Investigation of exchange market prediction model based on high-low daily data

Abstract

The model of Evolino recurrent neural networks (RNN) based on ensemble for prediction of daily extremes of financial market is investigated. The prediction distributions of each high and lows of daily values of exchange rates were obtained. Obtained distributions show an accuracy of predictions, reflects true features of direct time interval unpredictability of chaotic process. Changing of time series data from close to extremes allows to create new strategy of investment built on distributions basic parameters: standard deviation, skewness, kurtosis. Extension of close distribution to the pair of high-low distribution is opening extra capabilities of optimal portfolio creation and risk management for investors.