High-low Strategy of Portfolio Composition using Evolino RNN Ensembles

Abstract

Strategy of investment is important tool enabling better investor’s decisions in uncertain finance market. Rules of portfolio selection help investors balance accepting some risk for the expectation of higher returns. The aim of the research is to propose strategy of constructing investment portfolios based on the composition of distributions obtained by using high–low data. The ensemble of 176 Evolino recurrent neural networks (RNN) trained in parallel investigated as an artificial intelligence solution, which applied in forecasting of financial markets. Predictions made by this tool twice a day with different historical data give two distributions of expected values, which reflect future dynamic exchange rates. Constructing the portfolio, according to the shape, parameters of distribution and the current value of the exchange rate allows the optimization of trading in daily exchange-rate fluctuations. Comparison of a high-low portfolio with a close-to-close portfolio shows the efficiency of the new forecasting tool and new proposed trading strategy.