Trading Foreign Currencies Based on Long Range Dependence

The paper discusses a novel trading algorithm for trading foreign currencies. The algorithm is based on the fact that currency exchange rates exhibit long-range dependency i.e. time-series constructed of exchange rates deviates from complete randomness and contains memory. This memory can be useful in predicting future behaviors of the exchange rates. The type of correlation varies in time from positive to negative and vice versa, consequently it is important to accurately estimate correlation type. In long range dependent processes the strength and type of correlation is judge by value of so called scale exponent. Combing trend information and the estimated type of correlation we predict exchange rate behavior and thus we are able to determine best positions to sell and/or to buy. We tested the proposed trading algorithm on 8 currency pairs with the history ranging from 1.0x106 to 2.5x106 minutes where the minimum profit was 25430 pips for EURGBP pair and the maximum profit was 192070 pips for AUDCAD.