A consistent research design for value relevance studies

We argue that Ohlson’s linear solution to the residual earnings (RE) equation, a crucial component of a widely used value relevance research design, is generally not a linear regression. Moreover, its coefficients are firm-dependent. As such, its empirical specifications, the price-levels regression and the returns-earnings regressions are structurally ill-suited for consistent inference in cross-sections. To address this issue we, first, prove the existence of a regression solution to the RE equation and, second, introduce a valuation-based research design that builds on such a solution and warrants a consistent estimation of the empirical specification (which takes the form of a non-linear regression). Its estimation turns out to be an optimal implementation of the price-to-book (P/B ) multiple valuation, a technique that is easy to implement and familiar to the accounting community. The regression view on multiple valuation identifies P/B value with a price that incorporates earnings expectations formed only on the basis of the current levels of the RE drivers. Using a large sample of US non-financial firms over an almost 40 year-period, we document the usefulness of the alternative research design through a comparative testing of four economically-motivated and intuitively-appealing predictions: size, earnings predictability and volatility, and the quality of accruals are value-relevant. While the current research design does not validate them, the approach based on the regression solution shows a significant association between prices and the four attributes for most of the years in the sample.