Genetic algorithms in stock forecasting – deleting outliers

This work presents a proposal of usage of genetic algorithm to short-term forecasting of price and volume quotations. Presented algorithm resembles the naive method with seasonality but a lag of observation used as predictor can change in order to achieve best adjustment of ex post prognosis to data. The data were devoid of outliers with the help of hat matrix, taken from robust estimation. The results confirmed the earlier assumptions and gave better ex post forecasts after removing outliers. Much better results were obtained for the prices, compared to those obtained for the volume, due to smaller in the case of prices, caused by smaller random fluctuations.