nyasekai

Graveyard / Equity events / US stocks

Copying insider buying

Does buying stocks after company insiders buy beat the market?

Family
Equity events
Verdict
Dead
Cause
No edge
Buried
3 October 2026

The idea

Officers and directors who buy their own company's stock with their own money may know something the market has not priced yet. The idea is to buy the stock after the public filing and hold it while that information spreads.

How we tested it

DataSEC insider transaction filings (Form 4) and daily prices for 8,362 symbols, including delisted names. Events: clusters of 2 or more filings within 30 days, or officer buys of $100k or more. Long the stock, short an equal dollar amount of IWM, entry at the next open after the filing. Price at least $5 and 20-day median dollar volume at least $2M.
PeriodJanuary 2016 to October 2026 (9,835 tradeable events).
Costs0.13% per side, 0.26% round trip.
Holdout24 configurations (take profit, stop, hold of 60 or 90 days, filter). Each had to pass on the full period, in-sample and out-of-sample.
ControlNo random-entry control is stated in the notes. The benchmark was IWM. Against SPY the same events looked like −270 bp over 60 days (t −9), which is a small-cap versus large-cap effect, not an insider effect.

Result

Events9,835
Gross vs IWM, 5 days+23 bp (t 3.21)
Gross vs IWM, 20 days+19 bp (t 1.56)
Gross vs IWM, 60 days−21 bp (t −0.92)
Net of costs, 5 / 20 daysabout −3 bp / −7 bp
Configurations passing0 of 24
Best full-period Sharpe0.46
Out-of-sample Sharpe of the in-sample pick−0.05
Profit factorabout 1.0 everywhere

Why it failed

The effect is real but small: +23 bp gross over five days is smaller than the 26 bp it costs to get in and out. Two measurement traps also showed up. Against SPY the events looked terrible only because small caps lagged large caps in 2021-25. And log returns understate the dollar result of a volatile single stock, so long positions need simple returns.

What would make it worth retesting

Only with near-zero execution costs, or with a sharper subset chosen in advance from the research literature (for example first-ever purchases by a CEO), not by sweeping thresholds again on this data.

Past results of a backtest say nothing certain about the future.