FiClaw
-
Financial Agent Platform
中文

A-share research checklist

A-share factor research checklist

Use this before the backtest. It makes the data timestamp, investable universe, execution assumptions and validation record explicit; it does not imply that a strategy will be profitable.

Minimum record

Keep these five records with every study

Record the data version, universe rules, factor treatment, trading assumptions and validation result. That is the shortest path to explaining why two research runs differ.

CHECK 01

Data availability

Confirm when every field could have been known.

  • Apply financial data after actual announcement dates.
  • Use implemented dividends and documented ex-dividend conventions.
  • Version index membership, free float and sector classifications historically.

CHECK 02

Universe and corporate events

Make inclusion and event rules reproducible.

  • Document listing age, ST, suspension, delisting and missing-data rules.
  • Flag mergers, restructurings, asset sales and restatements.
  • State whether delisted stocks remain in the historical sample.

CHECK 03

Execution and costs

Model whether the portfolio can be traded.

  • Model price limits, suspensions and one-word boards.
  • Record fees, taxes, slippage, impact and participation limits.
  • Check turnover and capacity against the selected universe.

CHECK 04

Factor processing and exposures

Version every transformation from field to score.

  • Record missing-value, winsorisation, standardisation and ranking rules.
  • Review sector, size, beta, liquidity and valuation exposures.
  • Test individual factors before explaining a composite score.

CHECK 05

Validation and reporting

Use several diagnostics rather than one equity curve.

  • Report Rank IC, grouped returns, drawdown, turnover, cost and coverage.
  • Review market regimes and rolling out-of-sample periods.
  • Record failed windows and unresolved assumptions alongside positive results.

Take the checklist into selection and backtesting

Choose a small factor set from the research question, then carry these conventions into the prompt, specification and report. Review results against out-of-sample evidence and independent human judgment.

Open factor selectorBack to factor library