This paper studies generalized method of moments tests for the stochastic d
iscount factor representation of asset pricing models when one of the propo
sed factors is in fact useless, defined as being independent of the asset r
eturns. Analytic results on asymptotic distributions and simulation results
on finite sample distributions both show that(i) the Wald test tends to ov
erreject the hypothesis of a zero factor premium for a useless factor when
the model is misspecified, (ii) with the presence of a useless factor, the
power of the over-identifying restriction test in rejecting misspecified mo
dels is reduced, and in some cases a misspecified model with a useless fact
or is more likely to be accepted than the true model. (C) 1999 Elsevier Sci
ence S.A. All rights reserved.