Trading Macro-Cycles of Foreign Exchange Markets Using Hybrid Models

Abstract

Most existing studies on forecasting exchange rates focus on predicting next-period returns. In contrast, this study takes the novel approach of forecasting and trading the longer-term trends (macro-cycles) of exchange rates. It proposes a unique hybrid forecast model consisting of linear regression, multilayer neural network, and combination models embedded with technical trading rules and economic fundamentals to predict the macro-cycles of the selected currencies and investigate the predicative power and market timing ability of the model. The results confirm that the combination model has a significant predictive power and market timing ability, and outperforms the benchmark models in terms of returns. The finding that the government bond yield differentials and CPI differentials are the important factors in exchange rate forecasts further implies that interest rate parity and PPP have strong influence on foreign exchange market participants.