Earticle

현대 금융시장에서 신용평가는 금융건전성 평가뿐만 아니라 금융 기관의 대출 심사와 리스크 관리에도 필수적이다. 그러나 금융 환경이 급변하고 머신러닝 기술이 발전함에 따라 기존 신용평가 모형은 한계가 드러나고 있다. 본 논문에서는 2010년부터 2024년 까지 한국기업평가에서 부여한 제조업 기업들의 신용등급 데이터를 바탕으로 신용평 가 모형의 개선 방향을 논의하였다. 본 논문은 데이터 탐색을 통해 기존 신용평가 모형의 문제점을 파악하고, 다양한 머신러닝 기법을 적용하여 신용평가 모형을 개선하 고자 하였다. Random Forest, XGBoost, CatBoost를 활용해 주요 재무 변수의 중요도를 분석하고 신용위험 예측력을 향상시키는 데 초점을 맞추었다. 또한, 데이터 불균형 문제를 해결하기 위해 SMOTE를 적용하고, XAI 기법인 SHAP을 활용하여 신용등급 산정에 사용되는 재무 변수와 임계값 설정의 적정성을 평가하였다. 분석 결과, 실현된 신용위험과 기존 평가 방식에서 결정된 내재적 신용위험을 설명하는 주요 재무 변수가 다름을 확인하였다. 이는 특히 고 신용위험 기업의 평가 기준을 재정립할 필요성을 시사한다. 본 연구는 머신러닝 기반 신용평가 모형의 개선 가능성 을 제시하며, 금융 기관이 보다 정교한 신용위험 관리 전략을 수립하는 데 기여할 수 있다.

Credit scoring is essential for assessing financial soundness and serves as a fundamental tool for loan screening, capital allocation, and risk management in financial institutions. The accuracy and reliability of credit scoring models are directly linked to financial system stability, making their continuous improvement essential. Traditional models primarily rely on Generalized Linear Models (GLM), particularly Logistic Regression. While these models provide interpretable relationships between financial variables and default risk, they are constrained by their linear functional form and reliance on a limited set of features. This restricts their adaptability to evolving financial markets and the increasing availability of unstructured data sources. Advancements in machine learning (ML) and artificial intelligence (AI) have introduced various models to enhance predictive accuracy and address the limitations of conventional credit scoring models. ML-based approaches such as Random Forest, Support Vector Machines (SVM), XGBoost, and LightGBM, along with deep learning techniques, have been widely applied to credit risk modeling. These methods process large volumes of financial and transactional data, capturing complex patterns in credit risk assessment. However, their adoption requires further validation regarding interpretability and regulatory compliance. This study makes four key contributions to credit scoring research. First, unlike previous studies that relied on subjectively selected financial variables, we incorporate all financial features collected by credit agencies and adopt a data-driven selection approach, minimizing researcher bias and ensuring greater objectivity. This enables us to identify the most relevant predictors based on empirical evidence rather than predetermined assumptions. Second, we address the class imbalance issue, a common challenge in credit risk modeling. Since default cases are rare, traditional logistic regression models often suffer from biased estimates, where the model underweights defaulting firms. To mitigate this, we apply the Synthetic Minority Oversampling Technique (SMOTE) to balance the dataset before applying ML techniques. Third, we integrate multiple ML techniques to derive a comprehensive interpretation of feature importance. Specifically, we compare classification performance across Random Forest, Extreme Gradient Boosting (XGBoost), and Category Boosting (CatBoost). Unlike prior studies that analyze a single ML model independently, our approach integrates feature importance rankings across multiple models, providing a more robust estimation of the importance of financial variables in credit risk. Fourth, while ML models enhance predictive accuracy, their complexity can hinder interpretability, making adoption challenging for financial institutions. This study emphasizes the importance of explainable AI (XAI) in credit scoring. By applying Shapley Additive Explanations (SHAP), we provide insights into how key financial variables influence credit risk and default probabilities, offering practical guidance on the appropriateness of financial variables and threshold settings used in credit scoring. This study analyzes credit scoring data of manufacturing firms evaluated by Korea Enterprise Assessment from 2010 to 2024. By applying multiple ML techniques, we identify key financial variables influencing credit risk and integrate results for a comprehensive interpretation. Our analysis highlights differences between realized credit risk, which reflects actual defaults and missed payments, and implied credit risk, which is assessed by the current credit risk model. Realized credit risk is primarily driven by short-term liquidity and profitability indicators, such as inventory turnover period, current ratio, return on equity, and return on capital employed. In contrast, implied credit risk is largely influenced by firm size and long-term financial stability, with key variables including EBITDA, cost-to-sales ratio, pre-tax continuous operating income, total sales, and total liabilities. These findings suggest that while current credit scoring models emphasize long-term financial health, actual credit events are more influenced by short-term financial constraints. This discrepancy underscores the need to supplement credit scoring models by incorporating financial variables, particularly those related to short-term liquidity, especially for high-risk firms. Further analysis reveals that the importance of financial variables varies across rating levels. For A-level firms, short-term financial stability and debt repayment capacity are critical, emphasizing the importance of liquidity management. In contrast, B-level firms are more affected by structural financial indicators such as the debt-to-equity ratio and capital adequacy ratio, highlighting the significance of long-term solvency and debt management. These differences underscore the need to tailor credit scoring criteria based on risk levels. SHAP results indicate that while higher debt-to-equity and capital growth ratios generally reduce the likelihood of default, their impact on credit risk is nonlinear. This suggests that simple threshold-based classification may be insufficient for credit scoring. Instead, a more nuanced approach that accounts for interactions between financial indicators and their varying effects across credit risk levels is needed. Beyond feature importance analysis, we examine credit transitions. Credit scores evolve based on firms' financial conditions. Our findings show that while most firms maintain stable credit scores, downgrades occur more frequently than upgrades, particularly within the B-level category between 2022 and 2023. While some A-level firms experienced rating upgrades between 2019 and 2022, the trend shifted toward downgrades from 2022 to 2023. These patterns highlight the need for dynamic credit transition models that account for temporal changes in creditworthiness.

