Every activist screen produces the same shape of output, which is a list of filings with a flag on them. The flag is true or false. Downstream of that flag, most desks treat every campaign as the same object, which means a filer who has run twenty campaigns and settled fifteen of them with board seats carries exactly the same weight as a filer whose first 13D this is. That is a modelling choice, and it is a bad one, but it is not the screen's fault. The screen is reporting a form type and a stated intent, which is all the filing contains.
Turning that flag into something with information in it means building a record yourself, per filer, and the work is more about definitions than about data.
The tag is a form type and a stated intent
Start with what the filing actually asserts. Schedule 13D is filed by a beneficial owner crossing the five per cent threshold who is not eligible for, or not electing, the short-form route. Schedule 13G is the short form, available to qualified institutions and to passive holders who certify they are not holding with the purpose of influencing control. The distinction is a certification about purpose, made by the filer, at the time of filing.
So the activist label is a statement of intent, and intent is not skill. It is not even commitment. A 13D can be filed by somebody who will run a two-year campaign and win board representation, or by somebody who will drift below five per cent in four months having done nothing, and on the day of the initial filing those two look identical.
Two mechanical points matter for the record you are about to build. The amendment chain is where the campaign lives. Material changes to the position or the purpose require an amended filing, so the sequence of amendments against an initial 13D is your event timeline, and the only public one you get. And the deadlines for both schedules have moved with recent rulemaking, so check the deadline in force at the time of each filing rather than applying today's window to an old one. Getting that wrong biases any measure of filing promptness.
Be clear about what the overlay does and does not give you. It cross-references Schedule 13D and 13G filings and tender offers against insider buying, and it filters by form and by stated intent. I have found nothing in it that ranks filers or carries a track record column, so the record described below is a file you build and maintain yourself. The overlay is the event source, not the scoring layer.
The unit of observation is the campaign, not the filing
The most common way this fails is at the first step, where somebody builds a table keyed on filings. Filings are not comparable units. One campaign generates an initial filing and anywhere from zero to a dozen amendments, so a filing-keyed table silently overweights long campaigns and noisy filers.
Define the campaign as the tuple of filer and issuer, opening at the initial 13D and closing at whichever of three events comes first. A filing reporting the position has fallen below the reporting threshold. A filing reporting a settlement or standstill. Or a corporate outcome that ends the question, such as a completed sale of the company. If none of those has happened, the campaign is open and it stays out of the outcome statistics while still counting in exposure statistics. Treating open campaigns as failures is the single most common source of a pessimistic bias in these records.
One filer running two campaigns against the same issuer years apart is two rows. A group filing jointly is one row, and the group identity needs a rule that survives members joining and leaving, or your sample fragments.
The four fields that make a prior worth having
Outcome, classified rather than scored. Do not define a win by returns. Returns fold in market direction, sector direction and your own entry timing, and they will make the ranking unstable. Classify into a small set of mutually exclusive outcomes that are readable off the filings and the issuer's own disclosures. Board representation obtained. Settlement with governance changes and no seats. Strategic review or sale announced. Position exited without any observable change. Campaign still open. Five buckets is enough, and the discipline of forcing every campaign into exactly one is where most of the learning happens.
Duration. Median days from the initial filing to the closing event, computed separately per outcome bucket. Median rather than mean, because a couple of multi-year campaigns will otherwise dominate. Duration is the field that most directly informs your own holding period if you intend to trade alongside these events, and it is usually longer than desks expect.
Terms conceded. When there is a settlement, record what was actually given. Number of seats, whether they are independent designees or the filer's own people, the length of any standstill, and whether the standstill was later extended. This is the field that separates a filer who extracts real governance change from one who accepts a cosmetic settlement to exit gracefully, and it is available in the filings and the issuer's announcements without any inference about anyone's conduct.
Position economics. Size at initiation as a percentage of shares outstanding, and where you can establish it, the campaign as a share of the filer's disclosed book. A filer putting a fifth of their assets into one campaign behaves differently from one putting in a hundredth.
Turning a record into a weight without overfitting
Now the part where most of these projects go wrong, because the samples are small and the temptation to rank precisely is strong.
Count first. Many filers who look prolific have five or six closed campaigns. A win rate computed on six observations carries a confidence interval wide enough to swallow the ranking, and presenting it to two decimal places is a way of lying to your own investment committee. Set a minimum closed-campaign count below which a filer gets the pooled base rate, and state that threshold in the documentation.
Above the threshold, shrink toward the base rate rather than using the raw rate. The mechanics are standard and the parameter is a judgement, but the principle is not optional. Segment before you shrink, because outcomes vary by target size and sector, and a filer who only targets small caps belongs against the small cap base rate.
Then use the output as a tier rather than a score. Three or four tiers, assigned in a scheduled review, written down. Tiers invite an argument about which bucket a filer belongs in, and that argument resolves in a meeting and gets minuted. Scores invite an argument about weights, which never resolves.
The bias that will destroy the backtest if you let it
Every field above is known only after the campaign closes, and the campaign closes long after the filing you would have traded on. If you rank a filer today using outcomes that resolved after a historical trade date and then test a strategy on that historical date, you have written a look-ahead into the core of the model and it will produce a beautiful result that does not exist.
The fix is structural, not procedural. Store every field with the date on which it became publicly knowable, and make the ranking a function of a date. When the research layer asks for a filer's tier, it asks for the tier as of a date, and it gets the tier computed only from campaigns closed before that date. Filers with too few closed campaigns as of that date get the base rate, which means most filers get the base rate in the early part of any sample. That is not a defect in the method, it is the truth about how much you knew at the time.
Two other biases belong in the documentation because a good allocator will ask. Survivorship, in that accumulations abandoned before crossing five per cent never become filings, so you are measuring only the intentions that reached a threshold. And regime, in that the defensive toolkit available to targets and the disclosure rules themselves have both moved across any decade-long sample, so an early record is a prior about a game with different rules. Neither is fixable. Both are statable, and stating them is the difference between a record you can defend in review and one that falls apart the first time somebody asks how the sample was built.