We assess the predictive power of machine learning (ML) models for forecasting realised volatility using information from HAR model variables, limit order book (LOB) data, and news sentiment. Training and robustness checks on nearly seven million ML models show that high-dimensional ML models outperform HAR models in 90% of the out-of-sample period, except during extreme volatility. Explainable AI analysis identifies mid prices, mean bids, and mean asks as key predictors. Notably, incorporating ML into ensemble frameworks enhances HAR model performance, though caution is needed when using ML models as direct substitutes, since they may yield unreliable forecasts under certain market conditions.
This paper studies heterogeneity in flow-hedging behavior among active mutual funds. While recent evidence shows that the aggregate mutual fund industry hedges against common flow risk by tilting toward low-flow-beta stocks, I find that nearly half of U.S. active equity funds tilt their portfolios toward high-flow-beta stocks. To rationalize this finding, I propose an information-based model in which managers with more precise private signals about future flow shocks perceive lower uncertainty and hedge less aggressively. Empirically, funds with higher flow risk exposure are associated with proxies for informational advantages, and their flows predict aggregate industry and common flows. Moreover, funds that hedge against flow risk the least outperform those that hedge the most over 3% per year, suggesting that hedging against flow beta is costly.
This study reveals the dual impact of political connections within supply chains. While prior studies emphasize that political connections benefit directly connected firms, we show that they impose significant costs on their suppliers. Using close outcomes of US congressional elections as plausibly exogenous shocks to firm-level political connections, we find that firms gaining political connections earn positive abnormal returns around election dates, whereas suppliers with greater exposure to these connected firms suffer significant stock price declines. This adverse effect is especially pronounced when suppliers are highly dependent, lack sufficient countervailing power, or when connected firms face low switching costs. Further evidence indicates that connected firms gain value by strategically restructuring their supplier networks and renegotiating contracts to their advantage, leaving dependent suppliers with worse trade terms and squeezed profit margins, consistent with intensified bargaining pressures in competitive buyer–supplier relationships. Political connections thus generate negative spillovers within supply chains, redistributing value between connected firms and their suppliers.
We identify soft information acquisition by mutual fund managers based on detailed data on investor company visits in China and study the use of soft information in portfolio management. We find a clear divergence in fund managers’ preference for soft information in investment decisions. “Soft-information” managers hold fewer stocks and tend to invest in companies with high growth potential and significant idiosyncratic risk. Trades driven by soft information acquisition are profitable, resulting in superior performance by these managers, especially in their holdings of stocks rich with soft information. Fund managers’ distinct preferences for soft versus hard information create segmentation in both information acquisition and portfolio choice.
We investigate the reallocation of retail investor attention during firm-specific disruptive events, such as IPOs. While IPOs increase investor attention across listed firms, those geographically and industrially close to the IPO attract the most significant interest. Attention builds up prior to the IPO, peaks during the event, and remains elevated afterward, particularly for geographically proximate firms. Notably, geographic proximity plays a dominant and persistent role, even under firm-specific and market-wide uncertainty. Attention also responds systematically to information, increasing with negative and uncertain tone in IPO prospectuses, consistent with the negativity bias. Conversely, high levels of aggregate market attention divert attention away from IPO events, although this effect is attenuated for geographically proximate firms. We further show that heightened attention is associated with a positive contemporaneous return effect followed by a reversal over longer horizons, consistent with temporary price pressure. Our findings indicate that retail investor attention is allocated in a systematic and economically meaningful way, reflecting the role of local information advantages, social and geographic proximity.
Housing is a central household asset that shapes economic behavior. Using China Family Panel Survey data, we document a substitution effect between housing wealth and the value of sons as an alternative form of household security. A 10% increase in homeowners’ housing wealth raises the number of newborn daughters within the subsequent two years by 7.4% relative to the mean, with no corresponding effect on sons. Various mechanism tests indicate that households obtaining higher returns from the housing market become less sensitive to the gender difference in children’s expected returns. This insensitivity reduces the incentive for sex selection, leading to the birth of more girls. These findings offer a novel asset-market explanation for the decline in the sex ratio at birth in urban China and underscore the demographic consequences of wealth shocks.
This paper introduces an option-based, market-level signal derived from individual stocks' Implied Variance Asymmetry (IVA) using the Three-Pass Regression Filter (3PRF), which we refer to as IVA-3PRF. We find that IVA-3PRF is significantly positively related to future stock market returns: a one-standard deviation increase in IVA-3PRF predicts a 0.88% increase in market returns in the following month. Out-of-sample forecasts based on IVA-3PRF achieve an R2 of 3.79% (5.34%) using a minimum training window of five (ten) years. We show that the predictive power of IVA-3PRF arises from its ability to forecast future changes in economic conditions and, consequently, future cash-flow news. Importantly, the signal is most informative during pessimistic periods, when option investors' expectations appear more rational. This conditional rationality is driven by heightened investor attention to macroeconomic conditions in such times. By contrast, in optimistic periods, elevated sentiment induces overconfidence and biased expectations, reducing the signal's predictive power.