국내 주식시장에서 배당이 주식수익률에 대한 예측력을 가지는지에 있어 학계에서는 상반된 의견들이 대립해왔다. 이에 본 연구는 2010년부터 2024년 10월까지 KOSPI 및 KOSDAQ 배당주들의 월별 실적자료를 대상으로 배당 관련 주요 시점 (배당 기준, 공시, 확정 및 지급일)에 따른 배당수익률과 주식수익률간의 관계를 실증분석하였다. 그 결과, KOSPI 및 KOSDAQ 주식시장에서는 배당 기준일과 배당 공시일을 기점으로 배당수익률과 주식수익률간 통계적으로 유의한 관계가 나타나고 있음을 확인하였다. 또한 배당수익률과 다음달 주식수익률 간의 이러한 유의한 관계는 연간배당 기업들을 대상으로 발견되며, 배당수익률 상위 또는 하위 산업군에 국한되어 있거나 배당수익률의 개념에 따라 유의미함이 변동하는 것이 아닌 일관된 효과라는 결과를 얻었다. 한편 금융위원회에서는 작년 초 배당절차를 개선하겠다고 발표한 이래, 현재 기업들이 개선된 배당절차를 적극 채택할 수 있도록 유도하고 있다. 주주들에게 배당이 주기적으로 수익을 제공하는 원천일 뿐만 아니라 국내 주식시장에서 배당수익률을 기반으로 포트폴 리오 구성시 주가 차익도 얻을 수 있음을 나타내고 있는 본 연구결과는, 배당을 투자의사 결정에 영향을 미치는 의미 있는 요소로 활용할 수 있는지를 실증적으로 제시하고 있다는 점에서 더욱 그 의미가 부각되는 바이다.

The relationship between dividend yield and stock returns has been extensively studied, particularly in the U.S. market. Prior research, including Campbell and Shiller (1988) and Fama and French (1993), suggests that dividend yield predicts future stock returns. However, studies on the Korean stock market have produced mixed results. While some, such as Kim and Kim (2004) and Jung and Kim (2010), find no predictive power, others, like Oh (2021), provide empirical evidence supporting its significance. Given these conflicting findings, this study aims to clarify the issue by examining whether dividend yield predicts stock returns at specific dividend-related events, hypothesizing that its predictive power varies depending on event timing. To address this question, this study analyzes dividend-paying firms listed on KOSPI and KOSDAQ from January 2010 to October 2024, using financial and stock market data from FnGuide. Dividend yield is measured at four key dividend-related events: the ex-dividend date, the announcement date, the record date, and the payment date. The ex-dividend date marks the point at which shareholders eligible for dividends are determined. The announcement date is when firms publicly disclose their dividend decisions. The record date finalizes the dividend amounts, and the payment date is when dividends are distributed to shareholders. To examine whether dividend yield can predict stock returns, this study employs Fama-MacBeth (1973) cross-sectional regressions while controlling for firm characteristics such as profitability, asset growth, market capitalization, book-to-market ratio, and past stock returns. Additionally, five portfolios ranked by dividend yield are constructed to analyze return patterns across different yield levels. The results indicate that dividend yield significantly predicts stock returns under certain conditions. Specifically, dividend yield strongly predicts stock returns around the ex-dividend and announcement dates, with firms offering higher yields experiencing greater subsequent stock returns. This relationship remains statistically significant even after controlling for the Fama-French three-factor model and momentum effects. However, dividend yield does not predict stock returns around the record or payment dates, suggesting that market reactions occur primarily when dividend information is disclosed rather than when dividends are distributed. This study also examines whether dividend yield’s predictive power depends on payment frequency. The results show a strong and statistically significant relationship for firms with annual dividends, but not for those issuing quarterly or interim dividends. Furthermore, an industry-level analysis reveals that the predictive power of dividend yield is not confined to specific industries but applies broadly across the market, reinforcing its relevance in explaining stock return variations. To ensure robustness, this study tests alternative definitions of dividend yield, including expected future dividends rather than past dividends. The results remain consistent, confirming that the observed predictive effect is not driven by a specific measurement method. These findings provide strong empirical support for dividend yield as a meaningful indicator of stock return predictability in the Korean market. This study makes several key contributions to the literature on dividend yield and stock returns. First, it reconciles conflicting findings by distinguishing dividend yield’s effects at different event dates. By demonstrating that dividend yield significantly predicts stock returns around the ex-dividend and announcement dates, this study clarifies when dividend yield provides valuable information for investors. Second, it provides empirical evidence that dividend yield can enhance portfolio investment strategies, suggesting that investors can improve decision-making by incorporating dividend yield into stock selection criteria. Third, it challenges the notion that dividend yield is solely an industry-specific factor or a function of a firm’s dividend policy. Instead, it plays a crucial role in explaining stock return variations, reinforcing the idea that dividends are not just a mechanism for distributing earnings but also a valuable source of market information. This study also has policy implications for ongoing discussions on dividend reforms in Korea. As the Korean government seeks to enhance transparency and align domestic dividend policies with global standards, this study highlights the importance of clear and timely dividend announcements. By demonstrating that dividend yield contains meaningful information about future stock returns, it provides empirical support for policies encouraging firms to disclose dividend-related information timely and consistently. Improved disclosure practices could help investors make more informed decisions and enhance overall market efficiency. In conclusion, this study offers valuable insights for both policymakers and market participants. For investors, the findings suggest that dividend yield can be a useful tool in constructing profitable stock portfolios. For regulators, the study highlights the need for policies promoting better dividend disclosure. By addressing these issues, it contributes to a deeper understanding of dividends’role in stock return predictability and offers practical implications for investment strategies in the Korean stock market.

