Expert system to forecast exchange rate behaviour towards news surprises : an application to EUR/USD exchange rates

![CDATA[Time series forecasting has become a widely discussing area during the recent past. Most of the real life systems turn out to be more complex with the influence of numerous external factors. As a result forecasting such system becomes a challenging task. Even a sophisticated methodology may not produce results as expected unless it is supported with the correct set of inputs. Hence, the selection of inputs is viewed as a crucial step in the forecasting procedure. This study aims at proposing a novel approach to derive high frequency forecasts of exchange rates at times of news releases. The proposed approach selects the dataset dynamically with respect to the existing news release. A two step dynamic procedure comprise of identification of change points in variances and estimation of volatility using release time region data is proposed to quantify the major input which is the volatility. Moreover, the study develops an expert system driven by a feedforward neural netwok supplied with dynamic inputs. Comparison of forecasting performances of the proposed expert system with traditional ARIMA model and ANN fed with historical data concluded the superiority of the proposed system, with respect to the mean square error relevant to out of sample forecasts and percentage of profit opportunities.]]