Korea Capital Market Institute

Since the outbreak of COVID-19, Korea’s capital market has seen a rapid expansion of its retail investor base, with a growing number of individuals actively investing in overseas assets and structurally shifting from a domestic-equity-centered focus to broader global portfolios. Yet, there has been little account-level empirical evidence on how individuals conduct overseas direct investment, how foreign asset allocation reshapes their realized returns and risk, and through what channels characteristics such as age, gender, and asset size affect performance. To address this gap, this report reconstructs individual portfolios that include both domestic and foreign listed stocks and ETPs, using detailed account-level holdings and transactions for roughly 100,000 investors at a major securities firm between 2020 and 2022, and analyzes their investment behavior and outcomes in an integrated way.  The results show that, although domestic equities remain dominant overall, younger investors in their 20s and 30s and high-wealth investors hold significantly higher shares of foreign assets and are especially active in overseas ETPs. Male investors tend to hold fewer positions but concentrate more heavily in higher-risk assets compared with women, while high-wealth investors more often build systematic global portfolios centered on foreign ETFs. By contrast, small-scale investors frequently trade high-leverage, derivative-type foreign ETPs with limited capital. Even when the number of holdings is large, effective diversification is often weak, with assets clustered in a few names and specific U.S. stocks and ETPs, resulting in structures that superficially resemble international diversification but in practice are highly dependent on the U.S. market and leverage/inverse products, exposing investors to elevated volatility and uncertainty. In terms of performance, overall returns across domestic and foreign assets were generally weak relative to the broader equity market over the same period. After accounting for trading costs, more investors incurred losses than realized gains, and only a small minority meaningfully outperformed benchmark indices. While many participants in overseas markets did see improvements in portfolio returns and risk-adjusted performance, roughly half of them still failed to achieve satisfactory outcomes. A quantile analysis of performance drivers reveals strong heterogeneity: for low-performing investors, excessive trading without sufficient experience, concentration in a small number of stocks, and aggressive risk-taking further depress returns, whereas for top performers these same traits are not necessarily harmful and can sometimes be linked to outperformance. These findings indicate that overseas investment can expand diversification opportunities and improve results for some investors, but also embeds structural risks through excessive concentration and frequent trading of high-risk listed products.  Policy-wise, it is notable that relatively stable and favorable outcomes are observed when investors use savings-type accounts to invest long term and consistently in broadly diversified instruments such as plain-vanilla ETPs and overseas equity ETFs. This suggests the need to promote wider use of long-term, diversification-oriented accounts such as IRPs and ISAs, strengthen tax incentives for long-horizon investing, and tighten oversight of product design, disclosure, and sales practices for high-risk products including leverage/inverse ETPs. For young and small-scale investors in particular, expanding tailored financial education and digital, intuitive risk-warning systems is essential. Ultimately, enabling individuals to rely more on pooled investment vehicles and advisory or wealth-management services—rather than bearing complex investment decisions entirely on their own—will support both the financial soundness of households and the sustainable development of Korea’s capital market.  

