This paper investigates the validity of the Ohlson [Contemp. Account. Res. 11 (1995) 661] information dynamics (Linear Information Model: LIM) and attempts to improve the LIM. The difficulty concerning the empirical tests of the LIM lies in identifying ν t , which denotes information other than abnormal earnings. Recent papers, such as those of Myers [Account. Rev. 74 (1999) 1], Hand and Landsman [The pricing of dividends in equity valuation. Working paper, University of North Carolina, 1999], and Barth et al. [Accruals, cash flows, and equity values. Working paper (January) (July), Stanford University, 1999], all try to specify ν t by using various accounting information. Instead of tackling this difficult task, this paper focuses on serial correlation in the error terms caused by omitting the necessary variable ν t from the regression equation. The results indicate that adjustment for serial correlation leads to an improvement of the LIM.