CrossSectionCN
A China A-share companion to the Open Source Asset Pricing project
This site re-tests the documented cross-sectional return predictor literature against China A-share data. The intent is descriptive rather than prescriptive: each predictor is rebuilt under China-specific timing, disclosure, and listing-board constraints, and the resulting series, metrics, and known data limitations are published for inspection.
Scope and Relationship to Prior Work
The documentation structure follows Open Source Asset Pricing (Chen and Zimmermann, 2022) and the OpenSourceAP/CrossSection reference codebase: each signal carries a definition, literature attribution, expected return direction, portfolio construction rule, and a record of known data issues. None of the original U.S.-market results are republished here.
The narrower question is which documented predictors survive when rebuilt on China A-share data, and which adaptations are needed to make them well-defined locally. The site therefore separates imported predictors, China-adapted definitions, and any China-native signals as they are added. It is an independent companion experiment, not an extension of the original project.
Build Summary
- Factors documented: 94
- Supported factors: 87
- Factors with return series: 87
- Return months: 196
- Coverage rows tracked: 87
Research Contract
The current research contract is deliberately close to the academic anomaly convention:
- Signals are sampled at each stock’s last available trading day in a calendar month.
- Portfolios use next-month returns.
- Long-short returns are direction-adjusted using
factor_doc.csv. - Raw daily data remains the source for rolling-window signals such as momentum, reversal, volatility, maximum return, turnover, and illiquidity.
In notation, the published return series is
R^{LS}_{f,t+1} = \begin{cases} R^{High}_{f,t+1} - R^{Low}_{f,t+1}, & \text{if higher signal values predict higher returns} \\ R^{Low}_{f,t+1} - R^{High}_{f,t+1}, & \text{if lower signal values predict higher returns.} \end{cases}
Current Caveats
- Accounting ratios still need production filters for negative equity, tiny denominators, and restatement-sensitive fields.
- Suspension and relisting gaps can create extreme adjusted daily returns for momentum and lottery-style signals.
- Current returns are equal-weight research spreads before trading costs, capacity constraints, and full investability filters.
Factor Families
Each family below links to its rollup page (overlay chart + side-by-side metrics). The full list of factors lives in the Factor Library; per-factor return statistics are in Factor Returns.
| Family | Supported factors | Avg Sharpe | Avg cumulative |
|---|---|---|---|
| Earnings Quality | 7 | 0.29 | 9.29 |
| Financing | 1 | 0.13 | 1.11 |
| Investment | 4 | 0.01 | 0.95 |
| Leverage | 10 | -0.08 | 0.82 |
| Liquidity | 14 | 0.42 | 7.06 |
| Momentum | 10 | 0.11 | 1.18 |
| Profitability | 20 | 0.43 | 2.01 |
| Reversal | 3 | 0.76 | 4.16 |
| Risk | 10 | 0.42 | 2.42 |
| Shareholder Yield | 1 | 0.08 | 1.05 |
| Size | 2 | 1.23 | 35.72 |
| Valuation | 1 | 0.04 | 0.82 |
| Value | 4 | 0.24 | 1.51 |