Artificial intelligence (AI) is driving transformative change across the financial investment industry. The global AI market is projected to grow from USD 189 billion in 2023 to USD 4.8 trillion by 2033, and the financial sector has emerged as one of the most active adopters of AI technologies. In Korea as well, AI-related patent filings in the financial sector have surged rapidly, increasing from 20 cases in 2015 to approximately 380 in 2023—a nineteenfold rise—reflecting accelerating efforts to integrate AI into financial services. As expectations grow that AI will spur innovation across industries, it is essential to understand how AI can actually be applied in the financial investment industry based on objective data. In this regard, patent data serve as a highly valuable analytical resource. Patent data cover a broad range of industries, and each patent document provides detailed information on the data used, AI models employed, and technical specifications. Accordingly, patent analysis can go beyond identifying AI technology trends to forecasting the future development path of the financial industry and deriving meaningful policy implications. This study systematically analyzes the potential applications of AI in the financial investment industry using patent data. The results show that AI functions applicable to financial investment activities have expanded across diverse areas, including data analysis, anomaly and fraud detection, customer service, and report generation, with expected gains in operational efficiency and cost reduction. In particular, B2B startups are actively developing AI-based solutions tailored to financial services, suggesting that financial institutions’ reliance on external technology providers is likely to increase. The applicability of AI varies significantly across business segments within the financial investment industry and across different stages of each segment’s value chain. In repetitive and highly structured tasks, where data are well-organized and procedures are clearly defined, AI adoption is relatively straightforward. In contrast, in tasks that are unstructured and require contextual interpretation and judgment, AI applicability is lower and the expected benefits of adoption are more uncertain. Data accessibility is also a critical determinant of AI adoption: areas where data acquisition, processing, and sharing are smooth tend to exhibit higher concentrations of related patents. Moreover, AI adoption remains limited in high-risk areas, as AI errors in such domains can lead to large-scale financial losses or legal disputes. To overcome these limitations, three key policy directions can be proposed. First, regulatory sandboxes and pilot programs should be expanded to enable experimental AI adoption not only in routine tasks but also in unstructured and high-complexity activities. Such initiatives can support risk management while also generating empirical evidence for future policymaking. Second, data infrastructure should be upgraded at the national level to embed security and regulatory compliance capabilities and to establish support systems that facilitate data collaboration among financial institutions. Third, comprehensive financial AI guidelines should be developed, encompassing AI governance, principles of explainability, and third-party risk management, thereby reducing the psychological and institutional barriers that hinder active AI adoption by financial institutions. While patent analysis provides a forward-looking perspective on the potential integration of AI into the financial investment industry, case studies of overseas financial institutions offer practical insights into the processes of AI internalization and commercialization. These international cases demonstrate that real-world AI adoption is shaped not only by task structure, data accessibility, and risk levels, but also by three additional dimensions: the maturity of technologies and services within the industry, return on investment (ROI), and infrastructure constraints. Accordingly, effective AI adoption in the financial investment industry requires a phased approach that clearly recognizes both the potential and the limitations of AI while carefully weighing expected returns on investment. First, adoption strategies should be formulated based on the maturity of technologies and services within the industry, enabling rapid commercialization in mature areas while managing risk through pilot implementations in more uncertain domains. Second, investment priorities should be set with a strong focus on ROI—concentrating in the short term on areas that offer cost reductions and efficiency gains, while strategically fostering competitive technologies such as risk analysis and personalized advisory services over the long term. Third, continuous investment in data quality and computing infrastructure is essential to enhance AI performance and reliability, ultimately reducing operating costs and enabling stable service delivery. In conclusion, AI has already become a core strategic asset in the financial investment industry. Effective utilization of AI will require coordinated efforts to support AI experimentation by financial institutions, upgrade data infrastructure, establish robust ethical and accountability frameworks, and pursue gradual, practice-oriented diffusion strategies. With such a systematic foundation in place, Korea’s financial investment industry can leverage AI to achieve both innovation and stability, while further strengthening its global competitiveness.

