Efficiency of genetic algorithm as forecasting instrument of time series

The genetic algorithm is numbered among less formal methods which allows using it in different areas including forecasting. However, there is a question about efficiency of that instrument. We decided to check inner procedures of the algorithm affecting gaining speed and quality of solutions. We also tested exactness of expired forecasts. All the calculations were made using time series from the Stock Exchange. Comparing all chromosomes from the initial and final populations leads to a conclusion that preselection contributed to the unquestionable improvement of forecasts. It is possible because of a bigger diversity between chromosomes. RMSPE for the best chromosomes rapidly falls during the first twenty of all fifty generations. After that chromosomes with the best fitness dominate the whole population and the improvement of results is not so fast. The expired forecasts of close price turned out much better than volume although we observed deterioration for all stock instruments. The reason was big volatility of those volumes which influenced the efficiency of the algorithm.