Firm characteristics help estimate conditional factor loadings. Larger libraries contain new and overlapping information. We use the cross-sectional distribution of firm characteristics as a ruler to evaluate IPCA loading spaces. With factor counts fixed, expanding from 36 to 153 characteristics can improve in-sample fit while representing momentum and size positions less fully. Reallocating shared exposures changes attribution without changing joint fitted returns. With coefficients fixed, exposure references alter holdings, trading, and subsequent risk at the same characteristic target. Statistical factor models should be evaluated not only by return fit, but also by the exposures they represent and the investment decisions they support.

Traditional portfolio optimization typically relies on a single fitted return distribution, thus ignores non-normality and uncertainty across distributional states. Mean–variance–ambiguity (MVA) models, motivated by the smooth-ambiguity preferences of Klibanoff, Marinacci, and Mukerji (2005) and the KMM-to-MVA Arrow–Pratt reduction of Maccheroni, Marinacci, and Ruffino (2013), extend conventional risk penalties to account for distributional uncertainty, but they missed how state-conditional return distributions affect portfolio optimization stability. We develop Deep Mean–Variance–Ambiguity (Deep MVA) to resolve this gap by learning state-conditional return distributions and embedding the resulting conditional information loss into the ambiguity penalty. The resulting local certainty equivalent combines reduced-law mean–variance risk with a quadratic penalty for second-order disagreement. Conditional-mixture state resolution yields a Deep ambiguity covariance equal to the Traditional ambiguity covariance plus a positive-semidefinite state-feature increment. This increment is the conditional L2 loss from averaging state-specific distributions before optimization. On compact convex feasible sets, we show that the state-resolving program has stronger ambiguity-relevant curvature, lower mean-input sensitivity, and a unique global solution under restricted positive curvature or a common ridge; its projected iterations converge linearly. An LSTM-MDN learns state-conditional distribution densities from 60-minute bars, and Probability and VoV scores enter weekly MVA, with GARCH_Resid as a comparator. Tests on the SSE 50, CSI 300, and CSI 500 universes from 2017–2025 find statistically significant Sharpe out-of-sample improvements over Traditional MVA and MVO across all three universes and ambiguity specifications.

Can fully agentic artificial intelligence predict equity returns in real time? We build a novel dataset of daily AI-generated attractiveness scores spanning the entire Russell 1000 universe, continuously updated since April 2025. Each trading day, the model freely searches the live internet and produces quantitative scores for every constituent stock before the next market open, eliminating both look-ahead bias and attribution bias. A daily value-weighted portfolio of the 20 highest-ranked stocks earns a Fama-French six-factor alpha of 16.5 basis points per day (41.5% per annum), with an average bid-ask spread of 1.92 basis points. Predictability is sharply asymmetric: the model reliably identifies future outperformers but not future underperformers. The evidence points to stock picking rather than market timing: performance is strongest on down-market days, and the alpha does not come from betting on aggregate market direction. The AI’s reasoning traces help explain this predictability. Alpha is concentrated in traces with richer live-web evidence and stronger analytical structure. In the theme attribution, most active alpha reflects firm selection under an investment lens, not broad theme choice.

We examine whether the characteristics of economically linked firms predict stock returns in the Chinese A-share market. We construct two peer-based signals: a peeraverage (PA) signal that aggregates the characteristics of fundamentally similar firms, and a peer-deviation (PD) signal that summarizes a firm’s characteristic deviations from industry-peer averages. PA captures shared fundamental information, whereas PD benchmarks a firm’s relative position against economically comparable peers. Portfolio sorts, factor-spanning tests, and Fama–MacBeth regressions reveal economically meaningful and statistically significant return predictability for both signals, with generally larger portfolio returns among non-state-owned enterprises. Both signals retain significant incremental predictive power after controlling for the full set of own-firm characteristics and flexible machine-learning forecasts based solely on these characteristics. Changes in both signals also predict future earnings, while return predictability is generally stronger among firms receiving less investor attention, consistent with limited-attention-induced underreaction. These findings highlight the distinct informational and benchmarking roles of economic networks and their predictive value beyond firms’ own characteristics.