국민연금은 기금 고갈에 대한 우려와 함께 수익성 제고를 위해 주식 등 위험자산의 비중을 확대해 왔다. 그렇다면 지금까지 위험자산 투자에서 국민연금은 어느 정도 수익을 내고 있는가? 그 수익의 원천은 어디인가? 본 연구는 이에 대해 답하기 위해서 국민연금이 보유한 국내 주식 포트폴리오를 직접 분석하였다. 국민연금이 보고한 국내 주식 보유현황 신고 자료와 한국거래소의 매매내역 자료를 결합하여 2009년부터 2018년까지의 국민연금 국내 주식 월별 보유내역을 계산하였다. 이를 바탕으로 국민 연금의 10년간 운용성과를 계산하고 평가하였다. 분석결과, 국민연금이 보유한 국내 주식 포트폴리오는 월평균 0.57%(연 6.84%)의 수익률을 얻었다. 다양한 위험조정 수익률도 모두 통계적 유의성을 보였다. 보유주식을 시장별로 구분하였을 때, 유가증 권시장 주식의 수익률이 0.58%로 코스닥시장 수익률 –0.01% 보다 훨씬 컸다. Daniel et al.(1997)의 방법론에 따라 국민연금 투자성과를 분해하였을 때, 종목선택 능력(CS, characteristic selectivity)은 0.44%, 시장타이밍능력(CT, characteristic timing)은 –0.16%, 그리고 투자스타일(AS, average style)은 0.30%로 구성되었 다. 본 연구를 통해 국민연금의 투자수익률 대부분이 유가증권 주식에 대한 종목선택 능력(CS)와 투자스타일(AS)에서 기인함을 발견하였다.

The National Pension Service(NPS) has increased the share of risky assets such as stocks to raise profitability with concerns over fund depletion. If so, has the NPS been making a profit on investments in risky assets so far? If the NPS are making a profit, where is the source of that ability? To answer this question, we directly analyzes the Korean stock portfolio held by the NPS. The objective of this study is to examine the operational performance of NPS investments in the Korean stock market over the past decade, as well as the sources for this performance. The NPS invests in a variety of risky assets, including stocks, bonds, and alternative investments. Of these, domestic stocks account for about 14% of the total as of the end of 2023, amounting to KRW 148 trillion in investments. The NPS is also the largest institutional investor in the domestic stock market and has a huge influence on the market, so it is very important. We obtain the monthly holdings of NPS in domestic stocks from 2009 to 2018 by combining the report on large holdings of stocks reported by the NPS and the trading data from the Korea Exchange (KRX). Specifically, we calculate the transaction history of the NPS from the KRX. The KRX's intraday trading data contains information on the seller and buyer for every transaction, and we use the seller and buyer information to identify the NPS accounts. The process is as follows. First, we obtain the name, dates, quantity and price of purchase and sale stocks from 72,622 reports from 2008 to 2018 in the name of the NPS through the “Report on Large Holdings of Stocks and Other Securities” and “Report on Ownership of Certain Securities by Officers and Major Shareholders.” Second, we identify the NPS accounts by matching the transaction details of each account with the transaction data of the KRX. Based on 72,622 reports, we identify 12,449 NPS accounts from 46 securities firms. This study analyzes the performance of NPS using two methods in addition to the raw portfolio return. First, we evaluate the performance of the NPS using various benchmarks. We examine risk-adjusted returns by using single factor or multi-factor models. However, it has the disadvantage of evaluating the performance based on the assumptions of the risk factor model. Second, we use the Daniel et al. (DGTW, 1997) methodology to decompose returns by controlling for a benchmark portfolio based on firm characteristics. This methodology overcomes the shortcomings of the traditional risk factor model. The main results of the empirical analysis are as follows First, the domestic equity portfolio held by the NPS has an average monthly return of 0.57% over the period from 2009 to 2018. This is equivalent to an annualized rate of 6.84%. The risk-adjusted return using a one-factor model of market returns is 0.31% per month, which is statistically significant. The risk-adjusted return using the Fama-French (1993) three-factor model is 0.49% per month and the risk-adjusted return using the Carhart (1997) four-factor model is 0.48% per month, both statistically significant. This shows that even after controlling for risk, the NPS is still generating significant returns. By year, the risk-adjusted returns are higher and statistically significant mainly in 2012 and 2013. Second, we separate the returns of the NPS into the KOSPI and the KOSDAQ market. The return on the KOSPI stocks is 0.58%, which is much larger than the return on KOSDAQ stocks (-0.01%). The NPS earns most of its returns from investing in the KOSPI market stocks and loses money in the KOSDAQ market. The risk-adjusted returns of the KOSPI stocks are also statistically significant at 0.33%, 0.51%, and 0.51%, respectively, while the risk-adjusted returns of the KOSDAQ stocks are not statistically significant. Overall, the performance of the NPS is better in the KOSPI market than in the KOSDAQ. Third, we decompose the investment performance of NPS according to the methodology of Daniel et al. (DGTW, 1997). DGTW decomposes fund performance into characteristic selectivity (CS), market timing (CT), and average style (AS). The analysis shows that the portfolio return is composed of 0.44% for CS, -0.16% for CT, and 0.30% for AS. The largest return comes from stock selection, while market timing is not significant with a negative return. This result is similar to other studies in the U.S. and Korea that decompose the performance of active funds. When we decompose the portfolio returns of the NPS into the KOSPI and KOSDAQ markets, we also find that the returns of the NPS are mostly due to stock selection ability (CS) and investment style (AS). Our results show that the majority of the NPS's investment performance is attributable to its ability to select Korean equities and average style performance.

