Multi-Factor Stock Selection
A-share multi-factor stock selection: from signal combination to portfolio constraints
A multi-factor strategy should not be a pile of indicators. Each signal needs an economic rationale and demonstrable incremental information before it enters a combination with clear exposure and trading constraints.
Typical holding period
Often biweekly or monthly, depending on refresh cadence and turnover budget.
Rebalance and trigger
Use a fixed schedule and recompute coverage, weights and constraints each period.
Research universe
A stated investable universe with listing-age, liquidity and abnormal-status rules.
Research hypothesis
Write the hypothesis before reading the backtest
Factors with different economic sources may complement each other. Highly correlated factors or shared industry and style exposures can amplify, rather than diversify, risk.
This guide is for research education and backtest design. It is not investment advice or a return promise; results depend on sample, cost, regime and execution.
Strategy rules
Turn the idea into zero-discretion rules
Factor normalisation
Winsorise, standardise and neutralise each factor as specified.
Makes signals with different units comparable.
Composite score
Combine value, quality, growth and momentum through a pre-set or rolling rule.
Converts multiple signals to a reviewable ranking.
Portfolio constraints
Apply sector, name and turnover limits to targets.
Keeps the score from overriding implementation boundaries.
Implementation
From signal to portfolio
- 1Store raw values, processed values and coverage for every factor.
- 2Start with an interpretable equal-weight blend before testing more complex weights.
- 3Check factor correlation and exposure before construction to avoid duplicated bets.
- 4Translate target weights into constrained orders with a stated unfilled-order rule.
A-share constraints
Execution belongs in the test
- Financial, price and event inputs become available at different times and cannot be synchronised by assumption.
- Micro-cap effects can be capacity and liquidity effects; use trading-value and impact limits.
- Extended fields such as margin data need independent coverage and publication-time checks.
Validation checklist
Try to break the strategy first
Prefer stable regions that survive costs, regime changes and parameter perturbation to a single maximum return.
Risk map
How the strategy can fail
Research prompt
A first backtest specification
“In CSI 500 constituents, test value, quality, growth and momentum. Respect each field's actual availability, normalise within sector, begin with equal weights, rebalance monthly and apply name, sector and turnover limits. Output single-factor and composite IC, correlation, risk exposures, cost sensitivity and rolling out-of-sample results.”
FAQ
Common questions
Are more factors always better?
No. An additional factor needs data quality, economic interpretation and out-of-sample incremental value. Highly correlated factors can simply repeat the same exposure.
Can historical IC set dynamic weights?
It can be researched, but it adds model freedom. Use rolling and out-of-sample tests to see whether it is chasing recent noise.
Next research step
Fix research conventions before the backtest
Use the related factors, data boundaries and out-of-sample review to turn a strategy idea into an auditable research record.