Testing for jumps in GARCH models, a robust approach

Financial series occasionally exhibit large changes. To deal with those events, we assume that the observed return series consists of a conditionally Gaussian ARMA-GARCH (or -GJR) model contaminated by an additive jump component. In this framework, we propose a new test for additive jumps. The test is based on standardised returns, where the first two conditional moments of the non-contaminated observations are estimated in a robust way. Simulation results indicate that the test has very good finite sample properties, i.e. correct size and high proportion of correct jump detection. We apply our test on daily returns and detect less than 1% of jumps for the three exchange rates and between 1 and 3% of jumps for about 50 large capitalization stocks from the NYSE. Once jumps have been filtered out, all series are found to be conditionally Gaussian. We also find that simple GARCH-type models estimated on filtered returns deliver better out-of-sample forecasts of the conditional variance than GARCH and GAS models estimated on raw data. Maastricht University, School of Business and Economics, The Netherlands and CORE, Belgium. E-mail: [email protected] CeReFim, Université de Namur, Belgium and Maastricht University, School of Business and Economics, The Netherlands. E-mail: [email protected] Maastricht University, School of Business and Economics, The Netherlands. E-mail: [email protected] We are also grateful to Torben Andersen, Kris Boudt, Amélie Charles, Olivier Darné, Ángeles Carnero, Rogier Quaedvlieg, Esther Ruiz and participants at the CFE 2012 conference, econometrics seminar at CREATES in Aarhus, economics seminar in Antwerpen and 11th OxMetrics conference in Washington for helpful comments and suggestions. The usual disclaimer applies.