High frequency volatility forecasting: A new approach using a hybrid ANN‐MC‐GARCH model
提出一种混合ANN-MC-GARCH模型，通过将内生市场变量纳入前馈网络来改进MC-GARCH条件方差模型，利用1分钟高频数据对汇率、股指和金属商品指数进行实证，发现新模型预测精度优于传统MC-GARCH模型。
Abstract Risk permeates more and more financial markets around the world. It is an essential element that all financial market actors attempt to model and manage. This paper proposes a novel model that improves the predictive accuracy of high frequency volatility forecasts. The ANN‐MC‐GARCH model is therefore developed in this research. The hybrid model enhances the MC‐GARCH conditional variance model by including endogenous market variables in a feed forward network which models volatility in terms of past disturbances and variances. The forecasting accuracy of the novel ANN‐MC‐GARCH is evaluated against the classical MC‐GARCH model. The empirical investigation employs the 1‐min high frequency observations of four exchange rates (USD/EUR, USD/GBP, USD/JPY & AUD/JPY), three market indices (France 40, UK 100 & USA 500) and two metal commodity indices (Spot Gold & Spot Silver). The backtesting method employs the RMSE and MAE performance metrics, the out‐of‐sample , the Diebold‐Mariano test and the Model Confidence Set (MCS) procedure. The empirical findings show that the hybrid model is superior to the MC‐GARCH model for the nine datasets.