Predictive power of the daily Bangladeshi exchange rate series based on Markov model, neuro fuzzy model and conditional heteroskedastic model

Forecasting exchange rate is very important for many international agents e.g. investors, money managers, investment banks, funds makers and others. We forecasted the daily Bangladeshi exchange rate series for the period of January 1992 to March 2009 using popular non-linear forecasting models, namely Markov switching autoregressive model, fuzzy extension of artificial neural network model (ANFIS) and generalized autoregressive conditional heteroscedastic model. Our target is to investigate whether selected models can serve as useful forecasting models to find volatile and non-linear behaviors of the considered series. By most commonly used statistical measures: mean absolute percentage error, root mean square error and coefficient of determination, we found that ANFIS is a superior predictor than other two selected predictors. We believe findings of this paper will be helpful to make a wide range of policies for multinational companies who are involved with various international business activities.