본 연구는 투자자 심리가 주식 수익률에 미치는 영향을 탐구하기 위해 새로운 안도감 변수(REL)를 제시하고, 이를 후회 변수와 비교하여 분석한다. REL 변수는 동일 산업 군 내 최저 수익률과 개별 자산의 수익률을 비교하여 투자자가 느끼는 심리적 안정성 을 반영한다. 연구 결과, 투자자는 안도감이 높은 자산에 대해 낮은 기대 수익률을 수용하며, 반대로 안도감이 낮은 자산에는 더 높은 위험 프리미엄을 요구하는 경향이 있다. 이러한 효과는 후회 변수와 기업 특성 요인을 통제한 후에도 지속되며, 특히 소규모 기업이나 고유 변동성이 큰 자산에서 더욱 두드러진다. 행동 재무학의 관점에 서 본 연구는 투자자의 효용함수가 심리적 요인에 의해 어떻게 조정되며, 이것이 투자 행동과 자산 가격 형성에 미치는 영향을 분석한다. 특히, 후회와 안도감이 투자자의 효용을 변화시키며 자산 가격 결정 과정에 미치는 상호작용을 설명한다. 또한, 심리적 가격 장벽이 투자자의 감정을 증폭시켜 REL 변수의 영향을 강화한다는 점도 밝혀냈 다. 본 연구는 투자 전략 수립과 리스크 관리에서 심리적 요인을 반영한 새로운 접근을 제안하며, 개인 투자자가 보다 합리적이고 감정적으로 균형 잡힌 의사결정을 내릴 수 있도록 돕는 방향성을 제시한다.

This study introduces a new sentiment-based variable called Relief (REL) to examine the impact of investor psychology—particularly emotions such as relief and regret—on asset pricing and decision-making in financialmarkets. Rooted in behavioral finance, the REL variable aims to capture a specific dimension of psychological stability that investors experience when their chosen asset performs better than the worst-performing peer in the same industry group. The REL variable is constructed by comparing the return of an individual asset with the lowest return among its industry peers. This comparison reflects an investor’s emotional comfort—or relief—from having avoided the worst-case scenario, even if the absolute return is modest or negative. Essentially, the REL variable quantifies a positive psychological payoff derived from relative outperformance, even when that outperformance is not impressive in absolute terms. Empirical analysis reveals a consistent pattern: investors tend to accept lower expected returns for assets with high REL values, indicating a stronger sense of emotional relief. In contrast, when an asset has a low REL value—meaning it did not outperform even the worst peer—investors demand a higher risk premium. This behavior highlights how emotional states influence investment decisions beyond traditional risk-return trade-offs. Importantly, the effects of the REL variable remain statistically significant even after controlling for the regret variable and firm-specific characteristics such as size, volatility, and the book-to-market ratio. This suggests that REL captures a unique behavioral dimension not fully explained by existing psychological or financial models. Moreover, the impact of REL is especially pronounced in small-cap stocks and assets with high idiosyncratic volatility—markets that are inherently more uncertain and thus more susceptible to emotionally driven behavior. The study also explores how psychological price anchors—such as recent highs—can amplify emotional reactions associated with REL. Investors become more sensitive to relative performance comparisons when asset prices move away from the psychological price barrier. In such contexts, feelings of regret or relief are intensified, causing greater deviations from rational investment behavior. These findings demonstrate that emotional biases are not isolated anomalies but are systematically embedded in market dynamics. One of the study’s key theoretical contributions lies in its challenge to traditional financial models that assume stable preferences and rational utility maximization. Instead, the REL variable supports a dynamic model of investor utility, where emotional responses like regret and relief actively reshape decision-making preferences. Regret reflects the emotional cost of missing a better opportunity, while relief provides a positive emotional benefit from avoiding the worst outcome. These emotions function as psychological forces that alter how investors evaluate gains and losses. Incorporating REL into behavioral finance expands the framework by accounting not only for negative emotions such as regret but also for positive emotions like relief. This dual-emotion perspective helps explain why investors may favor certain assets—not necessarily due to their fundamentals—but because of how those assets make them feel in relative terms. Investors may feel reassured knowing that their asset did not perform the worst, even if the overall return was subpar. From a practical perspective, the findings offer valuable insights for investment strategy, portfolio management, and risk assessment. Investment professionals who consider sentiment-based variables such as REL may be better positioned to understand market dynamics shaped by investor psychology. By identifying assets that generate strong feelings of relief, investors can anticipate lower return expectations, while also recognizing the increased return demands associated with assets that trigger regret or emotional discomfort. Additionally, understanding the emotional reinforcement embedded in relative performance may lead to more psychologically resilient portfolios. Rather than relying solely on traditional metrics like beta, volatility, or book-to-market ratios, integrating psychological measures like REL enables investors to develop strategies that reflect real-world investor behavior—where emotions often have a stronger influence than pure analysis. In conclusion, this study contributes to a more comprehensive understanding of investor behavior by introducing the REL variable as a meaningful tool for capturing emotional responses to relative performance. The REL framework enhances traditional regret-based models by including the often-overlooked role of positive emotional feedback. This research demonstrates that asset pricing is shaped not only by fundamentals or rational expectations, but also by how investors feel about their performance in relation to the worst possible outcomes. By identifying and quantifying this emotional comfort, the REL variable illuminates the complex ways in which sentiment influences market outcomes. It offers both theoretical insight and practical guidance for building emotionally aware investment models, reinforcing the view that both positive and negative emotions play a central role in financial decision-making.

