Thompson Sampling for Adaptive Multi-Factor Investing

Summary

The report describes applying Thompson Sampling, an online learning method, to adjust a multi-factor investment model as market leadership changes. It uses China A-share constituents from the CSI 300 and CSI 500, with valuation, profitability, growth, momentum, reversal, volatility, liquidity, and size factors. The authors use stratified sampling to reduce industry effects and examine size-factor performance across historical periods.

The report compares buy-and-hold, scheduled allocation changes, Greedy, Epsilon-Greedy, and Thompson Sampling. It says Thompson Sampling led on the reported measures and adapted during the 2017 reversal in size-factor performance. A stock-and-bond mix is also presented as an example of adapting to cyclical asset rotation. These are historical findings reported by the authors; the provided text gives no detailed data, test design, transaction costs, or out-of-sample validation. It suggests the approach may suit cyclical or range-bound markets, while momentum-oriented methods may do better in strong trends, so the conclusions should not be treated as a guarantee of future performance.

Key ideas

• Thompson Sampling is used as an online method for adapting factor allocations as market conditions change.
• The model draws on eight factor groups and targets constituents of two major Chinese equity indices.
• The report compares Thompson Sampling with passive, scheduled, and other exploration strategies.
• The authors report that Thompson Sampling handled a historical size-factor reversal and cyclical asset rotation well.
• The text provides limited information on validation, implementation costs, and robustness beyond the reported historical tests.

Tags

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