This paper develops an autonomous framework for systematic factor investing via agentic AI. Rather than relying on sequential manual prompts, our approach operationalizes the model as a self-directed engine that endogenously formulates interpretable trading signals. To mitigate data snooping biases, this closed-loop system imposes strict empirical discipline through out-of-sample validation and economic rationale requirements. Applying this methodology to the U.S. equity market, we document that long-short portfolios formed on the simple linear combination of signals deliver an annualized Sharpe ratio of 2.75 and a return of 54.81%. Finally, our empirics demonstrate that self-evolving AI offers a scalable and interpretable paradigm.

Can fully agentic AI nowcast stock returns? We deploy a state-of-the-art Large Language Model to evaluate the attractiveness of each Russell 1000 stock each trading day, starting in April 2025 when AI web interfaces enabled real-time search. Our data contribution is unique along three dimensions. First, the nowcasting framework is completely out-of-sample and free of look-ahead bias by construction: predictions are collected at the current edge of time, ensuring the AI has no knowledge of future outcomes. Second, this temporal design is irreproducible once the information environment passes. Third, our framework is fully agentic: we do not feed the model curated news or disclosures; it autonomously searches the web, filters sources, and synthesises information into quantitative predictions. We find that AI possesses genuine stock-selection ability, but that its predictive power is concentrated in identifying future winners. A daily value-weighted portfolio of the 20 highestranked stocks earns a Fama-French five-factor plus momentum alpha of 19.4 basis points and an annualised Sharpe ratio of 2.68 over April 2025–March 2026. The same portfolio accumulates roughly 49.0% cumulative return, versus 21.2% for the Russell 1000 benchmark. The strategy is economically implementable: the average bid-ask spread of the daily Top-20 portfolio is 1.79 basis points, less than 10% of gross daily alpha. However, the signal remains asymmetric. Bottom-ranked portfolios generally exhibit alphas close to zero, while the strongest predictive content sits in the extreme top ranks. Delayed-entry tests further show that predictability does not vanish after a single day; rather, the signal remains positive over a broad window of subsequent entry dates, consistent with slow information diffusion rather than a fleeting overnight anomaly.

We introduce a real-time dispersion measure based on cross-sectional stock returns explicitly designed to capture retail-driven speculative episodes. Elevated return dispersion effectively identifies periods characterized by intensified retail investor trading behaviors, driven by salience, diagnostic expectations, and extrapolative beliefs. During these high-dispersion states, momentum strategies collapse, and short-term reversals become dominant. Conditioning momentum strategies on our dispersion measure resolves the longstanding puzzle of missing momentum in retail-intensive markets such as China, substantially enhancing profitability. A dynamic rotation strategy between momentum and short-term reversal portfolios guided by dispersion states achieves annualized Sharpe ratios nearly double those of static approaches. Extending our analysis internationally, we employ Google search trends as proxies for retail investor attention, confirming that dispersion robustly predicts momentum and reversal returns globally. Our findings underscore the behavioral channel through which retail-driven speculation conditions momentum dynamics, providing clear implications for dynamic portfolio management strategies.

We develop a deep-visualization framework for timing the factor zoo. Historical factor return trajectories are converted to two complementary image representations, which are then learned by convolutional neural networks (CNNs) to generate factor-specific timing signals. Using 206 equity factors, our CNN-based forecasts deliver significant economic gains: timed factors earn an average annualized alpha of about 6\%, and a high-minus-low strategy yields an annualized Sharpe ratio of 1.22. The outperformance is robust to transaction costs, post-publication decay, and factor category-level analysis. Interpretability analyses reveal that CNNs extract predictive signals from path boundaries and regime shifts, capturing patterns orthogonal to investor attention.

