Nak Hyun Jung
Research Subjects
Nak Hyun Jung has a total of 7 co-authors. Here are the top co-authors by number of shared publications:
강희 김
3 shared publications태연 오
3 shared publicationsKang Hee Kim
1 shared publication- Kyuwon Lee1 shared publication
SakJae Lim
Ph.D. Candidate in AI Convergence, Seoul AI School, aSSIST University
1 shared publicationTae Yeon Oh
1 shared publication
Latest Publications
- Open Access
- 30 September 2026
- Published by The Korean Society for Quality Management in Journal of the Korean society for quality management
- Vol. 54 (3) , 457-476
Evaluating Grounded Alignment of LLM Explanations in Financial Risk Early Warning: A Prediction–Explanation Separation Framework
Purpose: This study proposes a large language model (LLM) framework that separates prediction from explanation for early warning of high-volatility regimes in the Korean stock market, in which a dedicated model issues the forecast and an LLM produces auditable, evidence-grounded explanations. The goal is to evaluate and manage the grounded alignment of LLM explanations rather than to use the LLM as a forecaster.Methods: Daily Korea Composite Stock Price Index (KOSPI) and global-market data from January 2015 to May 2026 were used to construct 35 derived features; global variables were aligned at a one-trading-day lag to remove look-ahead bias, and a five-day embargo was applied at split boundaries. Under a strict chronological split, seven models were compared and the main predictor was selected by the area under the precision– recall curve (PR-AUC) on the validation set. The predictor estimates whether next-five-day realized volatility exceeds the top-25% training threshold; its predicted probability, SHapley Additive exPlanations (SHAP) attributions, a market-state summary, and a risk-factor taxonomy are assembled into a Structured Evidence Packet (SEP) that is the LLM's only input. Explanation quality was assessed by a Grounded Evidence Alignment Score (GEAS) and repetition stability, with input ablation (P1/P3/P4) and deletion/injection interventions.Results: XGBoost was selected as the main predictor (validation PR-AUC 0.760), with a test area under the receiver operating characteristic curve (ROC-AUC) of 0.665 and a PR-AUC of 0.782; pairwise model differences were not significant under a moving-block bootstrap. GEAS rose from 0.183 (no SHAP) to 1.000/0.998 when SHAP evidence was provided (paired Wilcoxon, Holm-adjusted p<1e-21), with no significant P3–P4 difference (a ceiling effect). On a common set of 120 sampled test dates, the dedicated model attained PR-AUC 0.799 versus 0.615 for direct LLM prediction (a reference comparison, because the two predictors received inputs of different information scope). Removing the top SHAP cue eliminated its mention (0%), whereas injecting an irrelevant distractor as the top cue produced 100% adoption.Conclusion: GEAS captures adherence to the provided evidence boundary rather than the LLM's internal reasoning faithfulness, and the LLM does not independently validate evidence quality. The findings support using LLMs as grounded, auditable explanation layers rather than autonomous forecasters, with system reliability contingent on the quality of the predictor's evidence.