This study investigates the relationship between the number of stock concepts and future returns, as well as the economic mechanisms underlying this association. Using novel collected data, we find that stocks with a greater number of concepts earn significantly higher returns in the subsequent month, generating a six-factor adjusted annualized alpha of approximately 9.6% for a long-short portfolio. Although these stocks exhibit higher turnover, return volatility, and investor attention - features commonly associated with speculative concept-driven trading - an analysis of cross-listed AH twin stocks reveals that concept counts do not widen the AH premium. Moreover, the return premium persists for up to 15 months without significant reversal, whereas firms that engage in opportunistic concept-chasing exhibit pronounced long-run reversals, suggesting that the positive concept-return relation is not driven by speculative motives. The higher returns are primarily attributable to strong industrial policy support and sustained above-expectation operating performance. Firms with more concepts are more likely to receive government subsidies and deliver positive earnings surprises, effects that are amplified when their core concepts receive stronger national policy backing. In contrast, firms that pursue concepts lacking substantive business relevance are significantly less likely to obtain government subsidies and fail to achieve above-expectation growth. Regarding investor composition, institutional ownership of high-concept stocks increases modestly, while ownership by government-guided funds rises substantially - particularly for stocks whose core concepts are strongly supported by national industrial policies. Conversely, concept-chasing behavior by listed firms significantly reduces the ownership ratio of government-guided funds. Overall, our findings indicate that concept stocks in China’s capital market are not purely speculative but signal underlying national policies.

This paper examines whether subjective macroeconomic expectations predict equity risk premia in China. Using professional macro forecasts from the Wind database over 2011- 2023, we construct a composite macroeconomic expectation index via PLS. We find that the index significantly and positively predicts future equity premia both in-sample and out-ofsample, implying a countercyclical equity premium. The predictive power is robust to a wide range of macro-financial controls and delivers economically meaningful gains in portfolio allocation. Further analysis shows that controlling for forecast errors does not attenuate predictability, and the subjective index largely subsumes the information in realized macro variables, suggesting that it captures forward-looking macroeconomic conditions rather than belief bias. Cross-sectional and long-horizon results reinforce this interpretation.

We propose a clustering-based dimensionality reduction approach to minimum variance portfolio optimization. Rather than constructing the global minimum variance (GMV) portfolio over the full stock universe, which is subject to severe estimation error due to high-dimensionality, we apply Ward’s hierarchical clustering to partition stocks into groups of similarly behaving assets, select one representative per cluster, and optimize on the resulting low-dimensional sub-universe. We show theoretically that clustering preserves the factor structure and yields a better-conditioned covariance matrix than random selection. Empirically, on the Chinese A-share market, the proposed strategies substantially outperform the full-universe benchmark, with gains robust to transaction costs.

Financial crises rarely erupt without warning; they are preceded by long periods of hidden fragility. Yet traditional market indicators such as the VIX capture only realized volatility, offering little foresight. This paper introduces the Stability Gap Model (SGM), developed iteratively from a simple intuition: fragility arises when risk-taking diverges from fundamentals and systemic buffers are insufficient. We trace the evolution of the model from its original formulation, through corrections and extensions, to its present form. The Classic SGM captures instantaneous imbalance, while the Beta SGM incorporates memory of past shocks. Empirical analysis demonstrates that the SGM provided clear early warnings ahead of the 2008 Global Financial Crisis, the 2011 Eurozone debt episode, the 2015 China/oil slowdown, and the 2018 tightening cycle, while also trending upward in 2019 before the COVID-19 crash. Furthermore, the paper demonstrates the model's utility in assessing fragility in hedge funds and proposes its application as a universal framework for stability analysis across diverse systems, from corporate finance to supply chains.

Investors’ prior experiences with a stock substantially affect their willingness to repurchase it. This paper explores the repurchase effect, a psychological bias in which investors are reluctant to repurchase stocks that have appreciated after a prior sale. To quantify this bias, we develop a novel stock-level measure, termed Repur, and investigate its implications for cross-sectional asset pricing. Our findings show that stocks with higher Repur tend to experience reduced future buying pressure from investors, which in turn results in lower subsequent returns. Economically, long-short portfolios based on Repur yield annualized abnormal returns exceeding 23% for equal-weighted and 11% for value-weighted risk-adjusted returns. Further analyses show that the pricing effect of Repur is more pronounced following periods of high investor sentiment, for stocks with greater arbitrage constraints, and for firms with smaller investor bases. Out-of-sample evidence from China confirms the significant pricing impact of the repurchase effect.

This paper investigates whether firm-level freight activity captures corporate fundamentals and predicts stock returns. Using smartphone geolocation data of truck drivers from 2019 to 2024, we construct a novel freight growth index (FGI) to quantify firms’ freight activity in the Chinese stock market. We find that firms’ freight growth is strongly associated with current operating performance and predicts future stock returns. A long-short portfolio sorted on FGI generates significant risk-adjusted monthly returns ranging from 57 to 72 basis points. Further evidence suggests that freight growth forecasts earnings announcement returns and is more predictive for firms with low information transparency. Moreover, freight growth provides incremental information in predicting stock returns beyond analyst forecasts. Our findings highlight that firm-level freight activity contains novel insights into firm fundamentals and stock pricing.