A Structured Factor Risk Model for Chinese A-Shares

Summary

The report presents DFQ-2018, a structured risk model for Chinese A-shares. It combines industry and style factors to identify exposures, estimate stock-return covariance, and support portfolio performance analysis. Among its design choices, it uses a state-owned-enterprise indicator as a partial proxy for policy risk, company characteristics such as analyst coverage and fund ownership to represent information uncertainty, and Bayesian shrinkage to improve beta estimates.

For covariance estimation, the report describes robust regression, Newey–West adjustments, Bayesian estimation of residual covariance, and GARCH-based variance adjustment. It reports average monthly cross-sectional adjusted R-squared values of 30.2% for CSI 300 constituents, 14.1% for CSI 500 constituents, and 17% for the broader market, and says its minimum-variance portfolios had lower volatility than those from a purely statistical model. The source warns that model failure and extreme markets remain risks; changing risk models also requires retuning portfolio risk aversion.

Key ideas

• The model combines industry and style factors to measure equity risk and estimate return covariance.
• It uses company attributes as proxies for policy risk and information uncertainty, and shrinks beta estimates toward a prior.
• Robust regression, Newey–West adjustment, Bayesian residual covariance estimation, and GARCH variance adjustment are among its refinements.
• Reported explanatory power varies across CSI 300, CSI 500, and broad-market stock universes.
• Portfolio risk settings may need recalibration when the risk model changes, and extreme markets can undermine model performance.

Tags

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