A Deep Learning Framework for Medium-Term Covariance Forecasting in Multi-Asset Portfolios
For institutional investors and risk managers, the framework offers improved medium-term covariance forecasts with practical benefits in portfolio allocation and risk management.
The paper proposes a deep learning framework combining 3D CNNs, BiLSTMs, and multi-head attention for medium-term covariance forecasting, achieving up to 20% reduction in distance metrics over classical methods on 14 ETFs from 2017-2023, with robust performance across market regimes and economic value in portfolio experiments.
Accurate covariance forecasting is central to portfolio allocation, risk management, and asset pricing, yet many existing methods struggle at medium-term horizons, where shifting market regimes and slower dynamics predominate. We propose a deep learning framework that combines three-dimensional convolutional neural networks, bidirectional long short-term memory layers, and multi-head attention to capture complex spatio-temporal dependencies. Using daily data on 14 exchange-traded funds from 2017 through 2023, we find that our model reduces Euclidean and Frobenius distance metrics by up to 20\% relative to classical benchmarks (e.g., shrinkage and GARCH approaches) and remains robust across distinct market regimes. Our portfolio experiments demonstrate significant economic value through lower volatility and moderate turnover. These findings highlight the potential of advanced deep learning architectures to improve medium-term covariance forecasts, offering practical benefits for institutional investors and risk managers.