Woojin Chang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Woojin Chang has a total of 30 co-authors. Here are the top co-authors by number of shared publications:
- Jae Wook Song
Department of Industrial Engineering, Hanyang University, Seoul, Republic of Korea
7 shared publications Cookhwan Kim
Department of Industrial Engineering, Seoul National University, 599, Kwanak Street, Kwanak-Gu, Seoul, Republic of Korea
3 shared publications- Poongjin Cho
School of Computing, Gachon University, Seongnam 13120, Republic of Korea
3 shared publications Sungsik Park
Department of Industrial Engineering, Seoul National University, 599, Kwanak Street, Kwanak-Gu, Seoul, Republic of Korea
3 shared publicationsBonggyun Ko
Samsung Elect, Big Data Analyt Grp, Mobile Commun Business, Suwon, South Korea
2 shared publications- Jungyoon Song2 shared publications
Latest Publications
- Open Access
- 7 October 2026
- Published by Elsevier in SSRN Electronic Journal
A hybrid graph neural network for realized volatility forecasting: Multi-relation attention and dual-path fusion
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.
- 7 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessA hybrid graph neural network for realized volatility forecasting: Multi-relation attention and dual-path fusion
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.
- 16 April 2026
- Published by Elsevier in Applied Soft Computing
- Vol. 198
Maximizing the risk-adjusted return of classification model-based investments
- 16 April 2026
- Published by Elsevier in Applied Soft Computing
- Vol. 198
Maximizing the risk-adjusted return of classification model-based investments
- Open Access
- 26 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 47301-47318
An End-to-End Deep Learning Model for Clustering-Based Statistical Arbitrage
We propose an end-to-end deep learning framework for statistical arbitrage in the crypto-currency futures market, which integrates clustering and trading within a unified structure. Statistical arbitrage seeks to exploit price deviations among similar assets to generate market-neutral profits. The cryptocurrency market, characterized by structural similarities among assets and market inefficiency, offers promising opportunities for such strategies. We incorporate clustering into the model to learn optimal arbitrage trading strategies that capture dynamic relationships within the cryptocurrency market and detect arbitrage opportunities. A transformer encoder is employed to model temporal dependencies among assets. The entire arbitrage process from portfolio selection to trade allocation is jointly optimized under a customized objective function that reflects the core principles of traditional statistical arbitrage. Empirical results demonstrate that the proposed model consistently outperforms benchmark strategies with negative Sharpe ratios, achieving a cumulative return of 3.310 and an annualized Sharpe ratio of 2.516 after accounting for transaction costs. Sub-period analysis shows that the strategy remains robust across varying market conditions with annualized Sharpe ratios ranging from 1.168 to 2.240 with transaction fees, supporting its market-neutral characteristics. Ablation studies further validate the crucial contribution of each model component, such as clustering, the end-to-end structure, and the proposed objective function, in enabling effective statistical arbitrage.
- 26 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 47301-47318
Open AccessAn End-to-End Deep Learning Model for Clustering-Based Statistical Arbitrage
We propose an end-to-end deep learning framework for statistical arbitrage in the crypto-currency futures market, which integrates clustering and trading within a unified structure. Statistical arbitrage seeks to exploit price deviations among similar assets to generate market-neutral profits. The cryptocurrency market, characterized by structural similarities among assets and market inefficiency, offers promising opportunities for such strategies. We incorporate clustering into the model to learn optimal arbitrage trading strategies that capture dynamic relationships within the cryptocurrency market and detect arbitrage opportunities. A transformer encoder is employed to model temporal dependencies among assets. The entire arbitrage process from portfolio selection to trade allocation is jointly optimized under a customized objective function that reflects the core principles of traditional statistical arbitrage. Empirical results demonstrate that the proposed model consistently outperforms benchmark strategies with negative Sharpe ratios, achieving a cumulative return of 3.310 and an annualized Sharpe ratio of 2.516 after accounting for transaction costs. Sub-period analysis shows that the strategy remains robust across varying market conditions with annualized Sharpe ratios ranging from 1.168 to 2.240 with transaction fees, supporting its market-neutral characteristics. Ablation studies further validate the crucial contribution of each model component, such as clustering, the end-to-end structure, and the proposed objective function, in enabling effective statistical arbitrage.
- Open Access
- 20 March 2026
- Published by Elsevier in SSRN Electronic Journal
Reward Inference for Portfolio Optimization via Multi-Expert Inverse Reinforcement Learning
Portfolio optimization requires navigating complex trade-offs among conflicting financial objectives, including return maximization, risk mitigation, and drawdown control. Reinforcement Learning (RL) provides a dynamic framework for asset allocation, but manually specifying reward functions that generalize across diverse market regimes remains a significant challenge. Inverse Reinforcement Learning (IRL) offers a principled alternative by recovering reward signals from expert behavior. However, when demonstrations originate from multiple experts with divergent investment objectives, the collective expert distribution becomes inherently multi-modal. Multiple distinct actions may be simultaneously rational for a given market state, and conventional IRL frameworks are not designed to accommodate this complexity. We propose Multi-Expert Inverse Reinforcement Learning (ME-IRL), a framework that employs an Energy-Based Model (EBM) as a reward network to model this multi-modal collective expert distribution. Rather than specifying objectives a priori or aggregating expert behaviors at the action-level, ME-IRL performs reward-level inference that encodes interactions among conflicting financial objectives. This is achieved through a Transformer-based architecture utilizing cross-attention between state and action representations to explicitly model cross-sectional asset dependencies. Empirical evaluations across ETF and stock portfolios consistently demonstrate that ME-IRL out-performs traditional financial models and single- and multi-objective RL baselines in both return and risk-adjusted metrics across diverse market conditions.