This study examines the performance of regime-based portfolio rebalancing strategies that utilize hidden Markov models (HMMs) to dynamically adjust asset allocations in response to volatility regime shifts. Based on a globally diversified portfolio reflecting the investment context of institutional investors with significant cross-border exposure, the analysis yields three key findings. First, regime-based strategies consistently outperform static strategic asset allocation (SAA) and buy-and-hold approaches in terms of risk-adjusted returns, while effectively reducing downside risk measures such as conditional value-at-risk (CVaR), maximum drawdown (MDD), and volatility. Second, the benefits of regime-based strategies are more pronounced at higher adjustment intensities, confirming their adaptability under shifting market conditions. Third, the performance advantages of regime-based strategies persist even after incorporating transaction costs. By integrating regime-based signals into the rebalancing process, this study provides empirical implications for institutional investors managing globally diversified portfolios in volatility-sensitive environments.

본 연구는 기업의 환경, 사회, 지배구조(ESG) 활동이 신용위험과 신용등급에 미치는 영향에 대한 실증분석을 수행하였다. 구체적으로, 신용위험의 대용치인 Merton(1974) 의 부도거리와 신용평가사의신용등급에 기업의 ESG 평가가 반영되고 있는지, ESG 활동이 신용위험에 미치는 영향에 대한 시장의 평가와 신용평가사의 평가 간에 차이가 있는지를 분석한다. 분석결과, 기업의 ESG 활동은 신용위험 및 신용등급과 통계적으로 유의한 관계가 있음을 발견할 수 있었다. ESG 개별항목 중 지배구조부문은 신용위험 및 신용등급에 유의한 영향을 미치는 것으로 나타났으며, 사회부문은 신용등급에만 유의하게 반영되는 것으로 나타났다. 반면, 상대적으로 최근에 관심도가 높아지고 있는 환경부문은 신용위험, 신용등급 모두에 유의한 영향을 미치지 않는 것으로 나타났다. 이와 같이 ESG 개별 항목별로는 신용위험 및 신용평가 반영에 차이가 있는 것으로 보이며, 신용평가사가 시장에 비해 선제적으로 반영하는 것으로 해석된다. 또한, 최근 일부 신용평가사가 환경부문을 신용평가에 적극적으로 반영하겠다는 의지를 표명한 바, 향후에는 환경부문의 성과가 신용등급에 반영될 것으로 기대되며 시장도 이를 점차 반영할 것으로 기대된다.

This study examines the implications of corporate ESG practices on credit risk and credit ratings of Korean listed firms. In particular, we investigate whether ESG performance is reflected in credit risk that is derived from stock price information and credit ratings assigned by credit rating agencies, respectively. This paper aims to advance our understanding of the relationship between ESG and credit risk and promote discussion about incorporating ESG factors into credit ratings. To measure the stock-based default risk, we calculate distance to default based on the widely used Merton (1974) bond pricing model. In the model, a firm’s equity can be viewed as a call option on the firm’s assets since at the maturity of a firm’s debt, the debt holders receive their debts, and the equity holders get the remaining amount. Our Merton distance-to-default measure is calculated using the numerical procedure of Bharath and Shumway (2008). Since the growth rate of a firm’s asset value is difficult to estimate, we use two estimates for the asset value drift rate, one using the risk-free interest rate and the other using CAPM. Regarding credit ratings, firms that issue corporate bonds in Korea are required to obtain credit ratings from at least two of the three separate official credit rating agencies (Korea Investors Service, Korea Ratings and NICE Investor Service). The lower of the two or three ratings is used as the bond credit rating. The credit rating system consists of a total of 22 rating grades ranging from AAA to D (10 investment grade ratings from AAA to BBB- and 12 non-investment grade ratings from BB+ to D). We convert the letter grades to numbers varying between 1 and 22 points, where 1 point corresponds to the highest rating, AAA. We obtain ESG ratings from the Korea Corporate Governance Service (KCGS). The KCGS’s ESG ratings include three categories: environmental, social and governance. Specifically, the environmental category includes environmental strategy and organization, environmental performance, and stakeholder relations. The social category includes employees, consumers, community, and partners and competitors. The governance category includes protections of shareholder rights, boards of directors, audit institutions and disclosure. The ESG rating isassigned in October every year, and after the final ratings are released, the ratings are adjusted in January, April, and July of the following year to reflect the latest issues if any. The KCGS ESG rating consists of seven grades: S, A+, A, B+, B, C, D, with S being the highest level. We convert the letter grades to numbers varying between 1 and 7 points, where 1 point corresponds to the highest rating, S. Our baseline panel regression model tests our hypothesis that ESG rating is negatively associated with the distance to default and positively associated with the credit rating. The dependent variables are the two distance-to-default measures and the credit rating. The independent variables are ESG ratings with a set of firm-level control variables. The firm-level control variables include size, leverage, cash flow, return on assets, stock beta, and asset volatility. We include firm and industry-year fixed effects for the regression. Using the abovementioned two measures of credit risk and ESG ratings from the KCGS, we find that corporate ESG performance is positively associated with the distance to default and a firm’s credit ratings. When we separate ESG into environmental (E), social (S), and governance (G) categories, we find that G is significantly associated with both the distance to defalutand credit ratings. On the other hand, S is significantly associated only with credit ratings. However, neither distance to default nor credit ratings are significantly associated with E, which has recently been attracting substantial attention. As such, credit rating agencies seem to incorporate ESG information when evaluating credit risk ahead of the stock market investors and considers governance ratings more importantly than the social and environmental ratings. It should be noted, however, that our results do not imply that corporate environmental performance is not an important factor when evaluating the credit risk of the companies. Given that credit rating agencies recently announced that they will considerenvironmental aspects of corporate activities when evaluating the credit risk of firms, the relationship between credit ratings and environmental performance will likely strengthen in the near future. Overall, depending on information availability and public attention toward ESG issues, there could be a time delay before environmental, social, and governance factors are fully incorporated into credit ratings. Last but not least, our findings suggest that ESG factors are increasingly important in corporate credit risk assessment and investors can improve credit risk management in corporate bond portfolios by considering ESG factors when evaluating corporate credit risk.

