Residual income (RI) valuation is a method of estimating firm value based o
n expected future accounting numbers. This study documents the necessity of
using linear information models (LIMs) of the time series of accounting nu
mbers in valuation. I find that recent studies that make ad hoc modificatio
ns to the LIMs contain internal inconsistencies and violate the no arbitrag
e assumption. I outline a method for modifying the LIMs while preserving in
ternal consistency. I also find that when estimated as a time series, the L
IMs of Ohlson (1995), and Feltham and Ohlson (1995) provide Value estimates
no better than book value alone. By comparing the implied price coefficien
ts to coefficients from a price level regression, I find that the models im
ply inefficient weightings on the accounting numbers. Furthermore, the medi
an conservatism parameter of Feltham and Ohlson (1995) is significantly neg
ative, contrary to the model's prediction, for even the most conservative f
irms. To explain these failures, I estimate a LIM from a more carefully mod
eled accounting system that provides two parameters of conservatism (the in
come parameter and the book value parameter). However, this model also fail
s to capture the true stochastic relationship among accounting variables. M
ore complex models tend to provide noisier estimates of firm value than mor
e parsimonious models.