Examination of Variables Affecting Dividend Forecast Using Hybrid Models of PSO-LARS and PSO-SVR Algorithms

Since one of the most important sources of information for investors and other beneficial is dividends forecast, this study tries to find models for predicting variables effective on dividend. To do this, information from chemical companies listed in Tehran Stock Exchange during the years 2006 to 2010 are used. The independent variables are accounting ratios and the dependent variable is dividend. The model framework is a combination of PSO-SVR and PSO-LARS algorithms. PSO algorithm identifies optimal combination of variables that influence the anticipated dividends. Then the data related to the variables selected by PSO are entered in to the SVR and LARS algorithms separately and train the algorithms. Then the algorithms are tested with evaluation data. Thus the prediction errors can be measured and the methods be compared. The research results show that combining PSO algorithm with LARS or SVR algorithm, as compared to using only SVR and LARS, can provide a better predict of considered affecting factors. Comparing the two combination methods, PSO-LARS and PSO-SVR, PSO-SVR shows that prediction error is less