Stock Market Technical Indicator Optimization by Genetic Algorithms

Abstract

Technical indicators are widely used in stock market forecasting, mostly to trigger the buy/sell rules in the technical analysis. Through some statistical analysis some key values for several indicator parameters are obtained. These values are generally adjusted to provide simple, round numbers, so they become part of easy-to-remember rules, such as 70-30 RSI rule, Crossover 50MA, etc. However, since these selections of indicator values are used as rule-of-thumb buy-sell triggers, it is not clear how changing market conditions affect them. For example, one indicator might provide good results for a particular stock in an uptrend market, but might fail miserably during downtrend. In this study, the performances of several different Exchange Traded Funds (ETFs) are analyzed using different technical indicators between the years 1993-2008. The indicator parameters are optimized against portfolio performance using genetic algorithms. Different analyses are implemented in different market conditions (uptrend or downtrend), using a basket of ETFs and different technical indicators. The trained indicators were tested between the years 2008-2010. The results indicate that even though the test performance is not as high as the training performance, the results are generally acceptable. Also, surprisingly, for several ETFs, the widely-used indicators, a lot of times, perform poorly indicating even though they are well-known and widely- implemented strategies; they should not be used blindly for any ETF or stock.