Economics Seminar: Non-parametric Likelihood Ratio Tests for Goodness-of-Fit and Specification

When?

Wednesday 14 November 2012, 16:00 to 17:30

Where?

40AD00

Open to:

Public, Staff, Students

Speaker:

Prof. Patrick Marsh

Prof. Patrick Marsh (University of Nottingham)

"Non-parametric Likelihood Ratio Tests for Goodness-of-Fit and Specification"

Abstract

Goodness-of-fit procedures, such as the Kolmogorov-Smirnov or Cramer-von-Mises, form the basis of tests in a variety of circumstances, such as model specification, stochastic dominance etc. In the presence of nuisance parameters, however, these tests are generally non-pivotal and their asymptotic properties need to be worked out case-by-case. In this work, instead, non-parametric likelihood ratio tests based on an exponential series density estimator are shown to be asymptotically pivotal and consistent against fixed alternatives. Direct finite sample numerical comparisons also indicate equivalence in size performance but significantly more power in some cases. Further examples involve application to specification tests in linear regression, specification of conditional variance and testing in the proportional hazards model.