This repository implements an innovative approach to stock market trend forecasting by combining regime detection models with diffusion models. The methodology captures market regime shifts through Hidden Markov Models (HMM) and Switching Jump Models (SJM), then leverages these insights with diffusion models to generate accurate trend forecasts.
Stock market forecasting presents significant challenges due to the volatile, non-stationary nature of financial time series data. This project addresses these challenges through a two-stage approach:
Regime Detection: Identifying distinct market states (regimes) using probabilistic models
Diffusion-Based Forecasting: Applying advanced diffusion models to generate accurate trend predictions conditioned on detected regimes
This hybrid approach allows for more robust forecasting by explicitly modeling the regime-switching behavior of financial markets while capturing complex distributions through diffusion models
