Linear Information Models: The Effect of Unconditional Conservatism

Abstract

A large literature develops and tests valuation models that assume a link between equity values and the time series behavior of accounting numbers. With respect to predicting equity values, linear information models (LIMs), such as the Ohlson (1995) model, have been outshone by models implemented in perfect foresight settings or ex-ante approaches based on analyst forecasts. The major challenge has been attempting to tackle the large negative bias reported by empirical studies testing LIMs. These studies suggest that the violation of the assumption of unbiased accounting underlying the Ohlson (1995) model causes a systematic negative bias. Hence, we expect that if conservative accounting is adequately incorporated into the models, valuation errors should be reduced. Recent research refines the conservatism corrections of LIMs, e.g. Choi/O’Hanlon/Pope, 2006. The findings are encouraging as bias is substantially reduced. Yet, it is puzzling that inaccuracy remains high. These results raise two questions, which we address in this study: First, do the conservatism corrections of different LIM implementations capture conservative accounting, and to what extent? Second, if conservatism is captured, then why is accuracy not markedly improved?