Title: On LASSO-GARCH approach
Authors: Iryna Okhrin - Dresden University for Technology (Germany) [presenting]
Sasa Zikovic - University of Rijeka Faculty of Economics (Croatia)
Abstract: It is a stylized fact that GARCH type models are good at describing the characteristics of financial data (leverage effects, volatility clustering, etc.). List of various XGARCH models has increased tremendously over the last decade, creating a serious model selection problem. Estimation of the most general model is a challenging task and the results are not easy to explain; it also leads to overestimation. We propose a LASSO-GARCH approach which allows for estimation and simultaneous model simplification. The question estimation and simplification relates not only to the number of lags to be taken into account, but also to the specific characteristics of the data. For example, if no leverage effect is present, there is no need to estimate an asymmetric GARCH model and it is exactly what a LASSO-GARCH procedure avoids. In our LASSO-GARCH study we consider a family of GARCH models and estimate them via a penalized ML approach. Simulation study results show consistency and efficiency of the estimator under different setups. An empirical study confirms the simulation results and shows the superior performance of the approach.