Title: Using quantile regressions for identifying mixed causal-noncausal models
Authors: Li Sun - Maastricht University (Netherlands)
Alain Hecq - Maastricht University (Netherlands) [presenting]
Abstract: Mixed causal-noncausal models are gaining attention, for instance, in modelling bubbles and asymmetric cycles in economic and financial time series. The aim is to extend a previous model selection approach for purely causal and noncausal models to mixed models with both lags and leads components. The estimation of mixed causal-noncausal models is carried out using quantile autoregressions in both direct and reverse time. Our framework is able to choose among several parameter sets those that minimize the sum of absolute rescaled residuals. Monte Carlo simulations and an application on real data illustrate the feasibility of our approach.