This paper examines whether the variables associated with Bitcoin returns are stable or change as the market matures. Using LASSO regression on 38 candidate variables covering Bitcoin-specific activity, public attention, macroeconomic conditions and cross-asset returns, we compare Bitcoin's earlier market-development regime (January 2012 to July 2021) with the subsequent market-transition period (July 2021 to February 2026). Before 2021, Bitcoin returns exhibit relatively weak associations with network activity and regulatory attention. After 2021, U.S. equity returns become the strongest and most consistently selected association: Bitcoin's equity beta more than doubles, equity returns are the only variable retained under the most conservative selection rules, and Bitcoin co-moves with equities on rising as well as falling days. Structural-break tests on the return model locate the shift within 2020–22, and the increase in the equity beta is robust to an alternative price alignment. Lagged specifications show no reliable predictive content, so these are contemporaneous associations. The results indicate that Bitcoin's return associations are regime-dependent and that Bitcoin has become increasingly integrated with traditional financial markets.
This study examines the non-linear association between AI-related disclosure and stock price crash risk. Using 5158 Chinese A-share listed firms from 2009 to 2023, we construct a firm-level measure of AI-related disclosure intensity from the Management Discussion and Analysis sections of annual reports. We interpret this text-based measure as a composite signal that may reflect both substantive AI-related activities and strategic disclosure around AI concepts, rather than as direct evidence of actual AI deployment. The results reveal a significant inverted U-shaped association: AI-related disclosure is positively associated with crash risk at relatively low levels, whereas the association weakens and becomes negative at higher levels. This pattern remains robust to a series of sensitivity tests and alternative specifications. Further information-channel tests suggest that the non-linear association is accompanied by changes in abnormal tone, analyst forecast errors, textual similarity, and analyst forecast optimism bias. Heterogeneity analyses show that the inverted U-shaped association is more pronounced among firms with higher financing constraints, weaker digital foundations, and lower internal control quality. Overall, the findings highlight the boundary between symbolic AI disclosure and genuine technological empowerment, and contribute to the literature on AI-related disclosure, strategic narrative management, and capital market risk.
Traditional global risk indicators (e.g., VIX) are limited in capturing multi-dimensional, forward-looking narrative risk. While authoritative macro reports provide such information, traditional text analysis techniques have faced challenges in systematically extracting and quantifying these complex narratives. This pre-registered study proposes a novel methodology to apply Large Language Models (LLMs) to quantify narrative risk from these reports, constructing a comprehensive system of Global Risk Metrics, decomposing narrative risks into growth, financial, and policy dimensions. Leveraging a dual-layer dataset comprising micro-level fund flows (EPFR) and macro-level aggregate capital movements (BOP) from 2002 to 2024, the empirical design will test whether these multidimensional risk indicators offer incremental predictive power for cross-border capital flows beyond established market, uncertainty, and quantitative benchmarks. We will further examine heterogeneous effects across country characteristics (such as financial openness, exchange rate flexibility, and political stability), investor risk aversion, and fund strategies.
Financing frictions and agency conflicts can generate different forms of inefficient investment, making the allocation of capital as important as access to finance. This study examines how supply chain finance (SCF), which draws on trade relationships and transaction information, is associated with corporate investment efficiency. Using Chinese A-share listed firms from 2015 to 2024, we estimate fixed-effects models with a residual-based measure of investment inefficiency. SCF is significantly negatively associated with investment inefficiency, with the association concentrated in investment shortfalls; the over-investment estimate is statistically insignificant. Channel regressions link SCF to lower financing constraints, more stable operating profitability, and greater research and development (R&D) expenditure. The association becomes more negative as supplier and customer concentration increases, and significant negative estimates are observed among non-state-owned firms, firms with higher financing constraints, and firms with concentrated suppliers. These results connect trade-related financing with investment shortfalls and highlight the relevance of firms' financing conditions and supply chain relationships to capital allocation.
This paper studies pricing errors in China's rapidly expanding convertible bond (CB) market. Unique market features—such as conversion price reset clauses, retail-investor dominance, and lagging regulatory norms—create substantial valuation challenges. We develop an improved pricing model using the least squares Monte Carlo method to capture key path-dependent contractual terms. Based on model-implied values, we find significant and left-skewed pricing errors, indicating widespread overvaluation, especially during conversion periods and for CBs linked to short-sellable stocks. Further evidence shows that mispricing is mainly driven by noise trading from retail investors, amplified by investor sentiment and heterogeneous beliefs. Institutional investors and major shareholders provide only limited correction. These results highlight how investor structure and trading system shape pricing efficiency and offer implications for reducing speculation and improving market regulation in China's CB market.
