A rich documentation is available in the GitHub repository.

volatility-gamm

GARCH-based MLP Mixer for high-frequency volatility forecasting and risk assessment.

volatility-gamm is a Python package for volatility forecasting using GARCH-based MLP Mixer models.

The package combines GARCH-family volatility information with MLP-Mixer style neural feature learning.

This package provides two primary components:

• GARCH-based volatility features: Utilities for using GARCH, GJR-GARCH, Threshold GARCH, AVGARCH, and FIGARCH conditional-volatility estimates as model inputs.
• GaMMixer neural model: A configurable MLP-Mixer style architecture for volatility forecasting.

Key Features

• GARCH-based MLP Mixer: Combines GARCH-family volatility features with neural MLP-Mixer modeling.
• High-Frequency Volatility Forecasting: Designed for intraday financial volatility forecasting tasks.
• Econometric Feature Integration: Supports GARCH-family conditional-volatility features.
• Neural Forecasting Model: Provides Keras-based GaMMixer model construction and compilation utilities.
• Hyperparameter Tuning: Includes Hyperopt/TPE-based tuning helpers.
• Risk Assessment: Includes Value-at-Risk utilities.
• Forecasting Metrics: Provides RMSE, MAE, MAPE, NMAE, and QLIKE.
• Research-Oriented Design: Suitable for volatility forecasting and financial risk modeling.

Installation

Downloading locally and installing:

git clone https://github.com/statsdl/GaMM.git
cd GaMM

Install dependencies:

pip install -r requirements.txt

Install the package:

pip install -e .

Using pip install from GitHub:

pip install git+https://github.com/statsdl/GaMM.git

Using pip install from PyPI:

pip install volatility-gamm

Optional full installation:

pip install "volatility-gamm[full]"

Development installation:

pip install -e ".[dev]"

Usage

1. GaMMixer model

import numpy as np
from gamm.model import compile_gamm_model

X_train = np.random.randn(200, 20, 6).astype("float32")
y_train = np.random.rand(200).astype("float32")

model = compile_gamm_model(
    input_shape=(X_train.shape[1], X_train.shape[2]),
    learning_rate=0.001,
    activation="relu",
    n_block=2,
    dropout=0.1,
    ff_dim=64,
)

model.fit(X_train, y_train, epochs=5, batch_size=32, verbose=0)

predictions = model.predict(X_train[:10], verbose=0)
print("Volatility forecasts:", predictions.reshape(-1))

2. Volatility forecasting metrics

import numpy as np
from gamm.metrics import (
    root_mean_squared_error,
    mean_absolute_error,
    mean_absolute_percentage_error,
    normalized_mean_absolute_error,
    quasi_likelihood,
    value_at_risk,
)

y_true = np.array([0.12, 0.15, 0.09, 0.18])
y_pred = np.array([0.11, 0.14, 0.10, 0.17])

print("RMSE:", root_mean_squared_error(y_true, y_pred))
print("MAE:", mean_absolute_error(y_true, y_pred))
print("MAPE:", mean_absolute_percentage_error(y_true, y_pred))
print("NMAE:", normalized_mean_absolute_error(y_true, y_pred))
print("QLIKE:", quasi_likelihood(y_true, y_pred))
print("VaR:", value_at_risk(0.05, y_pred))

3. Synthetic benchmark

python examples/run_synthetic_benchmark.py

API Reference

Model utilities

• build_gamm_model: Builds a GARCH-based MLP Mixer model.
• compile_gamm_model: Builds and compiles the GaMMixer model with Adam optimizer and MSE loss.

Tuning utilities

• gamm_tuning_space: Returns the default Hyperopt/TPE search space.
• tune_gamm: Tunes GaMMixer hyperparameters.

GARCH utilities

• append_garch_features: Adds GARCH-family conditional-volatility features.

Metrics

• root_mean_squared_error: Computes RMSE.
• mean_absolute_error: Computes MAE.
• mean_absolute_percentage_error: Computes MAPE.
• normalized_mean_absolute_error: Computes range-normalized MAE.
• quasi_likelihood: Computes QLIKE.
• value_at_risk: Computes normal Value-at-Risk from volatility forecasts.

Dataset Details

License

This project is licensed under the MIT License. See the LICENSE file for details.

Citation

If you use this package in your research, please cite:

@article{bhambu2025highfrequency,
  title={High frequency volatility forecasting and risk assessment using neural networks-based heteroscedasticity model},
  author={Bhambu, Aryan and Bera, Koushik and Natarajan, Selvaraju and Suganthan, Ponnuthurai Nagaratnam},
  journal={Engineering Applications of Artificial Intelligence},
  volume={149},
  pages={110397},
  year={2025},
  doi={10.1016/j.engappai.2025.110397}
}