Accurate volatility forecasting in energy markets is paramount for risk management, derivative pricing, and strategic policy planning. Traditional econometric models like the Heterogeneous Auto-regressive (HAR) model effectively capture the long-memory and multi-component nature of volatility but often fail to account for non-linearities and complex, unseen dependencies. Deep learning models, particularly Long Short-Term Memory (LSTM) networks, excel at capturing these non-linear patterns but can be data-hungry and prone to overfitting, especially in noisy financial datasets. This paper proposes a novel hybrid model, HAR-LSTM-GARCH, which synergistically combines the strengths of the HAR model, an LSTM network, and a GARCH model to forecast the realized volatility of crude oil futures. The HAR component captures the persistent, multi-scale volatility dynamics, the LSTM network learns the non-linear residual patterns, and the GARCH component models the time-varying volatility of the residuals themselves. Using high-frequency data on Brent Crude futures, we compute daily Realized Volatility (RV). Our empirical results demonstrate that the proposed HAR-LSTM-GARCH model significantly outperforms the benchmark HAR, GARCH(1,1), and standalone LSTM models in both statistical accuracy and economic significance, offering a robust framework for volatility forecasting in the complex energy sector.

Accurate forecasting of systemic risk is crucial for financial stability and regulatory oversight. Traditional parametric models often fail to capture the complex, nonlinear dependence structures in financial markets, particularly during periods of distress. This paper proposes a semi-parametric dynamic conditional correlation (DCC) framework that integrates nonparametric estimation of marginal distributions with a parametric DCC specification to improve the forecasting of systemic risk measures—Conditional Value-at-Risk (CoVaR) and Marginal Expected Shortfall (MES). We employ a kernel-based quantile regression approach for marginal modeling and combine it with a DCC-GARCH structure for capturing time-varying conditional correlations. Empirical results using U.S. financial institutions’ equity returns during the 2007–2009 crisis and the post-crisis period demonstrate that the suggested semi-parametric DCC framework yields more accurate and robust forecasts of CoVaR and MES compared to fully parametric alternatives. Importantly, the SP-DCC is the only model that maintains correct unconditional and conditional coverage at the 1% VaR level during stress periods, while Gaussian-DCC significantly over-rejects. The model also exhibits superior performance in identifying systemically important institutions and tail interdependencies during stress episodes. Our findings underscore the importance of flexible marginal modeling in systemic risk measurement and provide policymakers with an enhanced toolkit for macroprudential surveillance.

Accurate forecasting of systemic risk is crucial for financial stability and regulatory oversight. Traditional parametric models often fail to capture the complex, nonlinear dependence structures in financial markets, particularly during periods of distress. This paper proposes a semi-parametric dynamic conditional correlation (DCC) framework that integrates nonparametric estimation of marginal distributions with a parametric DCC specification to improve the forecasting of systemic risk measures—Conditional Value-at-Risk (CoVaR) and Marginal Expected Shortfall (MES). We employ a kernel-based quantile regression approach for marginal modeling and combine it with a DCC-GARCH structure for capturing time-varying conditional correlations. Empirical results using daily equity returns for a panel of 20 large U.S. financial institutions over the period January 2000 to December 2023, with emphasis on three stress episodes—the 2007–2009 global financial crisis, the 2011 European sovereign debt crisis spillover, and the March 2020 COVID-19 market disruption—demonstrate that the suggested semi-parametric DCC framework yields more accurate and robust one-day-ahead forecasts of CoVaR and MES compared to fully parametric alternatives, with the advantage widening during stress periods. During the 2007–2009 crisis sub-period, the SP-DCC model reduces the Fissler–Ziegel loss for CoVaR by 19.0% relative to the Gaussian-DCC benchmark and by 6.5% relative to the strongest parametric benchmark (skewed-t DCC). Importantly, the SP-DCC specification is the only model in our comparison set for which the null hypotheses of correct unconditional and conditional coverage are not rejected at the 1% significance level during stress periods, whereas the Gaussian-DCC benchmark significantly over-rejects. The model also exhibits improved performance in identifying systemically important institutions and tail interdependencies during stress episodes, with rankings that align more closely with official G-SIB designations and SRISK-based rankings than those implied by Gaussian-DCC benchmarks. Our findings underscore the importance of flexible marginal modeling in systemic risk measurement and provide policymakers with an enhanced toolkit for macroprudential surveillance.