Data transformation methods for genetic-algorithm-based investment decisions

In an earlier work, we examined the performance of genetic algorithms as a method for determining a strategy to invest in different financial instruments every month (S.K. Kassicieh et al., 1997). The inputs in the earlier work were differenced time series of 10 economic indicators where the genetic algorithm used the best three of these series to make the timing (or equivalently switching) decision. We use the same genetic algorithm with different data transformation methods applied to economic data series. These methods are the singular value decomposition (SVD) and principal component artificial neural network (PCANN) with 3, 4, 5 and 10 nodes. We report the result of a large number of runs to determine which of these methods works best. We find that the non standardized SVD of economic data yields the highest terminal wealth for the time period examined. The terminal accumulation is 78.75% of the dollar accumulation given by a perfect timing strategy