Mining Short-Horizon Equity Factors and Matching Rebalance Frequency to Costs

Summary

This study mines short-horizon price and volume factors for Chinese equities, then examines how transaction costs and rebalance frequency affect portfolio results. It uses genetic programming to generate candidate daily factors from conventional price and volume data as well as features derived from high-frequency trades, including buyer and seller activity. Correlation screening reduces the candidates to a pool of 17 factors, which are evaluated individually and combined into a composite stock-selection signal.

The reported tests find that factor information decays quickly, with most estimated half-lives near five days. The study recommends adjusting rebalance frequency to assumed costs: daily for low costs, every two days for moderate costs, and lower-frequency approaches when costs are high. In its 2010 to mid-2019 sample, a two-day portfolio with bilateral costs of 0.3% had strong reported returns relative to the CSI 500, but also high turnover and substantial drawdown. These are historical model results; data mining, changing market conditions, and implementation costs may limit their relevance to live trading.

Key ideas

• Genetic programming can generate candidate short-horizon factors from price, volume, and high-frequency trade features.
• Correlation screening was used to reduce redundant candidates to a 17-factor pool.
• The reported factors decay quickly, so rebalance frequency should reflect signal decay and trading costs.
• The tested portfolio had high turnover, and its historical results may not persist out of sample.

Tags

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