325Trades
54%Win rate
+1.21%Avg per trade
1.65Profit factor
+£1,967Profit, £500 a trade
−£317Max drawdown
When the bosses buy
Markets
Insider trading
Clusters of insiders buying their own company’s shares have been followed by three-day gains since 2021. Fresh data will show whether the edge survives
Company insiders in America must report their trades in their own firm’s shares to the Securities and Exchange Commission, on a filing known as Form 4, within two business days. Most of these filings say little: executives sell to diversify, pay tax or buy a house. Purchases are rarer and more telling, because there is only one obvious reason to buy. The question tested here is narrower: when several insiders buy on the same day, with real money, does the share price keep rising afterwards?
Over nearly five years of filings, it did, modestly and after costs. A cluster of insider purchases, bought at the next open and sold three days later, made an average of 1.2% a trade.
The rules
- Signal: a Form 4 filing date on which two or more different insiders made open-market purchases of common stock, totalling at least $500,000. Purchases through company plans (employee share schemes, 401(k)s, dividend reinvestment) do not count.
- No funds: the whole cluster is dropped if any buyer is a fund or company, or owns 10% or more of the stock. This was the biggest single improvement: those clusters made just 0.18% a trade.
- Size: market value between $300m and $5bn on the filing date.
- Entry: buy at the next day’s open after the filing. Share price at least $1.
- Exit: sell at the close of the third trading day (the entry day counts as day one).
Each trade is £500, with Interactive Brokers’ tiered commissions and a modelled bid-ask spread deducted.
Results
The equity curve is not smooth, and nor should it be. Most trades are small; the profit comes from a steady excess of modest winners over modest losers, plus the occasional large jump. Stripping out the best 1% of trades still leaves a profit of about £1,450.
| Year | Trades | Avg return | Win rate | P/L |
|---|---|---|---|---|
| 2021 | 27 | +0.26% | 52% | £35 |
| 2022 | 80 | +1.92% | 61% | £768 |
| 2023 | 50 | +1.67% | 54% | £419 |
| 2024 | 64 | +1.52% | 53% | £484 |
| 2025 | 68 | +0.47% | 53% | £161 |
| 2026 H1 | 36 | +0.57% | 44% | £101 |
The strategy made money in all six years, but the edge has thinned. Trades before 2025 averaged 1.55%; since then, 0.51%. That may be noise in a smaller sample, or it may be the signal being arbitraged away as insider-tracking tools multiply. The honest answer is not yet known.
What did not help
At least 16 variants were tested on the same data, which is itself a warning (see below). The ones that failed are as instructive as the one that worked:
- Routine versus first-time buyers. Almost no routine buyers appear in $500,000 clusters (10 of 635), so the distinction barely matters.
- Buying beaten-down shares. Clusters in stocks down 20% or more over three months averaged 2% a trade overall, but roughly nothing since 2025. Not used as a filter.
- Dropping private placements. Removing purchases made in company offerings lowered returns, so they stay in.
- Neighbouring settings. Thresholds of $250,000 or $1m, other size bands and two- or four-day holds were all positive. The result does not hinge on one lucky parameter.
The caveats
Three matter most. First, the bid-ask spread is modelled (15, 8 or 5 basis points a side, depending on company size), not measured from real quotes. Second, testing many variants on one dataset inflates the chance of finding something by luck; the real test is data the model has never seen, which arrives with the SEC’s filings for the third quarter of 2026. Third, trade size matters: at £100 a trade, Interactive Brokers’ minimum commission eats most of the edge. The strategy needs at least £250 a position, and works best at £500.
It is therefore promising, not proven. It will be paper-traded before any real money goes near it.
Hypothetical backtest, September 2021 to June 2026. Not investment advice. Past performance is not a reliable guide to future results. See the disclaimer.