Big dollar buys are the weakest insider signal we track
A $250,000-plus purchase sounds decisive. Across 9,000 of them it produced no edge in calm markets and a modest one in stress. Dollar size measures the insider's wealth, not their information.
- Trades: every open-market insider purchase reported to the SEC on Form 4 from 2006 onward that passes our routine-versus-opportunistic filter; for this piece, purchase value of at least $250,000.
- Entry: first session after the filing became public. Held 63 trading days (about three months).
- Benchmark: IWM (small caps), because insider-buy candidates skew small and cheap; beating the S&P alone would just be that tilt.
- Market state: stress = SPY at least 10% below its trailing 1-year high on the entry day; calm otherwise.
- Statistics: clustered by company (one vote per company) — five insiders buying the same stock in the same week count once. Cells under 40 trades or 15 companies are shown but greyed. No trading costs deducted.
- What is excluded: companies without a ten-year price history (delisted names are the main gap; see the survivorship note on the evidence page).
Three-month excess return versus small caps for large buy ($250k+)s made during sell-offs, by sector. Sectors with too few trades omitted from the chart, shown in the table.
| Sector (SEC industry group) | In sell-offs | In calm markets | Read |
|---|---|---|---|
| Transport & logistics | +29.2 pts n 58 · t 0.7 | −1.1 pts n 179 · t 0.1 | Too few trades to say |
| Telecom & media | +14.4 pts n 18 · t 1.3 | −2.0 pts n 190 · t -0.6 | Too few trades to say |
| Tech hardware & semis | +12.0 pts n 65 · t 2.0 | −2.1 pts n 228 · t -2.0 | No reliable edge |
| Mining & metals | +10.1 pts n 18 · t 1.6 | −3.6 pts n 51 · t -2.3 | Too few trades to say |
| Software & IT services | +8.5 pts n 73 · t 0.9 | −2.3 pts n 188 · t 0.5 | No reliable edge |
| Banks | +7.9 pts n 199 · t 2.1 | +1.4 pts n 574 · t 1.7 | Sell-offs only |
| Medical devices | +7.8 pts n 40 · t 1.3 | +8.2 pts n 118 · t 3.1 | Too few trades to say |
| Other | +7.7 pts n 105 · t 2.1 | +1.1 pts n 323 · t 0.1 | Sell-offs only |
| Food, drink & tobacco | +7.3 pts n 6 · t 0.6 | +1.8 pts n 111 · t -0.1 | Too few trades to say |
| Retail & consumer | +5.8 pts n 185 · t 2.0 | −0.6 pts n 410 · t 0.9 | No reliable edge |
| Energy | +2.3 pts n 175 · t 1.2 | −2.6 pts n 430 · t -2.4 | No reliable edge |
| Finance & real estate | +0.9 pts n 106 · t 0.7 | −0.1 pts n 267 · t 0.2 | No reliable edge |
| Construction | +0.6 pts n 18 · t 1.1 | −6.1 pts n 79 · t 0.1 | Too few trades to say |
| Industrials & manufacturing | −0.1 pts n 432 · t 1.5 | −0.2 pts n 1,277 · t -1.1 | No reliable edge |
| Insurance | −0.4 pts n 140 · t 0.3 | −1.0 pts n 248 · t -1.3 | No reliable edge |
| Services | −0.9 pts n 214 · t 0.9 | −1.7 pts n 519 · t -0.7 | No reliable edge |
| Biotech & pharma | −1.2 pts n 137 · t 0.4 | +4.7 pts n 431 · t -0.5 | No reliable edge |
| REITs | −3.7 pts n 177 · t -0.8 | −1.0 pts n 398 · t -0.7 | No reliable edge |
| Utilities | −3.7 pts n 96 · t 1.6 | −0.7 pts n 179 · t -2.4 | No reliable edge |
| Healthcare services | −6.0 pts n 10 · t -0.4 | −1.1 pts n 69 · t -0.1 | Too few trades to say |
t is the clustered t-statistic: above 2 means the result is unlikely to be luck; above 3, very unlikely. "n" is trades.
Market state now: NORMAL — the S&P 500 is 3.1% below its one-year high (as of 2026-09-16). The calm column above is the one that applies today; the sell-off numbers are what to expect if conditions change.
Sectors under pressure right now (their own ETF is in a drawdown):
No buy in the last 45 days matches this definition among names we cover.
SAVNG is an educational research publication, not a Registered Investment Adviser. Everything here is general-circulation information produced by an automated pipeline — the same for every reader — and is not personalized investment advice, an offer, or a recommendation to buy or sell any security. Any performance figure is hypothetical; past performance does not predict future results. Details.
Published by SAVNG Research. Every figure on this page is read from the study data at render time; if the study changes, the page changes with it.