Hybrid LSTM–HMM framework for regime-aware cryptocurrency price forecasting

Abstract

The rapid evolution of digital finance and blockchain technology has positioned cryptocurrencies as significant yet highly volatile financial assets. Their nonlinear and regime-shifting price behavior challenges traditional econometric and deep learning models, creating a need for more adaptive forecasting frameworks. Addressing this research gap, the present study proposes a hybrid Hidden Markov Model–Long Short-Term Memory (HMM–LSTM) architecture that jointly captures latent market regimes and temporal dependencies in cryptocurrency price dynamics. The HMM identifies latent market regimes, while the LSTM captures temporal dependencies. Technical indicators, including Bollinger Bands, EMA, and MACD, support feature benchmarking. Empirical results show that hybrid gains are asset-dependent and feature-dependent: performance is strongest for Ethereum and BNB, comparable for SOL, and weaker than the standalone LSTM for Bitcoin, indicating that regime information is conditionally beneficial for financial forecasting. Key findings highlight that the hybrid design effectively mitigates overfitting, improves generalization across assets, and captures regime-dependent price movements and robust regime-sensitive forecasting. The implications of this research extend to AI-driven financial forecasting, trading automation, and contributing to the advancement of intelligent, data-centric financial analytics.