Design a Novel Hybrid Model to Prediction Foreign Exchange Rate Based on GMDHMLP Network and Genetic Algorithm

Increasing the complexity of the foreign exchange market as the world's largest financial market, has led to a lot of instability and noise in exchange rate series as a result of a considerable increase in speculator participants. The exchange rate forecast will be disturbed by this great noise. Since the accurate exchange rate forecast is considered as one of the key elements in adopting effective and efficient trading strategies, the attempt to improve the exchange rate forecasting models have always been one of the topics discussed in the financial literature. Structural models, econometrics and artificial intelligence-based models separately did not perform well while facing with noisy series, but hybrid models have been able to provide far more accurate predictions. In this regard, we present a novel hybrid model with the ability to select optimal lags and noise resistance called GA-GMDHMLP, and we will look at its ability to predict exchange rates of EUR/USD and USD/JPY as compared to the Random Walk, ARIMA, GMDH and MLP models. Our results show that the proposed hybrid model provides higher prediction accuracy in out-of-sample data