- 30 September 2026
- Published by The Korean Society for Quality Management in Journal of the Korean society for quality management
- Vol. 54 (3) , 457-476
Open AccessEvaluating Grounded Alignment of LLM Explanations in Financial Risk Early Warning: A Prediction–Explanation Separation Framework
Purpose: This study proposes a large language model (LLM) framework that separates prediction from explanation for early warning of high-volatility regimes in the Korean stock market, in which a dedicated model issues the forecast and an LLM produces auditable, evidence-grounded explanations. The goal is to evaluate and manage the grounded alignment of LLM explanations rather than to use the LLM as a forecaster.Methods: Daily Korea Composite Stock Price Index (KOSPI) and global-market data from January 2015 to May 2026 were used to construct 35 derived features; global variables were aligned at a one-trading-day lag to remove look-ahead bias, and a five-day embargo was applied at split boundaries. Under a strict chronological split, seven models were compared and the main predictor was selected by the area under the precision– recall curve (PR-AUC) on the validation set. The predictor estimates whether next-five-day realized volatility exceeds the top-25% training threshold; its predicted probability, SHapley Additive exPlanations (SHAP) attributions, a market-state summary, and a risk-factor taxonomy are assembled into a Structured Evidence Packet (SEP) that is the LLM's only input. Explanation quality was assessed by a Grounded Evidence Alignment Score (GEAS) and repetition stability, with input ablation (P1/P3/P4) and deletion/injection interventions.Results: XGBoost was selected as the main predictor (validation PR-AUC 0.760), with a test area under the receiver operating characteristic curve (ROC-AUC) of 0.665 and a PR-AUC of 0.782; pairwise model differences were not significant under a moving-block bootstrap. GEAS rose from 0.183 (no SHAP) to 1.000/0.998 when SHAP evidence was provided (paired Wilcoxon, Holm-adjusted p<1e-21), with no significant P3–P4 difference (a ceiling effect). On a common set of 120 sampled test dates, the dedicated model attained PR-AUC 0.799 versus 0.615 for direct LLM prediction (a reference comparison, because the two predictors received inputs of different information scope). Removing the top SHAP cue eliminated its mention (0%), whereas injecting an irrelevant distractor as the top cue produced 100% adoption.Conclusion: GEAS captures adherence to the provided evidence boundary rather than the LLM's internal reasoning faithfulness, and the LLM does not independently validate evidence quality. The findings support using LLMs as grounded, auditable explanation layers rather than autonomous forecasters, with system reliability contingent on the quality of the predictor's evidence.
- Open Access
- 22 July 2026
- Published by Elsevier in SSRN Electronic Journal
MARC: A Multi-axis Robustness Check for Causal Discovery in Environmental Time Series, with Application to Hourly Air Quality
Observation-based causal discovery is increasingly applied to environmental time series, but its robustness to routine analysis choices remains unclear. We present MARC (Multi-Axis Robustness Check), a reusable audit framework that evaluates causal-discovery outputs across five axes: spatial replication, maximum lag, algorithm, effect-size cutoff, and run reproducibility, anchored by domain-knowledge positive controls. Applying MARC to three years of hourly air-quality observations from nine Seoul monitoring stations using PCMCI+ and LPCMCI, we found that only four contemporaneous pollutant couplings (NO 2-O 3 , PM 10-PM 2.5 , CO-NO 2 , and CO-PM 2.5) were invariant across all axes. Directional and lagged claims depended on analysis choices, and the largest lagged effects appeared only under one algorithm, indicating shadow edges. These results suggest that lagged causal claims in hourly environmental data require convergence across algorithms. MARC and notebooks are released for reuse.
- 22 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessMARC: A Multi-axis Robustness Check for Causal Discovery in Environmental Time Series, with Application to Hourly Air Quality
Observation-based causal discovery is increasingly applied to environmental time series, but its robustness to routine analysis choices remains unclear. We present MARC (Multi-Axis Robustness Check), a reusable audit framework that evaluates causal-discovery outputs across five axes: spatial replication, maximum lag, algorithm, effect-size cutoff, and run reproducibility, anchored by domain-knowledge positive controls. Applying MARC to three years of hourly air-quality observations from nine Seoul monitoring stations using PCMCI+ and LPCMCI, we found that only four contemporaneous pollutant couplings (NO 2-O 3 , PM 10-PM 2.5 , CO-NO 2 , and CO-PM 2.5) were invariant across all axes. Directional and lagged claims depended on analysis choices, and the largest lagged effects appeared only under one algorithm, indicating shadow edges. These results suggest that lagged causal claims in hourly environmental data require convergence across algorithms. MARC and notebooks are released for reuse.
