hmm_model_regime_change

Regime Detection in Bitcoin Using Hidden Markov Models (HMM) and Alpha-Stable Distributions

Overview

This repository demonstrates how to detect regime changes in financial markets, specifically Bitcoin, using Hidden Markov Models (HMM) with non-Gaussian Alpha-Stable distributions. The project uses the pomegranate library, which supports non-Gaussian distributions like Alpha-Stable, making it ideal for modeling Bitcoin’s highly volatile and non-normal return distribution.

Key Concepts:

• Hidden Markov Models (HMM): A statistical model where the system being modeled is assumed to follow a Markov process with hidden states.
• Alpha-Stable Distributions: Flexible distributions capable of modeling fat tails and asymmetry in financial data, unlike Gaussian distributions.
• Regime Detection: The identification of different market conditions (e.g., bull, bear, and range markets) and their transitions over time.

Features

• Simulated Bitcoin Data: Simulates market conditions with bull, bear, and range markets, each with varying volatility levels.
• Alpha-Stable HMM: Uses Hidden Markov Models with Alpha-Stable distributions to model the returns.
• Regime Prediction: Detects market regimes based on the modeled data.
• Visualization: Visualizes the identified regimes on the time series data.
• CSV Export: Exports the time series and corresponding regimes to a CSV file for further analysis.

Installation

To run this project locally, follow these steps:

1. Clone the repository:

   git clone https://github.com/your-username/regime-detection-bitcoin-hmm.git

2. Navigate to the project directory:

   cd regime-detection-bitcoin-hmm

3. Install the required dependencies: This project uses the following libraries:

   • pomegranate for HMM modeling
   • numpy, pandas for data manipulation
   • matplotlib for visualization

   Install them using the following command:

   pip install -r requirements.txt

Usage

Step 1: Simulate Data

Run the notebook hmm_regime_detection.ipynb, which simulates Bitcoin-like returns for 6 different market regimes (bull/bear/range with low/high volatility).

Step 2: Fit Hidden Markov Model

The notebook fits a Hidden Markov Model using pomegranate with Alpha-Stable distributions, making it ideal for modeling Bitcoin’s returns. It detects the underlying regimes in the simulated time series data.

Step 3: Predict and Visualize Regimes

The predicted regimes are visualized alongside the simulated returns, with each regime (market condition) highlighted using different colors.

Step 4: Export Results

The results (time series with predicted regimes) are exported as a CSV file for further analysis.

Example Notebook:

The core logic is implemented in the Jupyter notebook hmm_regime_detection.ipynb. Here’s a brief breakdown:

• Simulates market regimes using numpy.
• Trains an HMM with pomegranate using Alpha-Stable distributions.
• Predicts the hidden states (regimes) and visualizes them using matplotlib.
• Exports the results to regimes_prediction.csv.

File Structure