The Korean public fund industry has undergone significant structural changes amid the rapid growth of Exchange-Traded Funds (ETFs). Traditional mutual funds—once the main driver of market expansion—have seen stagnating growth and declining market presence, while ETFs, with their low fees and trading convenience, have surged to surpass public mutual funds in both net assets and market share around 2022. Since then, the overall growth of the collective investment market has been led by ETFs, widening the gap between the two sectors. This transition highlights ETFs’ position as an attractive investment vehicle but also signals the weakening competitiveness of traditional public funds. However, ETFs’ index-linked structure inherently limits portfolio diversity and active management. Even “active ETFs” are typically designed to maintain high correlations with benchmark indices, which constrains their ability to provide meaningful diversification or differentiated performance. In contrast, traditional public funds can actively deviate from benchmarks, engage in stock selection based on unique strategies, and expose investors to distinct sources of risk and return. Such flexibility allows them to pursue portfolio differentiation strategies that align with diverse investor preferences. This study focuses on portfolio differentiation as a sustainable growth strategy for public mutual funds amid intensified competition with ETFs. Its objective is to empirically assess how differentiated portfolio strategies—reflecting each fund’s unique management philosophy and style—affect (1) fund flows (investor subscriptions and redemptions) and (2) the market share of asset management firms. The analysis seeks to objectively evaluate how managerial originality and investment capability translate into performance and competitive outcomes in the industry, providing policy insights for enhancing the competitiveness of Korea’s mutual fund market. Using data from 2014 to 2024, the study quantifies the degree of portfolio differentiation for each fund, based on the overlap between its holdings and comparable ETFs. Differentiation indices are constructed by fund type (equity, bond, and fund-of-funds), and panel regression models are estimated to analyze the impact of differentiation on subsequent fund inflows and changes in asset manager market share. Control variables include fund returns, fees, commissions, and fund and firm size, while dynamic panel models (GMM estimators) are employed to address endogeneity concerns and ensure robustness. The main findings are as follows. First, greater portfolio differentiation significantly enhances fund inflows: funds with lower overlap with ETF holdings attract higher levels of new investment. Second, differentiation also contributes positively to asset managers’ market share growth. Firms that maintain distinctive portfolios—rather than replicating ETF-style compositions—show statistically significant increases in market share over time. The policy implications are clear. Promoting portfolio differentiation can strengthen the competitiveness of public mutual funds and yield positive spillover effects across the financial ecosystem. For investors, differentiation expands investment options and improves utility through better risk-return matching. For the industry, it mitigates excessive price competition and concentration in a few large firms, fostering a healthier, more diverse market structure.

Since the 2008 global financial crisis, central banks around the world have expanded their policy domains beyond the traditional mandate of price stability to encompass broader socio-economic issues such as climate change, digital currencies, and financial inclusion. This transformation demonstrates central banks' evolution into multifaceted policy institutions, but it has also raised new concerns. In particular, following the outbreak of the COVID-19 pandemic, critics argued that central banks failed to contain inflation partly because their policy capacity had become diluted as they began to act more like broad governmental agencies. Based on these concerns, this study analyzes 9,802 speeches delivered by senior officials of 25 central banks during the period 2009–2024, using natural language processing techniques to quantify the phenomenon of policy diversification and to empirically assess its implications for central banks' core mandate of inflation management. The analysis shows that central banks' policy priorities have evolved in response to changing economic circumstances and institutional environments. In the immediate aftermath of the financial crisis, speeches on financial stability accounted for nearly half of the total (excluding regional issues). However, after the global economy stabilized around 2017, references to climate change and digital currencies increased rapidly. Interestingly, differentiated priorities emerged across central banks, with the ECB focusing on climate change while the Federal Reserve concentrated on financial inclusion. Nevertheless, during the high inflation period following COVID-19, central banks returned to their core mandate, with references to monetary policy rising sharply once again. Analyzing how these shifts in policy attention affect price stability, we find that policy concentration measured using a theoretically-informed classification of nine topic categories significantly reduces inflation over the medium term (5-10 quarters). The effect of policy concentration begins to show statistical significance from the fifth quarter, reaches its peak around the seventh quarter (approximately -2.6 percentage points), and gradually diminishes thereafter. This suggests that excessive policy diversification may hinder achieving price stability objectives. Particularly in high-inflation environments, the impact of policy concentration emerges with a notable lag. No significant effect is observed for the initial six quarters, but disinflationary effects begin to appear from the seventh quarter and reach their maximum impact by the eighth quarter. This delayed response reflects the persistence of inflation expectations and the time required for central banks to rebuild credibility in high-inflation regimes. When inflation expectations have become entrenched, economic agents do not immediately trust new signals from central banks. Only when consistent and focused messages are maintained over a sustained period does credibility gradually accumulate, leading to a re-anchoring of inflation expectations. These findings underscore that the effectiveness of central bank communication is highly state-dependent and requires differentiated strategies tailored to prevailing inflation environments. In high-inflation periods, central banks must deliver strong, repetitive, and consistent messages on price stability to restore credibility—a process that requires a long-term perspective and sustained commitment. At the same time, central bank policy diversification cannot be understood simply as 'mandate drift'. In many countries, structural reforms have stalled and political actors have tended to avoid difficult decisions involving conflict resolution, thereby placing excessive expectations on central banks as independent and technically competent institutions. Moreover, when conventional monetary policy instruments are severely constrained, central banks themselves have incentives to engage more directly with structural issues, often leading to an unintended expansion of their policy domain. In conclusion, this study shows that central bank policy diversification has unfolded within an institutional context that makes such expansion almost inevitable. Nevertheless, central banks must continue to uphold a clear prioritization of their core mandate of price stability, even as they respond to diverse societal demands. To this end, carefully tailored communication strategies suited to prevailing economic conditions are essential.

