Title: Exact and heavy-tail robust inference in GARCH models
Authors: Richard Luger - Laval University (Canada) [presenting]
Abstract: A procedure is developed for building exact confidence intervals in GARCH models without any restrictions on tail heaviness. The considered GARCH models may even be subject to variance targeting. The approach uses profile quasi-likelihood ratios with Monte Carlo resampling techniques to obtain exact bounds tests. These level-exact tests are then inverted to produce conservative confidence intervals for the model parameters. The endpoints of the confidence intervals can be found quickly under certain conditions by a combination of bisection and grid search. Daily returns on major stock market indices are used to illustrate the exact inference procedure in a variety of GARCH specifications.