Build the universe that existed, not the one that survived

The universe is part of the strategy

A stock-selection rule operates on an opportunity set. Defining that set with today's listings or current index constituents gives the research information it would not have possessed historically. Companies that failed, merged, moved exchange or became too small can disappear from the sample. The resulting bias is not repaired by delaying an otherwise correctly timestamped signal.

CRSP's stock-database documentation describes active and inactive securities, permanent identifiers and corporate-action information. Those are relevant data capabilities, not a certification that any particular extract is unbiased. An analyst still needs to implement membership dates, security-type filters and eligibility decisions correctly. A database can contain delisted stocks while a query accidentally excludes every one of them.

Measure the effect of one missing failure

Imagine four fictional stocks, each assigned $100 at the start of a single period. Their ending values are $110, $105, $90 and zero. Total wealth falls from $400 to $305, a return of -23.75%. If the last stock is removed because it no longer exists at the end, the surviving three show $305 against $300, or approximately +1.67%.

This is not an estimate of typical survivorship bias. It is a deliberately transparent example of changing the denominator using future information. An actual delisting need not imply a total loss: a merger may deliver cash or stock, while a failed company may leave an uncertain recovery. Model the event's economic proceeds and dates rather than assigning the same outcome to every disappearance.

Store eligibility as dated evidence

Use a stable security identifier for prices and a separate issuer identifier for company-level exposures. Tickers can change or be reused, and multiple share classes can represent one issuer. Maintain dated links between accounting entities and traded securities. A many-to-many join that duplicates a company's earnings across several price records can distort both factor rankings and concentration limits.

Evaluate market capitalization, liquidity and price filters using information available before each formation decision. Record both eligible and excluded records with reason codes. If a rule requires a minimum trading history, a newly listed security becomes eligible only after satisfying it. Do not retroactively grant eligibility because a later download contains a long enough series.

Test the exclusions as carefully as the holdings

Reconcile constituent counts at every stage: raw securities, allowed types, sufficient history, valid features and final portfolio. A sudden count change should have an explanation such as a source-format migration or a declared rule change. Compare excluded companies by size and sector to identify systematic missingness. Publish those coverage limits rather than calling a partial sample the entire equity market.

Before interpreting results, trace a few difficult cases from original records through terminal proceeds: a ticker change, a merger and a delisting. Retain uncertain values as unresolved or apply a predeclared sensitivity treatment. Never fill a missing terminal price with the last quote indefinitely. A defensible universe is a dated decision history; a list of recognizable stock symbols is not.

Inspect the example

Sources & reproduction

Primary references support the definitions and method. Numerical examples are illustrative unless a source record is explicitly identified.

• Morningstar / CRSP: US Stock Databases

Continue the review