본 연구에서는 그래프 신경망 모형을 포함한 6개의 기계학습모형을 활용하여 한국 기업의 신용변동을 예측했다. 이를 예측하기 위해 수익성, 성장성, 안정성, 활동성 지표들을 활용하였고, 주성분 분석을 진행하여 지표별로 두 개씩 활용했다. 또한, 신용등급변동이 없는 경우가 대다수인 불균형 데이터를 처리하기 위해 SMOTE 방법 론을 활용하여 데이터를 오버샘플링했다. 본 연구에서는 산업구조를 반영하기 위해 이를 나타내는 상장시장, 데이터 시점, 현재 신용등급과 업종의 네 분류로 나누어 모형을 만들고 이를 최종적으로 결합하여 예측 결과를 도출했다. 결과를 살펴보면, 모든 모형이 데이터 비중에 맞게 선택하는 벤치마크 모형보다 우수하고, 그래프 신경 망 모형이 특히 더 우수한 성능을 나타내는 것을 확인할 수 있다. 또한, 변수 중요도를 분석한 결과, 성장성 지표와 활동성 지표가 중요한 것을 확인할 수 있었고, 그래프 신경망에서는 업종 그래프와 상장시장 그래프가 성능이 좋은 것을 확인할 수 있었다.

Credit rating is an essential component of the financial sector, as it evaluates a firm’s capacity to meet debt obligations. In the Republic of Korea, rating agencies assign levels ranging from AAA to D, with additional modifiers, and these ratings significantly affect financing conditions. Traditional methods typically rely on logistic regression and selected financial variables, yet these approaches often face difficulties in capturing the intricate or nonlinear patterns present in corporate financial data. In response to these challenges, researchers have increasingly turned to advanced machine learning algorithms that can account for more complex relationships. Nevertheless, their deployment is limited by relatively small datasets—particularly among smaller firms—and by concerns regarding model interpretability. The present study proposes a machine learning framework, including a Graph Neural Network (GNN), to predict rating changes in Korean firms. The data set spans 2010 to mid-2024 and includes 182 firms, yielding 1,417 year-level samples. Each annual observation is labeled according to whether its credit rating was upgraded (+1), downgraded (–1), or left unchanged (0). Because most observations lie in the unchanged category, the data are highly imbalanced. To address this imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) is employed, generating additional samples in the minority classes. In addition, Principal Component Analysis (PCA) is utilized to reduce the dimensionality of sixteen indicators representing profitability, growth, stability, and activity. Six models are assessed: logistic regression, LASSO, random forest, support vector machine, Light Gradient Boosting Machine (LGBM), and a GNN. The GNN-based approach is noteworthy for modeling financial indicators or corporate attributes as nodes within a graph, with edges delineating the relationships among these nodes. Such a representation allows for the capture of latent dependencies that are difficult to detect in methods that treat predictors independently. Furthermore, each corporate sample belongs to discrete subgroups determined by listing market (KOSPI or KOSDAQ), data period, rating band, and industry classification. The proposed hierarchical GNN merges these subgroup-specific networks to produce a consolidated prediction, thereby incorporating both firm-level attributes and group-level characteristics. When compared to a benchmark that classifies samples randomly in proportion to the observed class distribution, all six machine learning algorithms demonstrate superior performance. The GNN shows the highest precision and F1-scores, suggesting that it is particularly effective at identifying upgrades and downgrades, which are far less common than no rating changes. Nonetheless, like other models, it finds rare rating shifts more challenging to predict, highlighting the impact of data imbalance and the difficulty of forecasting uncommon events. An inspection of feature importance across models underscores the significance of growth and activity metrics, implying that sales expansion, equity growth, and the efficient use of assets offer robust signals of rating volatility. Moreover, the GNN indicates that distinguishing firms by industry group is especially influential, possibly because each sector’s distinctive regulatory, economic, and financial traits shape its credit risk profile. Compared to certain deep neural networks that demand extensive datasets, the GNN-based method presented here is relatively more practical in settings with limited data, including smaller firms with incomplete rating histories. Additionally, this approach provides improved transparency, as the graph architecture clarifies how different financial indicators or subgroups collectively affect rating transitions. Future work may benefit from enlarging the dataset, experimenting with alternative oversampling strategies such as ADASYN, and examining cost-sensitive learning to mitigate the imbalance problem further. Investigations might also consider alternative graph structures that connect entire firms as nodes and delineate inter-firm relationships or incorporate advanced architectures such as Transformers or LSTM networks. In summary, the findings suggest that a GNN-based framework can improve credit rating predictions by capturing complex interactions that traditional or other advanced machine learning methods may overlook. While data imbalance is still a problem, its consequences are somewhat mitigated by SMOTE. The significance of growth, activity, and sector-specific characteristics suggests that more accurate and comprehensible rating projections can be produced by integrating richer and more interconnected data. In the end, more investigation and more extensive data gathering should improve the precision and dependability of credit rating systems, leading to a better comprehension of the dynamics of corporate finance.

