The Application of GARCH-VaR models in Chinese Commercial Banks for the Foreign Exchange Rate Risks Management

This study is an attempt to construct and compare a comprehensive list of univariate-GARCH models and multivariate-GARCH models, and then the accuracy of the VaR results is evaluated through Kupiec tests. Nearly five years of exchange rate data of RMB from November 2015 to June 2020 are used. In the light of Kupiec test results on the RMB/USD, RMB/EUR, RMB/100JPY and RMB/HKD return sequences, the models that are most suitable for Chinese commercial banks to measure and forecast foreign exchange risks in the context of RMB internationalisation, the Sino-US trade war and the prevailing coronavirus are selected. Simultaneously, combining the actual net foreign exchange positions of 14 Chinese commercial banks could obtain the exchange rate risks of each bank. The empirical results reveal that the BEKK and DCC models that belong to the multivariate GARCH models are not applicable to predict exchange rate risks for the four groups of currency sequences at a 95% confidence level. Although there are possibilities of applicability at the 99% confidence level, the backtesting effects are still inferior to the ordinary and extended univariate GARCH (1, 1) models. Besides, the BEKK and DCC models tend to overestimate the exchange rate risks under the GED and T distributions since empirical backtesting results show that the failure times of these two models are too few. Keywords: exchange rates risks, Value-at-Risk, univariate GARCH, multivariate GARCH, Kupiec Tes