- 20 March 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessReward Inference for Portfolio Optimization via Multi-Expert Inverse Reinforcement Learning
Portfolio optimization requires navigating complex trade-offs among conflicting financial objectives, including return maximization, risk mitigation, and drawdown control. Reinforcement Learning (RL) provides a dynamic framework for asset allocation, but manually specifying reward functions that generalize across diverse market regimes remains a significant challenge. Inverse Reinforcement Learning (IRL) offers a principled alternative by recovering reward signals from expert behavior. However, when demonstrations originate from multiple experts with divergent investment objectives, the collective expert distribution becomes inherently multi-modal. Multiple distinct actions may be simultaneously rational for a given market state, and conventional IRL frameworks are not designed to accommodate this complexity. We propose Multi-Expert Inverse Reinforcement Learning (ME-IRL), a framework that employs an Energy-Based Model (EBM) as a reward network to model this multi-modal collective expert distribution. Rather than specifying objectives a priori or aggregating expert behaviors at the action-level, ME-IRL performs reward-level inference that encodes interactions among conflicting financial objectives. This is achieved through a Transformer-based architecture utilizing cross-attention between state and action representations to explicitly model cross-sectional asset dependencies. Empirical evaluations across ETF and stock portfolios consistently demonstrate that ME-IRL out-performs traditional financial models and single- and multi-objective RL baselines in both return and risk-adjusted metrics across diverse market conditions.
- 30 March 2025
- Published by Elsevier in Applied Soft Computing
- Vol. 175
Forecasting realized volatility using deep learning quantile function
- 30 March 2025
- Published by Elsevier in Applied Soft Computing
- Vol. 175
Forecasting realized volatility using deep learning quantile function
- 18 December 2024
- Published by Springer Nature in Applied Intelligence
- Vol. 55 (3) , 1-31
NQF-RNN: probabilistic forecasting via neural quantile function-based recurrent neural networks
- 18 December 2024
- Published by Springer Nature in Applied Intelligence
- Vol. 55 (3) , 1-31
NQF-RNN: probabilistic forecasting via neural quantile function-based recurrent neural networks
- 18 April 2022
- Published by SAGE Publications in Intelligent Data Analysis: An International Journal
- Vol. 26 (3) , 723-750
Stock market network based on bi-dimensional histogram and autoencoder
In this study, we propose a deep learning related framework to analyze S&P500 stocks using bi-dimensional histogram and autoencoder. The bi-dimensional histogram consisting of daily returns of stock price and stock trading volume is plotted for each stock. Autoencoder is applied to the bi-dimensional histogram to reduce data dimension and extract meaningful features of a stock. The histogram distance matrix for stocks are made of the extracted features of stocks, and stock market network is built by applying Planar Maximally Filtered Graph(PMFG) algorithm to the histogram distance matrix. The constructed stock market network represents the latent space of bi-dimensional histogram, and network analysis is performed to investigate the structural properties of the stock market. we discover that the structural properties of stock market network are related to the dispersion of bi-dimensional histogram. Also, we confirm that the autoencoder is effective in extracting the latent feature of the bi-dimensional histogram. Portfolios using the features of bi-dimensional histogram network are constructed and their investment performance is evaluated in comparison with other benchmark portfolios. We observe that the portfolio consisting of stocks corresponding to the peripheral nodes of bi-dimensional histogram network shows better investment performance than other benchmark stock portfolios.
- 18 April 2022
- Published by SAGE Publications in Intelligent Data Analysis: An International Journal
- Vol. 26 (3) , 723-750
Stock market network based on bi-dimensional histogram and autoencoder
In this study, we propose a deep learning related framework to analyze S&P500 stocks using bi-dimensional histogram and autoencoder. The bi-dimensional histogram consisting of daily returns of stock price and stock trading volume is plotted for each stock. Autoencoder is applied to the bi-dimensional histogram to reduce data dimension and extract meaningful features of a stock. The histogram distance matrix for stocks are made of the extracted features of stocks, and stock market network is built by applying Planar Maximally Filtered Graph(PMFG) algorithm to the histogram distance matrix. The constructed stock market network represents the latent space of bi-dimensional histogram, and network analysis is performed to investigate the structural properties of the stock market. we discover that the structural properties of stock market network are related to the dispersion of bi-dimensional histogram. Also, we confirm that the autoencoder is effective in extracting the latent feature of the bi-dimensional histogram. Portfolios using the features of bi-dimensional histogram network are constructed and their investment performance is evaluated in comparison with other benchmark portfolios. We observe that the portfolio consisting of stocks corresponding to the peripheral nodes of bi-dimensional histogram network shows better investment performance than other benchmark stock portfolios.
- 1 August 2020
- Published by Elsevier in Chaos, Solitons & Fractals
- Vol. 137
Fractal structure in the S&P500: A correlation-based threshold network approach
- 1 August 2020
- Published by Elsevier in Chaos, Solitons & Fractals
- Vol. 137
Fractal structure in the S&P500: A correlation-based threshold network approach
- 23 October 2019
- Published by Springer Nature in Pattern Analysis and Applications
- Vol. 23 (3) , 1183-1202
Instance-based entropy fuzzy support vector machine for imbalanced data
- 23 October 2019
- Published by Springer Nature in Pattern Analysis and Applications
- Vol. 23 (3) , 1183-1202
Instance-based entropy fuzzy support vector machine for imbalanced data
- 4 April 2019
- Published by Elsevier in Physica A: Statistical Mechanics and its Applications
- Vol. 525, 921-934
Explaining future market return and evaluating market condition with common preferred spread index
- 4 April 2019
- Published by Elsevier in Physica A: Statistical Mechanics and its Applications
- Vol. 525, 921-934
Explaining future market return and evaluating market condition with common preferred spread index