A Long-Short Portfolio Loss for Listwise Stock Ranking

Summary

The document summarizes research on a listwise learning-to-rank loss designed for constructing long-short stock portfolios. The proposed loss builds on ListMLE but gives extra emphasis to the top and bottom of the ranking, where long and short selections are made. It is described as inherently shift-invariant, and the paper supplies a probabilistic interpretation through a generalized Plackett-Luce model. The approach uses multiple factors to predict cross-sectional stock returns and rank equities.

The summary reports an empirical study using 68 factors from China’s A-share market over 2006–2019, with stated out-of-sample annual return of 38% and Sharpe ratio of 2. Those figures are reported claims in this short summary; it does not provide the portfolio construction details, transaction-cost assumptions, benchmark comparisons, or robustness tests needed to assess them. The underlying paper is identified as a 2021 publication, but the document itself is only a repeated abstract and provides no further methodological evidence.

Key ideas

• The method learns cross-sectional stock rankings from multiple factors for long-short portfolio construction.
• Its listwise loss places greater weight on the top and bottom of the ranking than standard ListMLE.
• The loss is described as shift-invariant and linked to a generalized Plackett-Luce probability model.
• The reported study uses 68 factors from China A-shares over 2006–2019.
• The summary reports out-of-sample return and Sharpe figures but omits costs, benchmarks, and robustness details.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.