Analysis of Enhanced Hidden Markov Models for Improved Stock Market Price Forecasting and Prediction

In the ever-evolving landscape of financial markets, the pursuit of accurate stock price predictions remains a formidable challenge. This study addresses the challenge of profitable stock market predictions by exploring modified HMM approaches involving fixed parameterisation and hyper-heuristic methods while also considering sequence lengths and adaptability to heuristic applications. This study extends its focus to the intricacies of determining optimal buy and sell times. Recognising the nonstationary nature of financial time series, the research explores threshold autoregressive models and mixture models for time series analysis. The results of this research indicate that the K-Fold method consistently exhibits strong performance in terms of fitting and robustness, with accuracy percentages exceeding 90% for certain stocks. The Sliding Window method proves effective for short-term forecasting but falls short in longer time horizons. Continuous HMM demonstrates impressive fitting capabilities but is susceptible to overfitting. The Hybrid HMM (with ARIMA) method offers above-average and relatively consistent results, although it may require customisation for specific scenarios.