Golden, R. M. (March 2, 2000). Discrepancy Risk Model Selection Tests for Comparing Possibly Misspecified or Non-nested Models Comments would be appreciated!! Submitted for publication.
Discrepancy Risk Model Selection Test (DRMST) Theory is an extension of Vuong's (1989) model selection theory developed using the methods of White (1994). The DRMST theory can be used to test the null hypothesis that two given probability models fit the underlying data generating process equally effectively with respect to a pre-specified significance level. The DRMST theory is applicable in situations where the models might be non-nested or misspecified. The DRMST theory also provides the opportunity for including model selection criteria penalty terms such as the AIC and BIC penalty terms. In addition, DRMST theory provides a methodology for hypothesis-testing in the presence of model misspecification for a large class of nonstationary environments where the observations are not necessarily independent or identically distributed. Additional applications of the theory include: (1) comparing models with different functional forms, (2) comparing local minima with respect to one model, (3) comparing competing preprocessing transformation, and (4) comparing nested models when the full model may be misspecified.