Multi-Lagged Specification of the Ohlson Model

This paper extends the study by Bar-Yosef, Callen, and Livnat (BCL) (1996), in which they test the single lagged formulation of the Ohlson (1995) linear valuation model by assuming that book values, earnings, and dividends are the underlying primitives in a multi-lagged Garman-Ohlson (1980) framework. Unlike BCL who focus solely on the information dynamic, this paper estimates both the information dynamic and the valuation equation. Also, since Akaike's (1969a, 1969b) final prediction error (FPE) and Akaike's (1973, 1974) information criterion (AIC) are known to overfit in small samples, the lag structure specification in this study is estimated by the AICe criterion suggested by Hurvich and Tsai (1989, 1991, 1994) rather than by the FPE criterion of BCL. Furthermore, to bias the case against a multi-lagged structure, the maximum lag structure is limited to three lags per variable rather than the five lags of BCL. Utilizing the result that the lag structure of the information dynamic determines the lag structure of the valuation equation, the internal consistency of the model is evaluated from the estimated lag structures. The median lag structure for the valuation equation implied by the lag structure of the estimated information dynamic is found to be somewhat inconsistent with the median lag structure estimated directly from the valuation equation. Further testing on a stationary subsample suggests that the inconsistency may be induced by nonstationarities. Overall, the results indicate strongly that the modeling of earnings and book values in a linear valuation framework necessarily requires a multi-lagged formulation rather than the single lag formulation common to the empirical accounting literature.