- Open Access
- 30 December 2025
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 20 (12) , e0332779
Factor-based deep reinforcement learning for asset allocation: Comparative analysis of static and dynamic beta reward designs
Traditional asset allocation rules, while effective in stable phases, tend to erode once markets enter volatile regimes or undergo structural breaks. Research in deep reinforcement learning (DRL) has usually emphasized raw-return rewards, leaving aside the role of factor exposures ( β ) that shape both risk-adjusted payoffs and adaptive responses. This paper advances a Factor-based Deep Reinforcement Learning for Asset Allocation (FDRL) framework in which β sensitivities—estimated via rolling regressions on momentum, volatility, deviation, and volume signals—inform both the state representation and the reward design. Five reward variants are examined (Sharpe, Sortino, Static- β , Dynamic- β , Momentum- β ) using PPO, SAC, and TD3 across equities, cryptocurrencies, macroeconomic instruments, and mixed portfolios. Empirically, β -based rewards generate heterogeneous but interpretable patterns. In equities, Dynamic- β improves annualized returns from roughly 20% (Sharpe baseline) to 23–24%, with Sharpe rising from 1.04 to about 1.27 across windows. In cryptocurrencies, Dynamic-/Momentum- β achieve 38–43% annual returns but remain highly regime-sensitive, with drawdowns often exceeding –35%. In macro instruments, Static- β delivers the most stable behaviour, maintaining volatilities near 8–9% and limiting drawdowns to roughly –18%. In mixed-asset portfolios, Momentum- β under TD3 produces the strongest gains (cumulative returns above 70–80%), exceeding equal-weight baselines whose CAGR remains near 19–22% with Sharpe ratios around 1.25. All findings were validated through beta-window sensitivity checks (30/60/90/120 days), regime-conditional analysis, and multiple robustness tests including HAC, Wilcoxon, jackknife Sharpe, moving-block bootstrap, and false-discovery-rate adjustments. These diagnostics confirm that the main performance patterns are not driven by window choice or serial dependence. Four contributions follow. First, a reward structure operationalizing time-varying β . Second, systematic benchmarking of factor-sensitive objectives. Third, evidence on asymmetric outcomes across asset classes. Finally, a framework that reconciles responsiveness with interpretability and risk discipline in allocation.
- 30 December 2025
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 20 (12) , e0332779
Open AccessFactor-based deep reinforcement learning for asset allocation: Comparative analysis of static and dynamic beta reward designs
Traditional asset allocation rules, while effective in stable phases, tend to erode once markets enter volatile regimes or undergo structural breaks. Research in deep reinforcement learning (DRL) has usually emphasized raw-return rewards, leaving aside the role of factor exposures ( β ) that shape both risk-adjusted payoffs and adaptive responses. This paper advances a Factor-based Deep Reinforcement Learning for Asset Allocation (FDRL) framework in which β sensitivities—estimated via rolling regressions on momentum, volatility, deviation, and volume signals—inform both the state representation and the reward design. Five reward variants are examined (Sharpe, Sortino, Static- β , Dynamic- β , Momentum- β ) using PPO, SAC, and TD3 across equities, cryptocurrencies, macroeconomic instruments, and mixed portfolios. Empirically, β -based rewards generate heterogeneous but interpretable patterns. In equities, Dynamic- β improves annualized returns from roughly 20% (Sharpe baseline) to 23–24%, with Sharpe rising from 1.04 to about 1.27 across windows. In cryptocurrencies, Dynamic-/Momentum- β achieve 38–43% annual returns but remain highly regime-sensitive, with drawdowns often exceeding –35%. In macro instruments, Static- β delivers the most stable behaviour, maintaining volatilities near 8–9% and limiting drawdowns to roughly –18%. In mixed-asset portfolios, Momentum- β under TD3 produces the strongest gains (cumulative returns above 70–80%), exceeding equal-weight baselines whose CAGR remains near 19–22% with Sharpe ratios around 1.25. All findings were validated through beta-window sensitivity checks (30/60/90/120 days), regime-conditional analysis, and multiple robustness tests including HAC, Wilcoxon, jackknife Sharpe, moving-block bootstrap, and false-discovery-rate adjustments. These diagnostics confirm that the main performance patterns are not driven by window choice or serial dependence. Four contributions follow. First, a reward structure operationalizing time-varying β . Second, systematic benchmarking of factor-sensitive objectives. Third, evidence on asymmetric outcomes across asset classes. Finally, a framework that reconciles responsiveness with interpretability and risk discipline in allocation.
