1. Three Hundred Years of Monthly Return Predictability: A Comprehensive Examination, Bai (2022)

  2. Download

    Presentations: FMA (Scheduled), University of Missouri (2022)

    Abstract: I search for return predictability that is consistent throughout time periods and across markets. Of the 22 predictors, only return on long-term bond [ltr], term spread [tms], past month equity premium [lag1], moving average (2,9) [ma_2_9], moving average (2,12) [ma_2_12] show consistent in-sample predictability with the 300-year UK data, but nothing shows consistent in-sample predictability with the 160-year US data. Although some predictors show predictability with their full availability in one market, the out-of-sample performance is not consistent. With the UK data, only lag1 shows consistent out-of-sample performance throughout time, but the trading turnover of the strategy based on lag1 is extreme. No predictor provides out-of-sample performance that is consistent throughout time with the US data. The cross-country tests show worse results in the UK market with US parameters but interestingly improved results in the US market with UK parameters. ma_2_9 survives the cross-country test in the US and even remains significant for the last 50 years. The long history does not support consistent return predictability. However, some predictors can be helpful to investors' portfolio allocation.

    UK 50-Year Out-of-Sample R² In 300 Years Yang Bai (yangbai@mail.missouri.edu)

    UK 20-Year Cumulative Return Ratio In 300 Years
    (Predictor Portfolio/Null Portfolio) Yang Bai (yangbai@mail.missouri.edu)

  3. Machine Learning Classification Methods and Portfolio Allocation: An Examination of Market Efficiency, Bai and Pukthuanthong (2020)

  4. Download

    Presentations: AFA Ph.D. Poster Session (2021), University of Missouri (2020), Crowell Prize 2020 Seminar (2021), University of Miami Winter Conference on Machine Learning and Business (2021), World Finance Conference (2021),SFA (2021)

    Awards: Crowell Prize 2020 (Third Prize), PanAgora Asset Management

    Abstract: We frame the asset pricing problem as a machine learning classification problem. The predictions on 3.34 million observations yield significant out-of-sample economic gains. Through directly measured accuracies, binomial tests suggest that the classifiers can extract forward-looking contents from historical information, implying imperfect information efficiency. The classifiers exploit the differences in return state transition uncertainties. As reflected by a pre-realization measure based on multi-class predicted probabilities, the classifiers are more confirmative in predicting high-trading-friction stocks. Consistently, only trading frictions contribute to out-of-sample predictability throughout 26,302 distinct stocks’ lifetimes. The adjustment of the classifiers’ favorance over certain return states increases the performance.

    10-Minute Audio Introduction:

  5. Females Look Innocent in the Eyes of SEC, Bai and Guo (2022)

  6. Download

    Abstract: We investigate whether SEC enforcement and private investigations are affected by differential enforcement for firms with gender-diverse board. Consistent with prior evidence in the legal literature regarding disparities in police and judicial enforcement, a one-standard-deviation increase in the percentage of females on board will reduce SEC enforcement risk by 0.3%, which is 27.3% of average SEC enforcement risk. We show that the reduced SEC enforcement is driven by issues in the SEC’s decision to initiate investigations. Compared to a board with less female directors, a board with more than 30% female directors is 1.2% less likely to be subject to investigation by the SEC. Subsequent analysis fails to support a reverse causality conjecture that female directors can reduce financial abnormalities.