We develop regression-based tests of hypotheses about out of sample pr
ediction errors. Representative tests include ones for zero mean and z
ero correlation between a prediction error and a vector of predictors.
The relevant environments are ones in which predictions depend on est
imated parameters. We show that standard regression statistics general
ly fail to account for errors introduced by estimation of these parame
ters. We propose computationally convenient test statistics that prope
rly account for such errors. Simulations indicate that the procedures
can work well in samples of size typically available, although there s
ometimes are substantial size distortions.