Can matching by company name tie the shares a data vendor's list left out to their filings?
Study 44 · Changed a rule or plan · recorded 8 October 2026 · answer: Partly · full paper with figures
The question
Study twenty-six tied only about half of 2,517 shares to their filings with the market regulator, because the vendor's list as first read covered just 1,310. We test, without that list, whether exact company-name matches find the right filer, and whether with the list they tie four in five shares in each period from 2014 to 2022.
Rules fixed before the result
The shares, the check, the reading rules and our expected answer were fixed before anything was counted. Name matching failed if over 5% of its accepted links could be wrong; otherwise each period counted if 80% of its shares were tied, with at least 200 shares needed to test. The full stored list, judged by the earlier study's rule, would become the next link if it passed and agreed with name matching on 95% of at least 200 shares.
Data and method
Of 2,517 development shares with a filing window in 2014-2022, the vendor's list as first read covered 1,310; of the rest, 1,033 had a single exact company-name match among the regulator's filers. A match was right if its filer filed a form marking a listing's start or end near the share's own, judged against mismatched pairs over 1,000 reshuffles. We also read the vendor's full stored list; every figure is a count over a fixed population, not a sample estimate.
The answer
Partly. Name matching passed its accuracy check: of 914 shares whose listing began or ended inside the window, about five in six had their name-matched filer file a form marking that start or end near the date, against about three in a hundred mismatched pairs, so at most about 0.6% of accepted links can be wrong, inside the 5% limit fixed in advance. With the vendor's list as first read, it tied at least four in five shares in 2017-2019 (81%) and 2020-2022 (85%), but 77% in 2014-2016, short of the bar. The vendor's full stored list, including its batches of shares no longer listed, tied 97% to 98% in every period and agreed with name matching on 99% of shares both tie. The plan expected this answer with low confidence, and the full list's result as most likely.
What it changes
By the rule fixed in advance, questions about company filings are now planned on the vendor's full stored list over the whole development sample, with name matching and its check kept as an independent cross-check. The earlier study's shortfall came from what had been read, not from what the vendor holds, so no purchase of further data is proposed.
What remains uncertain
The vendor's filer number carries no date, so a company that changed filer is tied only from the change; exact names miss renamed companies. Shares listed throughout the window, lacking a start or end to test, went unchecked. Any filing counts as a sign of life, not the event a later study needs; the 0.6% bound assumes wrong links file near listing dates no more often than random mismatched pairs; study twenty-six studied these shares before, in one development sample.
How it was made
Run took 2 minutes; independent code review: not recorded.
Fingerprints
plan 8f08104864a0f7c4c6c3042c7c710adb5552c0bdf08d88b3ad1293791a4e799c result not recorded
Exploratory work on development data; not a performance claim.