Applications of combined financial strategies based on universal adaptive forecasting

Abstract

We consider an online adaptive forecasting algorithm for time series elements. Based on this algorithm, we define a universal strategy for the financial market: such a strategy ensures asymptotically maximal profit compared to any trading strategy where decisions are made based on rules that depend continuously on the input information. To reduce risk, in simultaneous trading of several financial instruments we perform adaptive redistribution of the current capital among them according to the AdaHedge algorithm. We propose variations of a combined game with various algorithmic trading strategies. We give results of numerical experiments based on historical data of the MICEX and BATS (US) trading platforms.