과신 경영자는 자신의 능력에 대한 과도한 판단으로 인해 더 위험한 프로젝트를 수행하고 따라서 기업 가치에 부정적인 영향을 준다는 연구와 적극적인 혁신 활동을 통해서 기업 가치를 높인다는 실증 결과가 서로 상충되는 과신 경영자 퍼즐(overconfident manager puzzle)이 존재한다. 본 연구는 선행연구에서 사용한 과신 변수가 실제로는 미래에 대한 낙관적인 전망(optimistic perspective)을 측정하고 있음에 착안하여, 비현실적이거 나 근거가 희박한 낙관성을 과신(unrealistic optimism)으로, 현실에 근거한 긍정적인 전망은 낙관성(realistic optimism)으로 구분한다. 2009년부터 2020년 동안 8,382개 국내 표본을 대상으로 한 본 연구의 주요 결과는 다음과 같다. 경영자가 낙관적인 전망을 가질수록 주가 수익률의 변동성으로 측정한 투자 위험성이 줄어드는 음(-)의 관계를 나타냈는데, 이는 과신 경영자가 위험한 투자를 실행한다는 일부 선행 연구의 예측과 다른 결과이다. 또한 국내에서 낙관적 전망의 경영자는 더 많은 R&D 투자를 하고, 더 많은 특허를 만들고, 더 많은 현금 보유량을 가지는 것으로 나타났다. 그리고 혁신 활동의 결과로 인해 낙관적인 경영자의 기업일수록 Tobin’s Q로 측정한 기업 가치가 증가한다는 사실을 발견했다. 본 연구의 결과는 미래에 대해 낙관적 전망을 가진 경영자가 과도한 위험 투자보다는 혁신을 강화하는 효율적인 투자 활동을 통해 기업 가치를 높이고 있음을 의미하며, 미래에 대해 낙관적 전망을 가진 경영자가 비이성적 심리적 편향(bias) 보다는 현실에 근거한 낙관성(realistic optimism)을 가지고 있음을 시사한다.

Research suggests that overconfident managers often harm firm value by taking excessive risks due to overly optimistic future outlooks or inflated self-assessments (Griffin and Tversky, 1992; Malmendier and Tate, 2008). Overconfident CEOs are reported to overestimate the expected returns from uncertain ventures. However, Hirshleifer, Low, and Teoh (2012) present evidence that overconfident CEOs, who tend to be more enthusiastic about challenging and risky projects, invest more in innovation and often succeed, introducing the "overconfident manager puzzle" where such managers enhance rather than damage firm value. This study aims to test these conflicting findings in the context of Korean firms. It explores the hypothesis that optimistic managers—unlike overconfidence, which is traditionally viewed as harming firm value through overinvestment—might instead be innovators who enhance firm value. The study utilizes a novel methodology involving machine learning to measure managerial overconfidence. Unlike U.S. studies that often use stock options as indicators, this research leverages text analysis of managerial opinions disclosed in business reports, using the BERT machine learning model to quantify optimism. The ambiguity in measuring overconfidence in previous studies is another consideration. Psychologically, overconfidence is seen as an irrational bias, but empirical variables might capture a manager’s rational optimism about the future. If a manager’s outlook is rational, overconfidence could lead to positive outcomes, unlike the negative connotations typically associated with it. Many studies interchange overconfidence with optimism, where the former implies irrational excessive confidence and the latter signifies a positive future outlook. Generally, overconfident CEOs are believed to make poor decisions by overestimating future performance and underestimating risks, leading to value-destroying mergers and acquisitions (Malmendier and Tate, 2005, 2008). Such CEOs are often reported to overpay in M&A deals, necessitating strict control through compensation and governance structures. Overconfident managers also tend to invest more than their less confident counterparts, potentially harming firm value through overinvestment (Moez and Amina, 2008; Chen, Ho, and Ho, 2014). Conversely, some researchers highlight the positive roles of overconfidence. It can enhance decision-making execution, encourage necessary risk-taking for shareholder benefit, and stimulate entrepreneurial activities (Russo and Schoemaker, 1992; Goel and Thaker, 2008; Bernardo and Welch, 2001). Hirshleifer et al. (2012) found that overconfidence negatively impacts acquisitions but positively influences innovation. They suggest that overconfident managers in firms with innovation opportunities can achieve significant success, unlike those in firms without such opportunities who might make detrimental acquisition decisions. The study’s empirical analysis yielded several key findings. First, contrary to expectations, there was a negative relationship between CEO Optimism and risk, as measured by stock return volatility, suggesting that optimistic CEOs do not prefer riskier projects unlike overconfident CEOs. However, this relationship turned positive when controlling for capital availability, indicating that optimistic CEOs choose risky projects when capital is accessible. This implies that the measure of optimism might reflect rational optimism rather than irrational bias. Second, optimistic CEOs were found to increase R&D investments and engage more in innovation activities, such as filing patents. Third, they tend to hold more cash to seize future investment opportunities. Last, optimistic CEOs were confirmed to enhance firm value, aligning with Hirshleifer et al. (2012), suggesting that they are rational optimists driving innovation rather than irrationally overconfident leaders. In conclusion, the study reaffirms that optimistic managers can be seen as rational optimists whose confidence drives innovation and firm value enhancement, challenging the traditional view of overconfidence as purely detrimental.

본 연구는 국내 주요 유통·커머스 플랫폼 8개사를 대상으로 2020~2024년 재무제표 와 부도 사건 데이터를 활용하여 구조적·유동성 리스크를 조기에 진단하는 정량적 프레임워크를 제시한다. Altman Z″-Score, Ohlson O-Score, Piotroski F-Score 와 더불어 Burn Rate, Cash Runway를 주요 변수로 설정하고, 부도 발생 여부 예측에는 로지스틱 회귀와 베이지안 회귀, 발생 시점 분석에는 Cox 비례위험모형을 적용하였다. 분석 결과, Ohlson O-Score와 Cash Runway가 부도 가능성과 발생 시점 모두에서 통계적으로 유의한 핵심 변수로 확인되었으며, 베이지안 분석은 소표본 에서도 높은 예측 신뢰도를 제공함을 보여주었다. 이를 토대로 ‘긴급 모니터링→정량 진단→위험군 분류’ 3단계 조기경보체계(EWS)와 리스크 기반 핀포인트 규제를 설계 하였다. 또한 EU·미국·일본 사례 분석을 통해 플랫폼 거래 투명성 강화와 선제적 감독체계를 위한 정책 방향을 도출함으로써, 유통·디지털 커머스 생태계의 지속가능 성과 거래 신뢰성 제고에 기여하고자 한다.

