Vendor Equity Data: Check Coverage
Which vendor data filters can silently change a historical equity universe?
Audit identifiers, delistings, splits, duplicate rows and vendor revisions before selecting a historical equity universe. Record each rule and its effect on available securities; a cleaned current constituent list can remove failed firms from the past.
The paper concerns the studied Datastream and CRSP panels; the proposed row checks are a manual workflow and are not paper-verified universal deletion rules or available app nodes.
Evidence map
| Aspect | Finding |
|---|---|
| What it is | Ince and Porter compare individual US equity returns from Thomson Datastream with the Center for Research in Security Prices (CRSP). |
| What problem it answers | A clean-looking return table can silently change its universe. |
| Practical check | Obtain the vendor's field definitions, identifier history, corporate-action policy, exchange coverage and delisting treatment. |
What it is
They use that comparison to assess whether Datastream can support studies of large equity universes outside the United States. Their reported issues concern coverage, classification and data integrity. Naive use of the vendor file changes economic inferences in their tests. After their screening, inferences from the two US databases become more similar. They then apply their screens to data from four European equity markets. This is evidence about the datasets and research designs they examined, not an estimate of the error rate in every vendor feed.
What problem it answers
Securities can be absent, misclassified or represented by observations whose provenance needs checking. A strategy that selects stocks by size, liquidity or return history can therefore be sensitive to which identifiers entered the file and when they entered. A result from one database should be accompanied by a coverage and classification audit before treating a difference from another database as a market effect. The paper establishes that these checks can materially change inference; it does not establish that a particular row in a user's file is wrong.
Practical check
For a dated sample of securities, compare identifiers and observation windows against an independent primary record where access permits. Count unmatched securities by year and by the characteristic used for selection. Inspect large return reversals, long unchanged-price runs, duplicates and currency changes as flags for investigation. Those example flags are a proposed audit workflow here; this page does not attribute that exact checklist or a universal deletion threshold to Ince and Porter. Record every corrected row and the rule applied, then recompute the original analysis both before and after correction. Report which difference came from the universe, the classification or an individual value.
For a Stochastly study
A user-supplied bar file needs the same provenance discipline before a backtest is interpreted. Export a dated inventory of files and instruments, document the adjustments made outside the application, and compare results under the original and reviewed inputs. These steps describe a manual research protocol. They do not claim that a vendor-equity audit, CRSP comparison or automatic screening node is available in the application. If an independent reference series is unavailable, say so and limit the claim to checks that were actually performed.
Source
Ince & Porter, "Individual Equity Return Data from Thomson Datastream: Handle with Care!", Journal of Financial Research 29(4), 2006, 463-479. DOI 10.1111/j.1475-6803.2006.00189.x. Primary source