Evaluating direction-of-change forecasting: neurofuzzy models vs. neural networks

This paper investigates the nonlinear predictability of technical trading rules based on a recurrent neural network as well as a neurofuzzy model. The efficiency of the trading strategies was considered upon the prediction of the direction of the market in case of NASDAQ and NIKKEI returns. The sample extends over the period 2/8/1971–4/7/1998 while the sub-period 4/8/1998–2/5/2002 has been reserved for out-of-sample testing purposes. Our results suggest that, in absence of trading costs, the return of the proposed neurofuzzy model is consistently superior to that of the recurrent neural model as well as of the buy & hold strategy for bear markets. On the other hand, we found that the buy & hold strategy produces in general higher returns than neurofuzzy models or neural networks for bull periods. The proposed neurofuzzy model which outperforms the neural network predictor allows investors to earn significantly higher returns in bear markets.