A0543
Title: Testing for relevance of partially parametric models with parametric nulls
Authors: Daniel Henderson - University of Alabama (United States) [presenting]
Jiancheng Jiang - UNC Charlotte (United States)
Abstract: Tests of relevance are considered for partially parametric models. Specifically, it is tested that the entire nonparametric function is irrelevant in predicting the outcome variable. This results in parametric null models, which can be estimated via (non-linear) least-squares. The asymptotic theory is developed for our test statistics (both under the null and versus local alternatives), and valid bootstrap procedures are proposed for use in finite samples. Further testing for the relevance of a control function in simultaneous equation models (i.e., test for exogeneity) is considered. This requires us to extend our theory to account for generated regressors. Then a joint test for correct parametric specification and irrelevance is considered. Finally, an omnibus version of our test with improved power is considered. Simulations and an empirical exercise suggest that the tests perform well in finite samples.
