Enhancing intraday trading performance of Neural Network using dynamic volatility clustering fuzzy filter

We extend Neural Network (NN) trading models with an innovative and efficient volatility filter based on fuzzy c-means clustering algorithm, where the choice for the number of clusters, a frequent problem with cluster analysis, is selected by optimizing a global risk-return performance measure. Our algorithm automatically extracts fuzzy rules from past trades by taking into account the predicted return size and intraday time varying realized volatility, the latter used as a proxy for uncertainty. The model identifies unique intraday scenarios and subsequently creates a dynamic and visually apprehensible risk-return search space to control algorithmic trading decisions. Our results show that this method can be successfully applied to support high-frequency intraday trading strategies, outperforming both standard NN and buy-and-hold models.