Methodology
Everything below is a stated assumption, not a proven fact. The weights are a starting hypothesis to be tested against history, and the backtest in the open-source worker exists to do exactly that.
Sources
United States: SEC Form 4 filings, both the live EDGAR feed and the quarterly bulk datasets covering 2006 onwards. Europe: managers’ transaction notifications filed under the Market Abuse Regulation, beginning with the Swedish register, which covers July 2016 onwards. Regulators publish these records without prior review and state plainly that they cannot guarantee accuracy.
The filter
Only open-market purchases survive. In US filings that is transaction code P. Europe has no standardised code, so the wording of the declared nature is tested, with disposals excluded first so a sentence containing both words cannot pass as a purchase. Volumes expressed as a nominal amount, used for debt instruments, are dropped because they are not comparable to a share count.
The signal
| Rule | Points |
|---|---|
| Chief financial officer | +30 |
| Chief executive officer | +25 |
| Other executive officer | +15 |
| Ten percent shareholder | +10 |
| Board member | +8 |
| Purchase above 1M | +30 |
| Purchase above 250K | +22 |
| Purchase above 100K | +14 |
| More than half of existing position | +20 |
| Four or more insiders in the same window | +20 |
| Market capitalisation under 500M | +8 |
| Market capitalisation above 10B | −10 |
| Pre-scheduled plan | −25 |
The scheduled-plan penalty rests on published work separating opportunistic insiders from routine ones, which finds the abnormal return concentrated almost entirely in the opportunistic group. A trade decided months in advance describes a calendar.
Measuring what happened next
- Entry at the first close available on or after the publication date, never the transaction date.
- Returns compared with a market index over the same window, otherwise you are measuring the market rather than the filter.
- Positions whose horizon has not elapsed are excluded, not carried forward to today. Otherwise recent buys inflate the result.
- No fees and no slippage are simulated.
Known limits
Filing deadlines mean the transaction is always older than the disclosure. Spouse and trust accounts appear under the manager’s name. Delisted companies drop out of the measurement, which biases results upward. European filings carry an ISIN rather than a ticker, and an unmapped listing cannot be measured. Below ten measured purchases, an individual track record is mostly noise.