Risk Predictions: A Re-examination.

Abstract ABSTRACT: Previous studies by Beaver et al. 1970 and Eskew 1979 indicate that the inclusion of accounting-based risk measures in models used to predict the systematic risk of equity securities enables better predictions than security-market-based models which exclude accounting risk measures. This study finds that accounting risk measures do not improve upon market-based systematic risk predictions. The earlier findings of superior predictive ability for accounting-based forecasts are reinterpreted as due to the instability over time of systematic risk, coupled with a fortuitous "shrinking effect" of the ordinary least-squares regression model. Replications of the analysis using portfolios, risk levels, and contemporary values of the accounting risk variables fail to reveal any advantage for accounting-based predictions. Additional tests for the stability of the relationship between accounting and market-based risk measures indicate instability over time and across groups of companies.