CMStatistics 2022: Start Registration
View Submission - CFE
Title: Testing the ATV-GARCH model Authors:  Niklas Ahlgren - Hanken School of Economics (Finland) [presenting]
Alexander Back - Hanken School of Economics (Finland)
Timo Terasvirta - Aarhus University (Denmark)
Abstract: It is common for long financial time series to exhibit a gradual change in conditional and unconditional volatility. The additive time-varying (ATV-)GARCH model allows for structural change with remarkable flexibility. Instead of making all GARCH parameters time-varying, the intercept is parameterised by a logistic transition function with rescaled time as the transition variable. This specification is a parsimonious parameterisation of the very general nonparametric time-varying GARCH. It provides a simple and flexible way to capture deterministic nonlinear changes in the variance, and is particularly well suited for situations in which volatility of an asset or index is systematically increasing or decreasing (or both) over time. The model is unidentified if the intercept is constant. It is, therefore, imperative to test the constancy of the intercept before attempting to fit the model to data. We derive Lagrange multiplier (LM) tests of GARCH against ATV-GARCH. A testing-based modelling strategy is introduced and illustrated by two empirical examples.