Abstract:
We suggest a robust form of conditional moment test as a constructive test for functional misspecification in multiplicative error models. The proposed test has power solely against violations of the conditional mean restriction but is not affected by any other type of model misspecification. Monte-Carlo investigations show that an appro priate choice of weighting function induces high power against various alternatives. We illustrate how to adapt the framework to test also out-of-sample moment restrictions, such as orthogonalities of prediction errors.