MSc Defence: Mathematical Models for Algorithmic Trading: Evaluating HMM and Neural Network-Based Strategies

Candidate: Areebah Muhammad

Supervisor: Dr. You Liang

Abstract
Pairs trading and multiple trading strategies are market-neutral approaches that utilize the co-integration or co-movements of stocks and other financial instruments to generate potential profits that are independent of overall market direction. In this paper, we introduce three novel resilient trading strategies. The first combines the Kalman Filter (KF) algorithm with KF innovation volatility interval forecasts (KFI) using neural networks, including KF innovation volatility point forecasts (KFP) as a special case. The second employs the hidden Markov model combined with the data-drive innovation volatility forecasting (DDIVF) approach introduced in [15] (HMM-DDIVF), compared to the KF algorithm using DDIVF only (KF-DDIVF). Our strategies are implemented and tested on hourly prices of Bitcoin, Ethereum, and Bitcoin Cash during a bear market. These cryptocurrencies are chosen for their long-term co-movement and high trading volumes. Experimental results reveal that both KFI and HMM-DDIVF strategies consistently outperform their respective benchmarks KFP and KF-DDIVF. These trading strategies demonstrates enhanced robustness and higher profitability with and without transaction costs, thus provides robust and risk-averse trading solutions.