Extraction of investment strategies based on moving averages: A genetic algorithm approach

Investment strategies as rules for buy and sell are introduced as conditional statements involving inequalities of various moving averages. Different conditional statements on moving averages are represented as strings, encodable as chromosomes in an approach based on genetic algorithm. The data mining of good investment strategies corresponds to the extraction of rules that are fit in the sense of evolutionary computation. Three different kinds of moving averages (simple moving average, exponential moving average and adaptive moving average) of various durations have been used in the conditional statement involving the closing price of a given stock. The performance of a given investment strategy is evaluated using the rate of overall return in both the training set and the test set, thereby converting the problem of discovering good investment strategies to an optimization problem in combinatorics, which is solved with a genetic algorithm approach. Stock data from NASDAQ, including Microsoft, Intel, Oracle, and Dell are tested and comparisons of genetic algorithms with benchmark methods such as random walk, buy and hold, and exhaustive search are performed. Results show evidence of superior performance in genetic algorithm in term of the rate of overall return for the test set. Within the confine of limited data, all three moving averages show similar results for the four technology stock investigated.