Improving risk-adjusted performance in high-frequency trading: The role of fuzzy logic systems

In recent years, algorithmic and high-frequency trading have been the subject of increasing risk concerns. A general theme that we adopt in this thesis is that trading practitioners are predominantly interested in risk-adjusted performance. Likewise, regulators are demanding stricter risk controls. First, we scrutinise conventional AI model design approaches with the aim to increase the risk-adjusted trading performance of the proposed fuzzy logic models. We show that applying risk-return objective functions and accounting for transaction costs improve out-of-sample results. Our experiments identify that neuro-fuzzy models exhibit superior performance stability across multiple risk regimes when compared to popular neural network models identified in AI literature. Moreover, we propose an innovative ensemble model approach which combines multiple risk-adjusted objective functions and dynamically adapts risk- tolerance according to time-varying risk. Next, we extend our findings to the money management aspects of trading algorithms. We introduce an effective fuzzy logic approach which dynamically discriminates across different regions in the trend and volatility space. The model prioritises higher performing regions at an intraday level and adapts capital allocation policies with the objective to maximise global risk-adjusted performance. Finally, we explore trading improvements that can be attained by advancing our type-1 fuzzy logic ideas to higher order fuzzy systems in view of the increased noise (uncertainty) that is inherent in high-frequency data. We propose an innovative approach to design type-2 models with minimal increase in design and computational complexity. As a further step, we identify a relationship between the increased trading performance benefits of the proposed type-2 model and higher levels of trading frequencies. In conclusion, this thesis sets a framework for practitioners, researchers and regulators in the design of fuzzy logic systems for better management of risk in the field of algorithmic and high-frequency trading.