JSJungyoon Song

Jungyoon Song

*This is an algorithmically generated author record based on publicly available information.

Jungyoon Song has a total of 5 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    A hybrid graph neural network for realized volatility forecasting: Multi-relation attention and dual-path fusion

    by Jongu Lee, Woojin Chang, Jungyoon Song

    Accurate realized volatility (RV) forecasting is important for risk management and portfolio allocation, yet existing models typically capture either temporal dynamics within individual assets or a single form of cross-asset dependence. This study proposes HyGNN-RV, a hybrid graph neural network that jointly learns nonlinear temporal patterns and cross-asset volatility interdependencies. Daily, weekly, and monthly RV and the jump component each define a separate asset-relation graph, and relation-level attention learns how much each relation contributes. A parallel graph convolutional path preserves the current volatility level, which attention-based aggregation tends to smooth away, and the two paths are combined through an RV-conditioned gate. Sliding-window experiments on Dow Jones Industrial Average constituents from July 2014 to June 2025 compare HyGNN-RV with econometric, deep-learning, and graph-based benchmarks. The proposed model attains the lowest mean squared error(MSE) and quasi-likelihood loss(QLIKE), reducing the losses of the second-best model by approximately 9.9% and 3.4%, and ranks first in all six windows under QLIKE, including the COVID-19 and tariff-shock periods. The two graph paths prove complementary: removing either degrades a different error metric. Out-of-sample relation weights further show that the model shifts toward the daily relation as market volatility rises and selectively toward the jump relation during discontinuous shocks.

  • Open Access

    DeepHAR: Heterogeneous Attention-Conditioned Regression Network for Realized Volatility Forecasting

    by Jungyoon Song, Jae Wook Song

    Accurate volatility forecasting is essential for risk management and portfolio allocation; yet existing methods face key limitations. Linear Heterogeneous Autoregressive (HAR)-type econometric models lack adaptability to regime shifts, while general-purpose deep learning models fail to capture volatility-specific structures, often underperforming compared to simple linear benchmarks. We propose DeepHAR, a heterogeneous attention-based regression framework that embeds the Heterogeneous Market Hypothesis (HMH) into a neural architecture for realized volatility forecasting. DeepHAR constructs horizon-specific representations via temporal attention pooling and models asymmetric cross-horizon dependencies through heterogeneous cross-attention, conditioning short-term volatility shocks on mid- and long-term market memory. A dual-path decomposition further separates this representation into trend and shock components, capturing volatility persistence and mean reversion. Experiments on major U.S. equity index ETFs over 2005–2025 show that DeepHAR consistently outperforms strong econometric and deep learning benchmarks, with statistically significant error reductions confirmed by Diebold–Mariano tests. DeepHAR also maintains robust performance under extreme stress, including the COVID-19 crash and the 2025 Tariff Shock. Attention-based analyses reveal economically meaningful, regime-dependent reallocation of cross-horizon focus consistent with the HMH, while utility-based evaluations confirm that forecasting gains translate into tangible portfolio benefits. These findings indicate that embedding domain-specific market structure into attention mechanisms offers a more effective path toward robust, interpretable volatility forecasting than purely linear or purely generic deep learning approaches.