- 31 August 2024
- Published by People and Global Business Association in GLOBAL BUSINESS FINANCE REVIEW
- Vol. 29 (7) , 1-15
AI-based Pairs Trading Strategies: A Novel Approach to Stock Selection
Purpose: This study aims to explore the optimization of Stock Pairs Trading Strategies' performance using AI techniques, with a focus on accurately evaluating stock similarities and selecting the most suitable pairs. Design/methodology/approach: A variety of AI models, including Autoencoders (AE), Vector Embeddings (VE), Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and Gated Recurrent Units (GRU), are utilized to assess the similarity between stocks, which is crucial in the stock pairs selection process for implementing the Pairs Trading strategy. Findings: The implementation of the Pairs Trading strategy with stock pairs selected through AI models showed higher profitability than conventional methods. Strategies utilizing LSTM models demonstrated the highest performance, achieving an approximate cumulative return of 51.25353%. This indicates that AI models are capable of accurately assessing similarities and establishing effective trading strategies. Research limitations/implications: The study highlights the potential of AI-based stock pair selection methods to enhance Pairs Trading Strategies' performance. This approach surpasses traditional statistical methods by better reflecting the stock market's complexity and dynamism, potentially offering investors more stable and higher returns. Originality/value: The research contributes to the field by demonstrating the effectiveness of AI models in the stock pair selection process, suggesting a novel approach to enhancing Pairs Trading Strategies that could provide valuable insights for investors seeking more sophisticated investment strategies in the financial markets.
- 31 August 2024
- Published by People and Global Business Association in GLOBAL BUSINESS FINANCE REVIEW
- Vol. 29 (7) , 1-15
AI-based Pairs Trading Strategies: A Novel Approach to Stock Selection
Purpose: This study aims to explore the optimization of Stock Pairs Trading Strategies' performance using AI techniques, with a focus on accurately evaluating stock similarities and selecting the most suitable pairs. Design/methodology/approach: A variety of AI models, including Autoencoders (AE), Vector Embeddings (VE), Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and Gated Recurrent Units (GRU), are utilized to assess the similarity between stocks, which is crucial in the stock pairs selection process for implementing the Pairs Trading strategy. Findings: The implementation of the Pairs Trading strategy with stock pairs selected through AI models showed higher profitability than conventional methods. Strategies utilizing LSTM models demonstrated the highest performance, achieving an approximate cumulative return of 51.25353%. This indicates that AI models are capable of accurately assessing similarities and establishing effective trading strategies. Research limitations/implications: The study highlights the potential of AI-based stock pair selection methods to enhance Pairs Trading Strategies' performance. This approach surpasses traditional statistical methods by better reflecting the stock market's complexity and dynamism, potentially offering investors more stable and higher returns. Originality/value: The research contributes to the field by demonstrating the effectiveness of AI models in the stock pair selection process, suggesting a novel approach to enhancing Pairs Trading Strategies that could provide valuable insights for investors seeking more sophisticated investment strategies in the financial markets.
- Open Access
- 31 December 2023
- Published by The Korean Society for Quality Management in Journal of the Korean society for quality management
A Study on DRL-based Efficient Asset Allocation Model forEconomic Cycle-based Portfolio Optimization
Purpose This study presents a research approach that utilizes deep reinforcement learning to construct optimal portfolios based on the business cycle for stocks and other assets. The objective is to develop effective investment strategies that adapt to the varying returns of assets in accordance with the business cycle. Methods In this study, a diverse set of time series data, including stocks, is collected and utilized to train a deep reinforcement learning model. The proposed approach optimizes asset allocation based on the business cycle, particularly by gathering data for different states such as prosperity, recession, depression, and recovery and constructing portfolios optimized for each phase. Results Experimental results confirm the effectiveness of the proposed deep reinforcement learning-based approach in constructing optimal portfolios tailored to the business cycle. The utility of optimizing portfolio investment strategies for each phase of the business cycle is demonstrated. Conclusion This paper contributes to the construction of optimal portfolios based on the business cycle using a deep reinforcement learning approach, providing investors with effective investment strategies that simultaneously seek stability and profitability. As a result, investors can adopt stable and profitable investment strategies that adapt to business cycle volatility. Key words: Portfolio Theory, Deep Reinforcement Learning, Business Cycle, Asset Allocation
