AI Research COINCOIN_insider

COIN insider net‑selling extremes → forward 20‑day return (weekly, top decile)

148
Weeks analyzed (with return data)

After weeks of heavy insider selling, COIN has tended to drift lower over the next month — but the signal is far from crystal clear. The top-decile selling weeks averaged a forward 20-day return of -4.4%, while all other weeks returned +6.0% — a gap of more than ten percentage points. Yet with only 15 extreme-selling weeks in 148, the result carries a p-value of 0.062, meaning it could be noise about six times in a hundred.

This study tests whether insider net-selling extremes, as reported on Form-4, reliably precede below-baseline returns — the idea that smart money exits near local peaks. The directional support is there, but the evidence is borderline. For the full breakdown of data, charts, and the reasoning behind this finding, read the analysis below.

The research question

For COIN over the past ~3 years, do weeks with top-decile net insider selling as reported on Form-4 predict below-baseline forward 20-day returns, signaling that insiders reliably offload near local peaks? Thesis: Consistent insider selling marks price tops, so the stock underperforms in the following month as the smart money quietly exits.

How this was measured

Form‑4 insider transactions for COIN (last ≈3 years) are aggregated to calendar weeks (Mon‑Sun). Net selling per week = total disposal notional − total acquisition notional, with zero filled for weeks with no insider activity. Weeks are ranked by net selling; the top decile (highest selling) is tagged as heavy‑selling weeks. For every calendar week, the forward 20‑trading‑day return is measured from the close of the first trading day on or after that week’s Friday. A Welch two‑sample t‑test (unequal variance) compares the forward returns of heavy‑selling weeks against the remaining (non‑top‑decile) weeks. The thesis is that heavy net selling signals insider distribution near local highs, predicting subsequent underperformance.

The key numbers

Weeks analyzed (with return data)
148
Heavy‑selling weeks (top decile)
15
threshold ≥ 28,654,849.88 USD net selling
Rest weeks (non‑top‑decile)
133
Mean fwd 20‑day return — heavy weeks
-4.4041%
N=15; mean return after extreme insider selling
Mean fwd 20‑day return — rest weeks
5.9776%
N=133
Difference (heavy − rest)
-10.3817%
negative would mean selling weeks underperform
Welch t‑statistic
-1.974
positive = heavy returns higher (opposite thesis) – negative favors thesis
Welch p‑value (two‑sided)
0.0618
p=0.0618 ≥ 0.05 → no statistically‑clear separation

Reading the numbers

After extreme insider selling weeks, COIN's average 20-day return was -4.4%, versus +6.0% for all other weeks — a roughly 10-percentage-point gap. But that gap just misses conventional statistical significance (p=0.06), so it's suggestive, not conclusive.

The charts

COIN forward 20‑day returns: heavy‑selling weeks vs rest
What this chart says

This histogram stacks the frequency of forward 20-day returns for heavy-selling weeks (orange) against all other weeks (blue). The orange bars cluster below zero, meaning most heavy-selling weeks were followed by negative returns, while the blue bars stretch further into positive territory. The key takeaway is that the two distributions clearly separate at the zero line, but the small number of orange bars (just 15 weeks) makes the pattern fragile.

Forward 20‑day return distribution: heavy vs rest
What this chart says

The box plot lays out the full return distribution for each group. The heavy-selling box sits entirely below zero (median negative) with a mean of -4.4%, while the rest group's box straddles zero with a mean of +6.0%. The most telling detail is that the heaviest-selling weeks' entire middle 50% of returns is negative, whereas most other weeks see a mix of gains and losses. This aligns with the thesis that insiders offload near peaks, but the wide whiskers and small heavy-selling sample (15 vs 133) remind us not to over-interpret.

Top‑decile heavy‑selling weeks (sorted by net selling high→low)

week_startanchor_datenet_selling_usdfwd20_return
2024-11-182024-11-22332,784,358-0.1144
2025-07-142025-07-18219,448,062-0.242
2024-11-112024-11-15175,701,0590.0351
2025-06-232025-06-27164,317,8780.0757
2024-03-042024-03-08143,640,9890.008
2024-02-262024-03-01137,385,5270.2218
2025-11-032025-11-07103,076,229-0.113
2024-12-022024-12-0660,364,121-0.2194
2025-11-102025-11-1457,025,478-0.1287
2026-02-022026-02-0655,074,3170.1923
2026-01-122026-01-1643,876,969-0.3137
2023-12-182023-12-2242,749,690-0.3067
2023-12-112023-12-1534,051,596-0.0996
2026-02-162026-02-2030,233,1060.162
2024-05-132024-05-1729,369,6910.1819

The takeaway

There's a pattern suggesting that COIN's worst insider-selling weeks precede a soft month, but it's not statistically airtight. Over 148 weeks, the top-decile selling weeks (15 total) averaged a forward 20-day return of -4.4%, while the rest of the weeks averaged +6.0% — a gap of over 10 percentage points. But with a p-value of 0.062, that gap could be random about 6 times out of 100, which is borderline and not conclusive by usual standards. Given only 15 heavy-selling weeks, the result is more of a suggestive lean than a reliable signal: one or two extremes (like the -24% after July 2025's heavy selling) are pulling the average down. For the thesis that insiders sell near local peaks, the directional support is there, but the evidence isn't strong enough to trade on alone.

The fine print