Hybrid HMM\u2013Gradient Boosting Signals for Short-Horizon Equity Returns and Prices

We address the trade-off between interpretability and accuracy in return forecasting by combining a regime detector with a flexible predictor. First, a Gaussian hidden Markov model (HMM) infers daily market regimes (e.g., bull, bear, high-volatility) from returns and volatility, providing an interpretable state indicator. Second, these regime labels, together with technical features, feed a gradient-boosting regressor to forecast next-dayreturns with nonlinear interactions. We use non-cumulative price evaluation and benchmark against Random Walk, ARIMA, and RBF-SVR. On daily data for AAPL, JPM, GS, and MSFT (2022\u20132025), the hybrid model generally improves directional accuracy and achieves competitive error metrics. Diagnostics (bootstrap tests for regime count; Ljung\u2013Box, ADF\/KPSS, Durbin\u2013Watson) and sensitivity checks support robustness, with two to three regimes preferred. The approach yields regime-aware, decision-relevant signals for short-horizon trading and risk management and is extensible to macro factors and multi-asset settings.