Working Papers
Embedded Files
Add Headings and they will appear in your table of contents.
Disentangling Market Timing from Stock Picking---A Machine Learning Based Approach (with Michael Gallmeyer and Andrea Lu)
Presentations: FMA Annual Meeting (2023) , Goizueta Doctoral Research Conference (2023), University of Georgia (2023), Georgia Institute of Technology (2023), FMA European 2022, FMA Asia/Pacific 2022, AsianFA 2022, SFA 2022, AFBC 2021
Traditional performance attribution approaches may attribute mutual fund market timing to stock picking due to new risk factors, characteristics, and industry rotations. Our theoretical framework shows that successful timing (picking) strategies entail buying stocks with high future systematic (idiosyncratic) returns. Therefore the covariance between fund holding weights and future stock systematic (idiosyncratic) returns measures timing (picking) performance. Our regression tree approach accurately distinguishes systematic from idiosyncratic returns, accommodating complexities in the return structure. Novel to the literature, funds have significant and persistent timing and picking skills. By buying past winners, investors can achieve an annual risk-adjusted return of 2.93%.
Presentations: Emory University (2022)
This paper studies the partitioning of stocks into groups with distinctive expected returns based on ex-ante firm characteristics, which can be used as comparable groups to compute the abnormal part of returns, that is, UNexpected returns. In order for stock expected returns to be similar within groups and disperse across groups, I introduce a methodology to select characteristics that best distinguish expected returns, and cutoffs points where returns are most sensitive to the underlying characteristics. I show that: 1) the combination of chosen characteristics changes over time; 2) fewer fund managers are identified to be stock pickers once the time-variation in comparable groups is incorporated; 3) and the resulting portfolios exhibit desirable properties as basis assets.
Interesting Interactions Between Size and BM
When BM and size are considered jointly, interesting patterns arise: when BM is low, the relationship between size and returns is increasing and concave; and when BM is high, the relationship between size and returns becomes decreasing and convex. The right panel of the figure, representing regression tree-based comparable groups efficiently picked up this pattern while the ones sorted with even breakpoints do not.
Failure Mimicking Portfolios (with Neal Galpin)
Presentations: Monash University (2023*)
Regressing a constant on a set of excess returns gives portfolio weights for the minimum-variance stochastic discount factor (SDF). We show that discounting returns by a given model, then applying the same procedure to these discounted returns gives loadings to mimic SDF errors. We compare these "failure mimicking portfolios" for leading consumption-based asset pricing models. Models that use some market price data or future consumption perform well overall and in both large- and micro-cap samples but not in mid-cap strategies. We show a handful of strategies survive all the consumption-based models.
Which Portfolios are Most Important to Asset Pricing? (with Neal Galpin and Lin Wu)
Presentations: The University of Melbourne (2017), Monash University (2017*)
We estimate a non-parametric stochastic discount factor (SDF) from a set of portfolios, then test whether excluding a portfolio changes the implied SDF. Though related to traditional asset pricing tests, our approach has several advantages: we test all portfolios jointly and can incorporate trading costs easily. We show four portfolios provide independent information about the SDF after accounting for trading costs: the Market and Profitability factors, an Investment-based portfolio, and the Value-Momentum-Profitability anomaly portfolio. The remaining portfolios are redundant. We show both the joint testing and transaction cost adjustments are important for inference, and provide a simple way to implement our tests.
Work in Progress
Human-specific Components and Company Value in the Age of AI: Evidence Based on Textual Analysis
Factor, Price Thyself (with Neal Galpin)
Page updated
Google Sites
Report abuse