Combining High-Frequency Stock Factors in a CSI 500 Enhancement Portfolio

Summary

This weekly Chinese equity research note reviews long-short screens built from high-frequency and technical factors. The listed signals include return skewness, downside volatility share, opening-period buying interest and large-order flows, reversal, late-session trading, outflow size, and price moves attributed to large orders. It also covers four GRU and neural-network factor variants. Most factor examples use monthly rebalancing, while the learned factors are described with weekly rebalancing; the note says conventional factors were orthogonalized but does not include the underlying charts or their numerical results in the supplied text.

For a CSI 500 benchmark enhancement portfolio, the method combines turnover, reversal, volatility, intraday flow, and deep-learning signals. It estimates factor premia by cross-sectional regression, uses their trailing 12-month average to forecast next-period premia, and combines those estimates with current factor values. Portfolio construction maximizes expected excess return under benchmark-weight, exposure, industry-neutrality, and turnover constraints, assuming next-day average-price execution and stated trading costs. The note flags factor failure and liquidity risks; the missing figures limit assessment of performance.

Key ideas

• The note surveys high-frequency factors based on return distribution, volatility, intraday trading flows, reversal, and large-order activity.
• It describes four GRU and neural-network variants evaluated as weekly rebalanced long-short signals.
• The CSI 500 enhancement portfolio forecasts factor premia using cross-sectional regressions and trailing averages.
• Portfolio optimization constrains benchmark weights, factor exposures, industry deviations, and turnover.
• Performance charts are referenced but absent from the supplied text, and factor failure and liquidity are stated risks.

Tags

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