Abstract:
|
We propose a new test for dynamic asset pricing models using a generalized spectral approach embedded in the stochastic discount factor framework. The procedure can be viewed as a substantive extension of the GMM conditional moments tests to frequency domain analysis. It incorporates the conditional moments at all lags and weights down higher order lags, which is expected to increase the power of the test in most cases, because economic agents usually discount remote past information. Our test can detect a wide range of model misspecifications, including those that prices assets correctly, spatially, but suboptimally over time. It can be applied to both linear and nonlinear asset pricing models. The test allows for heterogeneity and instantaneous spacial dependence across assets. It also allows for ARCH effects in pricing errors for each asset, which many existing test approaches usually ignore. A simulation study shows that the test has reasonable finite sample performances. We apply it to evaluate the dynamic CAMP model and a variety of factor models.
|