This study examines the characteristics and market relevance of a Capital Market Sentiment Index (CMSI) constructed by fine-tuned transformer-based large language model (LLM) on a large corpus of Korean economic and financial news articles. By utilizing data from January 2010 to March 2025, the CMSI co-moves significantly with major global sentiment indices, Korean equity market returns, option-implied measures of investor fear and hedging demand, speculative demand for derivative-based ETFs, and key indicators of equity market trading activity and IPO dynamics. In particular, the CMSI is strongly associated with individual investor trading and with IPO first-day premiums—suggesting that shifts in sentiment propagate more prominently through noise traders and firms operating under high information asymmetry. Analyses of the relationship between sentiment and stock returns find that while the CMSI exhibits limited medium-horizon predictability for aggregate stock market returns, it meaningfully forecasts short-horizon market movements in the same direction as contemporaneous sentiment shifts. Cross-sectionally, optimistic (pessimistic) sentiment is associated with overpricing (underpricing), especially among firms that are hard-to-value or exhibit strong lottery-like features. These results indicate that the CMSI contains economically meaningful information for explaining the cross-section of stock returns. Overall, this study provides empirical evidence on the validity and usefulness of an LLM-based sentiment index in the context of equity markets. As the use of unstructured data and advanced NLP methods becomes increasingly important in financial research, these findings illustrate that LLMs can serve as an effective tool for capturing nuanced, latent investor sentiment embedded in textual information.

This study provides a comprehensive discussion of the “Capital Market Sentiment Index (CMSI),” regarding its conceptual background, construction methodology, statistical properties, and potential applications. As macroeconomic uncertainty and market volatility continue to rise, the need to incorporate unstructured information alongside conventional metrics has become increasingly important. In response, the index is designed to quantitatively measure the qualitative judgments, emotions, expectations, and concerns of capital market participants. This study provides a detailed explanation of the method to construct an index that quantifies market participants’ perspectives by training a transformer-based, general-purpose large language model (LLM) on Korean news data. The significance of this study lies in its enhancement of the transparency and explainability of an index built using an LLM. By providing detailed explanations of the news text preprocessing procedures and learning algorithms, the study strengthens model’s transparency and reproducibility. Furthermore, through Shapley value analysis, outputs are decomposed into the contribution of each input word, enabling quantitative interpretation. Sentiment measured at the sentence-level is used to construct multi-layered index time series based on daily averages and variances, and application strategies are proposed based on their statistical characteristics. The Capital Market Sentiment Index is valuable in supporting evidence-based decision-making. In particular, as a high-frequency indicator derived from unstructured textual information, it offers strong use case for real-time monitoring of investor sentiment and for evaluating policy effects. The model developed in this study can serve as a foundation for future research aimed at advancing the index framework through the introduction of multidimensional classification schemes and the application of multimodal models.