Predicting mutual fund performance using artificial neural networks

This study utilizes an artificial neural network (ANN) approach to predict the performance of equity mutual funds that follow value, blend and growth investment styles. Using a multi-layer perceptron model and GRG2 nonlinear optimizer, fund-specific historical operating characteristics were used to forecast mutual funds' risk-adjusted return. Results show that ANN generates better forecasting results than linear models for funds of all styles. In addition, our model outperforms that of Chiang et al. [Chiang WC, Urban TL, Baldridge GW. A neural network approach to mutual fund net asset value forecasting. Omega Int J Manage Sci 1996:24;205-215.] in predicting the performance of growth funds. We also employed a heuristic approach of variable selection via neural networks and compared it with the stepwise selection method of linear regression. Results are encouraging in that the reduced ANN models still outperform the linear models for growth and blend funds and yield similar results for value funds.