improved neural network forecasting models for foreign exchange rates using volatility indices

the emphasis of this paper is the role of volatility indices on improvement artificial neural networks (anns) forecasting models for the daily usd/eur and usd/gbp exchange rates two volatility indices are used. first; the realized volatility, which is based on intra-daily data, and second the garch volatility. they are applied into the model in two ways. firstly, the lagged volatility index is added to the model. secondly, some levels for the volatility are defined and the time series are partitioned according to the level of volatility, and then different models of exchange rate forecasting are built for each level of volatility. the forecasting results demonstrate that the models with low and middle volatility are not preferred to the model without volatility index. however, in case of high volatility, the level models improve forecasting power. this means that high volatility provides new information for foreign exchange market. jel classification: f31, f37, c63