An Adaptation of Evolutionary Strategies for Forecasting the Exchange Rate

In this paper, we propose an adaptation of evolutionary strategies (ES) for forecasting the exchange rate. The proposed method employed the evolution of the functional form as well as its coefficients. Using mutation, the functional form is evolved from an initial population. Evolution strategies is used to search for coefficients of functions. We used the data of bath/us-dollar exchange rate from Bank of Thailand during the year 1998. The result is validated using 10-fold cross validation method. The error is less than 5% on training data and testing data. The error on best predictor on testing data is 1.19%. The main contribution of the proposed method is the evolution of the functional form as a separate phase and then using ES to evolve the real value coefficients. The proposed method has been shown to work well in the forecasting task. It is expected to be suitable for variety of tasks that the functional form are not known apriori.