High frequency volatility forecasting and risk assessment using neural networks-based heteroscedasticity model
The paper introduces the GaMM model, a hybrid GARCH-MLP approach, improving high-frequency volatility forecasting and risk assessment.
What it examines
The paper introduces GaMM, a novel hybrid model merging GARCH-based volatility representation with an MLP-Mixer architecture to forecast high-frequency financial volatility. It employs Bayesian optimization for tuning and compares its performance against various benchmark models to enhance risk assessment and prediction accuracy.
What it concludes
The study concludes that GaMM outperforms traditional and hybrid models in forecasting volatility and Value-at-Risk. Its efficient, lightweight design is promising for financial risk management, portfolio optimization, and market analysis, with future work suggested on interpretability enhancements and advanced layer integration.
Evidence objects
Researchers introduce GaMM, a novel hybrid model integrating GARCH representations with an MLP-based mixer, achieving superior high-frequency volatility forecasting, risk assessment, and reducing forecast errors and $VaR$ estimates across models.
key_findings bullet 1 · key_findings · validation V0
The study employs advanced hyperparameter tuning via Bayesian Optimization, incorporating time-mixing and feature-mixing techniques alongside diverse GARCH-type representations, resulting in significantly lower RMSE, MAE, QLIKE metrics, and enhanced model precision.
key_findings bullet 2 · key_findings · validation V0
Utilizing data from major U.S. equity indices during volatile years, the paper conducts rigorous tests, like the Friedman test, yet notes challenges during extreme volatility, calling for further interpretability improvements.
key_findings bullet 3 · key_findings · validation V0
This paper introduces a novel hybrid model that creatively unites traditional GARCH techniques with a multi-layer perceptron architecture $\text{MLP-Mixer}$. Its innovative stacking of MLPs captures temporal and feature dimensions, enhancing prediction accuracy and risk assessment measures. The approach offers fresh perspectives for derivative modeling and volatility forecasting with empirical impact.
key_findings bullet 4 · key_findings · validation V0
Raw abstract and provenance
- … The proposed model is evaluated on three high frequency financial times series datasets over three different years. The computational results demonstrate the proposed …
Source row: 1034 · abstract type: snippet