We examine the risk-return trade-off in market anomalies within the A-share market, showing that even decaying anomalies may proxy for latent risk factors. To balance forecast bias and variance, we integrate the 1/N and mean-variance frameworks, minimizing out-of-sample forecast error. Treating anomalies as tradable assets, we construct optimized long-short portfolios with strong performance: an average annualized Sharpe ratio of 1.56 and a certainty-equivalent return of 29.4% for a meanvariance investor. These premiums persist post-publication and are largely driven by liquidity risk exposures. Our results remain robust to market frictions, including shortsale constraints and transaction costs. We conclude that even decaying market anomalies may reflect priced risk premia rather than mere mispricing. This research provides practical guidance for academics and investors in return predictability and asset allocation, especially in the unique context of the Chinese A-share market.

We examine the risk-return trade-off in market anomalies within the A-share market, showing that even decaying anomalies may proxy for latent risk factors. To balance forecast bias and variance, we integrate the 1/N and mean-variance frameworks, minimizing out-of-sample forecast error. Treating anomalies as tradable assets, we construct optimized long-short portfolios with strong performance: an average annualized Sharpe ratio of 1.56 and a certainty-equivalent return of 29.4% for a mean-variance investor. These premiums persist post-publication and are largely driven by liquidity risk exposures. Our results remain robust to market frictions, including short-sale constraints and transaction costs. We conclude that even decaying market anomalies may reflect priced risk premia rather than mere mispricing. This research provides practical guidance for academics and investors in return predictability and asset allocation, especially in the unique context of the Chinese A-share market.

We utilize the Bayesian model scan approach to examine the best performing models in a set of 15 factors discovered in the literature, plus principal components (PCs) of anomalies unexplained by the initial factors in the Chinese A-share market. The Bayesian comparison of approximately eight million models shows that HML, MOM, IA, EG, PEAD, SMB, VMG,PMO, plus the four PCs, PC1, PC6, PC7, PC8 are the best supported specification in terms of marginal likelihoods and posterior model probabilities. We also find that the best model outperforms existing factor models in terms of pricing tests and out-of-sample Sharpe ratio.

This paper proposes the concepts of ESG information flows and a predictable framework of ESG flows based on AR process, and studies how ESG information flows are incorporated into and affect a dynamic portfolio with transaction costs. Two methods, called the ESG factor model and the ESG preference model, are considered to embed ESG information flows into a dynamic mean-variance model. The dynamic optimal portfolio can be expressed as a traditional optimal portfolio without ESG information and a dynamic ESG preference portfolio, and the impact of ESG information on optimal trading is explicitly analyzed. The rich numerical results show that ESG information can improve the out-of-sample performance, and ESG preference portfolio has the best out-of-sample performance including the net returns, Sharpe ratio and cumulative return of portfolios, and contribute to reducing risk and transaction costs. Our dynamic trading strategy provides valuable insights for sustainable investment both in theory and practice.

Behavioral theories suggest that investor misperceptions and market mispricing will be correlated across firms. We use equity and debt financing to identify common misval- uation across firms. A zero-investment portfolio (UMO, undervalued minus overvalued) built from repurchase and issue firms captures comovement in returns beyond that in some standard multifactor models, and substantially improves the Sharpe ratio of the tangency portfolio. Loadings on UMO incrementally predict the cross-section of returns on both portfolios and individual stocks, even among firms not recently involved in external fi- nancing activities. Further evidence suggests that UMO loadings proxy for the common component of a stock’s misvaluation.

In this paper, we show that only the systematic component of exchange rate expectations of professional investors is a strong predictor of the cross-section of currency returns. The predictability is strong in short and long horizons. The strategy offers significant Sharpe ratios for holding periods of 1 to 12 months, and it is unrelated to existing currency investment strategies, including risk-based currency momentum. The results hold for forecast horizons of 3, 12, and 24 months, and they are robust after accounting for transaction costs. The idiosyncratic component of currency expectations does not contain important information for the cross-section of currency returns. Our strategy is more significant for currencies with low sentiment and it is not driven by volatility and illiquidity. The results are robust when we extract the systematic component of the forecasts using a larger number of predictors.