Foreign Exchange (FX) Rate Forecasting Using Hybridization

Abstract

Foreign Exchange (FX) market provides a forum where one currency rate is converted into other currency rates, which is based on selling and buying the currency in exchange rates like Indian Rupee to United State Dollar (INR/USD), European Dollar to United State Dollar (EUR/USD), Hong Kong Dollars to United State Dollar (HKD/USD), etc. Hybridization of intelligent techniques are proposed for forecasting FX data with feature extraction and feature selection technique. Here hybridization of various techniques and ANFIS is used to construct a predictive model for next week ahead forecasting (Jang, 1993). Feature extraction is used to extract the new features from existing features. In this work 5 new features: Simple Moving Average (SMA), Exponential Moving Average (EMA), Weighted Moving Average (WMA), Variance and Standard Deviation are extracted. Feature extraction is usually the first step for constructing any predictive model based on time series data. These extracted data are then partitioned dynamically using k-fold cross validation as explained in section 3.5 of chapter 3. In this work, data is partitioned into 10 folds and each subset has taken part as training as well as testing data. These features are given to the ANFIS, ANN, RNN, SVR and models with ensemble regression techniques individually to get the best and accurate predictive result. The predictive results are based on certain performance measures: Mean Absolute Percentage Error (MAPE), Mean Absolute Deviation (MAD), Mean Absolute Error (MAE) and Root Mean Square Error (RMSE).