This paper evaluates bootstrap simulation techniques for calculating the distribution of the maximum drawdown (MDD), an important risk indicator. Using stochastic dominance tests, we examine the complete distribution properties of the MDD in both the stock and cryptocurrency markets. The standard Efron (1979) bootstrap method, which assumes that the random variables are independent and identically distributed, systematical ly underestimates the true MDD. While the moving block bootstrap provides reasonable estimates, it is subject to non-stationarity bias, particularly when large drawdowns occur at the boundaries of a return series. The stationary bootstrap of Politis and Romano (1994) produces the most accurate and robust results, especially for longer block lengths. Alternative procedures, such as the block-block bootstrap, the tapered bootstrap and robust resampling, do not lead to better results.
We construct global currency volatility risk as the equal-weighted average of currency optionimplied or realized volatility across 17 major currencies and find that it significantly and positively predicts currency returns, especially for horizons beyond three months. The in-sample R2 statistics are 2.49%-5.02% (2.64%-7.13%) for global currency option-implied (realized) volatility in univariate regressions. They also perform well for out-of-sample cases, with R2 statistics of 3.00%-6.84% (3.53%-10.06%) for global currency option-implied (realized) volatility. Additionally, the currency return predictability systematically increases with rising inflation risk. More interestingly, global currency volatility risk is closely related to U.S. economic activity, uncertainty, the VIX, and sentiment.
This study investigates whether sovereign credit rating downgrades increase corporate default probabilities. Employing a difference-in-differences (DID) design, we compare firms rated at or above their sovereign (bound firms) with those rated below (non-bound firms). Our findings reveal that bound firms exhibit a significantly higher increase in default probability following a sovereign downgrade. This effect is amplified in firms with weaker financial fundamentals and in countries with underdeveloped banking sectors or internal economic crises. The results underscore the pivotal role of sovereign ceilings in shaping corporate creditworthiness and provide novel insights into the spillover effects of sovereign risk on the private sector.
This study examines how artificial intelligence (AI) adoption is associated with corporate investment efficiency among U.S. firms. We measure AI adoption using AI mentions, defined as the proportion of AI-related terminology in senior management remarks during earnings calls, and validate this measure by showing that it is positively associated with 10-K discussions of AI applications in operational decision-making, product development, and business expansion. We construct an instrumental variable based on exposure to university-authored AI research—measured as a normalized score of textual similarity between non-corporate academic publications and industry descriptions—to address concerns that AI adoption may be endogenous to firms’ investment decisions and investment efficiency. Our baseline results indicate that higher AI adoption is associated with greater investment efficiency, and this association remains in two-stage least squares regressions using the AI technology exposure instrument. We also find that firms with higher AI mentions tend to have higher Tobin’s q over longer horizons, consistent with a gradual reflection of AI-related benefits in firm value. Additional analyses suggest several mechanisms: AI adoption is associated with more accurate management sales forecasts, higher financial reporting quality, and greater process and product innovation intensity. Cross-sectional evidence further indicates that regulatory frictions, such as exposure to data-privacy regulation, attenuate the association between AI adoption and investment efficiency. Overall, the findings suggest that AI adoption is an emerging factor in corporate investment and capital allocation.
The main purpose of this paper is to revive the behavioral explanation of the long-term return reversal anomaly based on overreaction and arbitrage asymmetry. We demonstrate that: (i) winners (losers) with higher (lower) mispricing rankings experience stronger and more persistent reversals; (ii) the reversal pattern of winners (losers) is stronger following periods of optimism (pessimism), and (iii) because of arbitrage asymmetry, winners experience stronger reversals than losers. This behavioral evidence remains strong after controlling for a comprehensive list of competing hypotheses built on risk or rationality-based arguments.
We develop a machine learning framework that uses technical indicators derived from international stock data to forecast the performance of international equity ETFs. To address the limited history of ETFs, we train a random forest model on global stock data and apply it to rank ETFs according to the probability of outperformance. Portfolios formed on this ranking are associated with economically meaningful gross-of-cost return spreads, averaging 0.76% per month (t-stat = 2.76) and 0.63% after market risk adjustment. Predictability is strongest at short horizons and decays with longer holding periods. Volatility and momentum indicators contribute the most to model performance. In line with limits to arbitrage, predictive strength is more pronounced in less efficient markets and among lower-liquidity ETFs. Results are robust across portfolio construction methods and alternative models, and the signal performs well out-of-sample from 2011 to 2022, with extended evidence up to 2024.