- 31 December 2023
- Published by The Korean Society for Quality Management in Journal of the Korean society for quality management
Open AccessA Study on DRL-based Efficient Asset Allocation Model forEconomic Cycle-based Portfolio Optimization
Purpose This study presents a research approach that utilizes deep reinforcement learning to construct optimal portfolios based on the business cycle for stocks and other assets. The objective is to develop effective investment strategies that adapt to the varying returns of assets in accordance with the business cycle. Methods In this study, a diverse set of time series data, including stocks, is collected and utilized to train a deep reinforcement learning model. The proposed approach optimizes asset allocation based on the business cycle, particularly by gathering data for different states such as prosperity, recession, depression, and recovery and constructing portfolios optimized for each phase. Results Experimental results confirm the effectiveness of the proposed deep reinforcement learning-based approach in constructing optimal portfolios tailored to the business cycle. The utility of optimizing portfolio investment strategies for each phase of the business cycle is demonstrated. Conclusion This paper contributes to the construction of optimal portfolios based on the business cycle using a deep reinforcement learning approach, providing investors with effective investment strategies that simultaneously seek stability and profitability. As a result, investors can adopt stable and profitable investment strategies that adapt to business cycle volatility. Key words: Portfolio Theory, Deep Reinforcement Learning, Business Cycle, Asset Allocation
- Open Access
- 30 September 2023
- Published by The Korean Society for Quality Management in Journal of the Korean society for quality management
A Study on AI-based Composite Supplementary Index for Complementing the Composite Index of Business Indicators
Purpose The main objective of this research is to construct an AI-based Composite Supplementary Index (ACSI) model to achieve accurate predictions of the Composite Index of Business Indicators. By incorporating various economic indicators as independent variables, the ACSI model enables the prediction and analysis of both the leading index (CLI) and coincident index (CCI). Methods This study proposes an AI-based Composite Supplementary Index (ACSI) model that leverages diverse economic indicators as independent variables to forecast leading and coincident economic indicators. To evaluate the model's performance, advanced machine learning techniques including MLP, RNN, LSTM, and GRU were employed. Furthermore, the study explores the potential of employing deep learning models to train the weights associated with the independent variables that constitute the composite supplementary index. Results The experimental results demonstrate the superior accuracy of the proposed composite supplementary index model in predicting leading and coincident economic indicators. Consequently, this model proves to be highly effective in forecasting economic cycles. Conclusion In conclusion, the developed AI-based Composite Supplementary Index (ACSI) model successfully predicts the Composite Index of Business Indicators. Apart from its utility in management, economics, and investment domains, this model serves as a valuable indicator supporting policy-making and decision-making processes related to the economy. Key words: Composite Index of Business Indicators, Deep Learning, Business Cycle
- 30 September 2023
- Published by The Korean Society for Quality Management in Journal of the Korean society for quality management
Open AccessA Study on AI-based Composite Supplementary Index for Complementing the Composite Index of Business Indicators
Purpose The main objective of this research is to construct an AI-based Composite Supplementary Index (ACSI) model to achieve accurate predictions of the Composite Index of Business Indicators. By incorporating various economic indicators as independent variables, the ACSI model enables the prediction and analysis of both the leading index (CLI) and coincident index (CCI). Methods This study proposes an AI-based Composite Supplementary Index (ACSI) model that leverages diverse economic indicators as independent variables to forecast leading and coincident economic indicators. To evaluate the model's performance, advanced machine learning techniques including MLP, RNN, LSTM, and GRU were employed. Furthermore, the study explores the potential of employing deep learning models to train the weights associated with the independent variables that constitute the composite supplementary index. Results The experimental results demonstrate the superior accuracy of the proposed composite supplementary index model in predicting leading and coincident economic indicators. Consequently, this model proves to be highly effective in forecasting economic cycles. Conclusion In conclusion, the developed AI-based Composite Supplementary Index (ACSI) model successfully predicts the Composite Index of Business Indicators. Apart from its utility in management, economics, and investment domains, this model serves as a valuable indicator supporting policy-making and decision-making processes related to the economy. Key words: Composite Index of Business Indicators, Deep Learning, Business Cycle