Advances in forecasting realized volatility: a review of methodologies
A major survey of 41 studies finds deep learning models, especially hybrids like CNN-LSTM and Transformers, now surpass traditional methods in forecasting realized volatility. The review compares 32 models, introducing new hybrids such as HARNet, which combines classic and neural network approaches. Deep learning models excel in turbulent markets and capture all six key volatility features. However, their high computational cost and data needs remain challenges, while linear models still offer simplicity and interpretability with limited data.
What it examines
This survey reviews all major models used to forecast realized volatility in finance from 2000 to mid-2024. It compares traditional linear models, machine learning, and deep learning methods, aiming to identify which models work best and what features of volatility they capture.
What it concludes
Deep learning models, especially CNNs and Transformers, provide the most accurate volatility forecasts by capturing complex patterns. These models are useful for risk management, investment strategies, and pricing derivatives. Future research should focus on hybrid models and improving computational efficiency for broader practical use.
Evidence objects
A sweeping survey of 41 studies finds deep learning modelsespecially hybrids like CNN-LSTM and Transformersnow consistently outperform traditional linear and machine learning methods in forecasting realized volatility with greater accuracy.
key_findings bullet 1 · key_findings · validation V0
Notably, new hybrid models such as HARNet, which merges the classic HAR model with convolutional neural networks, show empirical superiority in capturing complex market dynamics and all six key volatility features, including autocorrelation and co-movement.
key_findings bullet 2 · key_findings · validation V0
Despite their accuracy, deep learning models face high computational costs and data demands; linear models remain valuable for their simplicity and interpretability, especially with limited data, highlighting the need for scalable, hybrid approaches.
key_findings bullet 3 · key_findings · validation V0
This paper uniquely surveys realized volatility forecasting, systematically comparing traditional linear models and advanced machine/deep learning methods (CNNs, LSTMs, HARNets, transformers). Its novelty lies in breadth and empirical analysis, not new models. Compelling for quantitative finance, it identifies research gaps, guiding practitioners and researchers in model selection and future work.
key_findings bullet 4 · key_findings · validation V0
Raw abstract and provenance
- … processing, and financial modeling have transformed the financial industry from a … in the derivatives market, it is critical for financial institutions (eg, hedge funds, trading …
Source row: 117 · abstract type: snippet