In recent years, digital commerce platforms have grown at an unprecedented pace, fundamentally reshaping retail and service industries worldwide. While these platforms prioritize rapid user acquisition and market share expansion, often at the expense of immediate profitability, their business models inherently carry heightened financial vulnerability. Aggressive marketing expenditures, heavy investments in logistics and technology, and extended deferred payment arrangements with vendors contribute to elevated cash burn rates and liquidity risk. Traditional insolvency prediction models, designed for asset-intensive or manufacturing firms, fail to capture the nuanced risk dynamics of digitally mediated, two-sided marketplaces. Against this backdrop, this study develops and empirically validates a hybrid risk-diagnosis framework tailored to digital commerce platforms, combining conventional insolvency scores with liquidity-specific indicators. Specifically, we integrate Altman’s Z″-Score, Ohlson’s O-Score, and Piotroski’s F-Score with burn rate and cash runway metrics to holistically assess structural solvency, operational health, and short-term survival capacity. Our empirical analysis covers eight major Korean platform firms—Homeplus, Woowa Brothers (Baemin), Coupang, TMON, Wemakeprice, Balan, skplanet, and Kurly—over the 2020–2024 period. Using publicly disclosed financial statements and, where necessary, reputable media reports to supplement missing data, we first employ logistic regression to identify the most predictive variables for default occurrence. The results reveal that the Ohlson O-Score (β≈+0.84, p≈0.10) and Cash Runway (β≈–0.60, p≈0.12) exert statistically meaningful effects on default odds, with one-unit increases in O-Score and one-month increases in runway corresponding to roughly 2.3× higher and 0.56× lower default odds, respectively. Conversely, Z″-Score and F-Score showed limited predictive power in this binary setting. To extend beyond binary classification and capture the temporal dimension of default risk, this study applies both frequentist and Bayesian Cox proportional hazards models. Consistent with prior logistic regression results, both models identify Ohlson O-Score and Cash Runway as statistically significant predictors of time-to-default. The frequentist model estimates a hazard ratio of approximately 2.22 for O-Score and 0.56 for Runway, indicating a higher default hazard with increased financial stress and a protective effect from greater liquidity. To enhance robustness under small-sample constraints and quantify uncertainty, we implement a Bayesian Cox model using Markov Chain Monte Carlo (MCMC) sampling via the No-U-Turn Sampler (NUTS). Posterior estimates confirm the predictive power of O-Score and Runway, with 95% credible intervals excluding zero and posterior probabilities exceeding 99%. This Bayesian approach provides not only directional validation but also interpretable uncertainty bounds critical for risk-sensitive policy decisions. Building on these insights, we propose a three-stage Early Warning System (EWS) and a complementary risk-based regulatory framework (“pinpoint regulation”). Stage 1 real-time monitoring flags acute liquidity stress via burn rate spikes or runway collapse. Stage 2 quantitative diagnosis confirms structural weakness using O-Score and Z″-Score thresholds. Stage 3 synthesizes signals to classify platforms into high-risk, recovering, and stable cohorts. For high-risk firms, we recommend emergency liquidity support (conditional lending, escrow mandate), shortened settlement cycles, and external management oversight. Recovering firms receive buffer instruments (guarantees, tax incentives) and quarterly indicator reviews, while stable firms benefit from annual stress tests and disclosure expansion to foster transparent, sustainable growth. Internationally, we compare EU, US, and Japanese regulatory approaches: the EU’s PSD2 and Late Payment Directive mandate client fund segregation and penalize payment delays; the US relies on private escrow services, trust-account rules under Money Transmitter Laws, and trade-credit insurance; Japan’s Transparency Law enforces upfront disclosure and administrative guidance. These case studies underscore the effectiveness of segmented regulation—combining pre-emptive oversight for high-risk entities with marketfriendly self-regulation for sound firms. Academically, our research enriches platform finance literature by adapting legacy credit risk models to the liquiditysensitive, digitally mediated economy. Practically, it delivers a replicable methodology and policy toolbox for regulators, investors, and platform operators to preempt financial distress, safeguard transactional ecosystems, and promote resilient, inclusive growth in the digital commerce sector.

This study investigates the relationships between three leading cryptocurrencies— Bitcoin, Ethereum, and Ripple—and six major implied volatility indices as proxies for stock market uncertainty: VIX, VXD, VSTOXX, VDAX, VXEFA, and VXEEM. We employ nonparametric causality-in-quantiles and quantile-on-quantile approaches to examine nonlinear causal effects and dependence structures across various cryptocurrency market states and levels of uncertainty. Our findings reveal a one-way information flow from cryptocurrency returns to stock market uncertainty, with stronger predictive power during periods of low uncertainty. This suggests that while stock market uncertainty may not reliably predict cryptocurrency returns, cryptocurrency-related information can significantly influence stock markets, particularly during low uncertainty, likely driven by investor attention to cryptocurrencies for diversification or speculative purposes. The quantile-on-quantile analysis shows that changes in implied volatility generally have a negative impact on cryptocurrency returns, with these effects being more pronounced at lower quantiles. Furthermore, cryptocurrencies’potential to act as hedges or safe havens against stock market uncertainty emerged only under extremely bullish market conditions and has significantly diminished over time due to the increasing integration of cryptocurrency and stock markets.