COP top-decile weekly net insider buying → forward 20-day return
Insiders at ConocoPhillips are often thought to read the commodity tape better than the street. The logic is straightforward: when their buy clusters appear, a local bottom may be near. But a computed test on three years of COP Form-4 activity suggests otherwise.
Only six weeks qualified as top-decile net insider buying. The average forward 20-day return after those weeks came to 2.71% versus a baseline of 0.44% — a 2.27-point gap. Yet with a p-value of 0.466 and outcomes ranging from solid gains to a -7.3% drawdown, that gap is statistically indistinguishable from noise.
The full methodology and caveats are laid out below. The short version: the story of clever insider timing doesn't survive contact with the numbers.
Over the past ~3 years, do weeks with top-decile net insider buying in COP predict above-baseline forward 20-day returns? Thesis: COP insiders time the commodity cycle well, so clustered buying marks local bottoms that outperform.
How this was measured
Aggregated COP insider Form-4 transactions (market trades only: share_price > 0) to ISO calendar weeks, computing net notional (acquisition $ − disposal $). Selected weeks in the top 10% of weekly net notional (i.e., largest net buying). For each such week, we anchored on the last trading day of that week (Friday or the latest available day) and measured the forward 20-trading-day close-to-close return. Compared the mean of these event returns against the unconditional forward-20-day return distribution over all trading days with sufficient future data, using a Welch two-sample t-test. Overlapping forward returns (if events are within 20 days) are not corrected for; see caveats.
The key numbers
Reading the numbers
Top-decile insider-buying weeks averaged about 2.7% over the next 20 days versus 0.4% for all weeks, but with only 6 event weeks and a p-value of 0.47, that gap is not statistically reliable.
The charts
This histogram spreads out the six event-week outcomes, and the first thing to notice is how wide they are: returns range from about -7.3% to +10.7%, with the average landing at +2.7%. With only six observations, one or two big winners drive that average, so the picture is more scattered than consistent. That matters because it means the apparent edge could easily be a small-sample accident rather than a real pattern.
The taller bar on the left, about 2.7%, is the average 20-day return after top-decile insider-buying weeks; the shorter bar on the right, about 0.4%, is the all-week baseline. So event weeks did beat the baseline by roughly 2.3 percentage points on average. But the statistical test says this difference is not clear-cut: with a p-value of 0.47, there's a good chance a gap this size would appear by chance even if insider buying had no real predictive power.
High-insider-buying event weeks and their forward-20d returns
| anchor_date | net_notional_usd | forward_20d_return |
|---|---|---|
| 2024-01-19 | 2200389.08 | 0.0328 |
| 2024-12-20 | 1256835.31 | 0.092 |
| 2025-01-17 | 2200894.15 | -0.0732 |
| 2025-06-20 | 499,472 | -0.0358 |
| 2025-11-14 | 168439.96 | 0.0401 |
| 2026-01-16 | 2640457.2 | 0.1066 |
The takeaway
No — over the last three years, COP's top-decile insider-buying weeks have not shown a meaningful edge over the stock's typical 20-day return. Only six weeks qualified as top-decile net buying, and the average forward 20-day return after those weeks was +2.71%, versus +0.44% across 734 trading-day anchors. That is a +2.27 percentage-point gap, but with just six events and a p-value of 0.466, the result is well within the range of pure chance — basically a coin flip, not a real signal. The individual outcomes were all over the place, including a -7.3% forward return after a strong buying week in January 2025. The idea that COP insiders time the commodity cycle may be plausible, but this sample is far too thin and noisy to demonstrate it. The practical takeaway is to treat clustered insider buying as an interesting story rather than a reliable predictor of near-term returns.
The fine print
- Only 6 qualifying weeks over ~3 years — far too few for reliable inference; a couple of outliers could flip the average.
- Event weeks sometimes sit within 20 trading days of each other, so forward returns overlap and the significance test is a bit too generous.
- Returns are anchored to the last trading day of the week; using the actual transaction day would capture the market's reaction more precisely.
- Only priced market transactions were counted, so zero-price grants/vestings were left out and may carry different information.