Multi-dimensional pattern discovery in financial time series using sax-ga with extended robustness

Abstract

This paper proposes a new Multi-Dimensional SAX-GA approach to pattern discovery using genetic algorithms (GA). The approach is capable of discovering patterns in multi-dimensional financial time series. First, the several dimensions of data are converted to a Symbolic Aggregate approXimation (SAX) representation, which is, then, feed to a GA optimization kernel. The GA searches for profitable patterns occurring simultaneously in the multi-dimensional time series. Based on the patterns found, the GA produces more robust investment strategies, since the simultaneity of patterns on different dimensions of the data, reinforces the strength of the trading decisions implemented. The proposed approach was tested using stocks from S&P500 index, and is compared to previous reference works of SAX-GA and to the Buy & Hold (B&H) classic investment strategy.