Should You Copy a Single Insider's Trades? What 102,000 Insider Purchases Tell Us
Key findings
- The typical lone insider purchase does not beat small caps after costs: median 60-day return +0.7%, which is 1.7% behind the Russell 2000 and 3.0% behind the S&P 500 (51,240 trades).
- An insider's track record does not predict their next trade. The rank correlation is essentially zero (0.035 and 0.018).
- The "buys near the 52-week high" effect that looked promising turned out to be mostly momentum, not an insider effect, once we compared with matched stocks.
- Insider purchases do carry a thin edge over similar stocks: about +1.65% per 60 days, roughly the size of small-cap trading costs.
- Stocks under $5 do badly: a median 7.7% to 10.3% behind the Russell 2000.
- Filtering and context matter more than copying one person.
"Just follow the insiders" is one of the most repeated ideas in retail investing. A company's own executives and directors buy their stock with their own money, the purchase is public within days on a Form 4, and so the obvious plan is to track a good insider and copy the trade. We tested that plan on about 102,000 insider purchases. This is what held up, and what did not.
How we measured
- Data: 102,083 open-market purchases ("P - Purchase") from the OpenInsider historical screener, filed from January 2019 to October 3, 2026, after removing duplicates. We excluded 2,020 rows that looked like errors (zero or negative value or price, over $500 million, or no trade date).
- Test window: filings from January 2021 to June 2026 (2019 and 2020 were used only as history for insider track records). 51,240 purchases in that window have price data and a 60-day result.
- Entry: the next market open after the Form 4 was filed. A filing before 9:30 a.m. ET on a trading day is entered at that day's open, any other at the next session's open.
- Return: price change over 60 trading days (about 90 calendar days), using adjusted prices, compared with the Russell 2000 (IWM, the fairer benchmark for this small-cap-heavy universe) and the S&P 500 over the same dates.
- Honesty checks: we found rules in 2021 to 2023 and then tested them on untouched 2024 to mid-2026 data. We re-derived 5 random trades from a fresh price download, and all 5 matched exactly.
Finding 1: the typical lone insider buy does not beat small caps
Across all 51,240 priced purchases, the median 60-day return was +0.7%. That sounds fine until you compare it with what a small-cap index did over the same dates.
| Measure (60 trading days) | Result |
|---|---|
| Trades | 51,240 |
| Median return | +0.7% |
| Median vs Russell 2000 | -1.7% |
| Median vs S&P 500 | -3.0% |
| Trades that beat the Russell 2000 | 46.3% |
| Mean return (a few big winners pull it up) | +3.5% |
| Mean excess vs Russell 2000 | +0.8% |
The mean looks better than the median because a few large winners lift it. After subtracting the Russell 2000 the mean edge is only +0.8%, before costs. Trading small caps costs roughly 0.5% to 2% round trip depending on liquidity, which turns a thin positive mean into zero or worse. This is before any correction for the survivorship bias described below.
Finding 2: an insider's track record does not predict the next trade
If one insider is consistently good, copying that insider should work. To test it, we scored each trade using the insider's earlier purchases, counting only purchases whose 60-day window had already ended before the new trade (no look-ahead), and requiring at least 3 of them. That covered 25,308 trades, 49.4% of the window.
The result is close to zero. The rank correlation between an insider's past record and the next trade's excess return was 0.035 using past median excess and 0.018 using past hit rate. The top quintile of "proven" insiders did not do better than the rest, it did worse:
| Insider's past median excess | Next trade, median vs Russell 2000 |
|---|---|
| Top quintile (best past record) | -3.7% |
| Everyone else (not top quintile) | -0.6% |
| Bottom quintile | -6.2% |
The ordering is not consistent enough to trade on: the bottom group is worst, but the top group is also below average. A good past record is not a signal.
Finding 3: the "near the 52-week high" effect was momentum
One pattern looked real in our first pass. Purchases made within 10% of the stock's 52-week high beat the Russell 2000 in both test periods, by a median +2.5% in 2021 to 2023 and +1.2% in 2024 to mid-2026 (about 4,000 trades in each). Because we found it after looking at many tables, we treated it as a hypothesis and tested it properly.
The test: for every insider trade, we drew 5 other stocks with no insider buying that were bought the same day. They had the same distance from their 52-week high, the same price range and similar trading volume, and no insider purchase in the 90 days before or the 60 days after. If insiders add something beyond "strong stocks keep going up", they should beat those controls near the high.
| Period | Median edge vs matched stocks | 95% interval |
|---|---|---|
| 2021 to 2023 | -0.10% | -0.86% to +0.69% |
| 2024 to June 2026 | +1.45% | +0.78% to +2.12% |
The edge shows up in only one of the two periods, so the pre-registered test was not confirmed. We also checked whether the insider edge is bigger near the high than elsewhere. It is not: if anything it is smaller there (-0.69%, interval -1.44% to -0.02%). Our reading is that most of the original pattern was momentum and market conditions, not something insiders add.
Finding 4: insider buys do carry a thin edge over similar stocks
Pooling every bucket, we compared each insider purchase with each of its 5 matched stocks. The insider purchase came out ahead by a median +1.65% per 60 days (95% interval +1.24% to +2.08%), and won 52.7% of 242,280 comparisons. It held in both periods: +1.39% in 2021 to 2023 and +1.99% in 2024 to mid-2026.
So insiders do seem to pick slightly better than similar stocks. But 1.65 points is about the size of the 0.5% to 2% round-trip cost on small caps. It is a real but thin signal, and not a standalone strategy.
One measurement caveat: our pre-specified comparison, which used the average of the 5 controls, gave about zero (+0.01%). We later found that this method is biased against the insider by about 1.4 to 1.6 points, because an average of 5 stocks is less spread out than one stock and a skewed distribution then pulls the median down. We tested this with a placebo and switched to the head-to-head comparison above. Treat the +1.65% as suggestive.
Finding 5: stocks under $5 do badly
The most stable pattern in the data is not about insiders at all. It is about price.
| Entry price | Median vs Russell 2000 | Beat the Russell 2000 |
|---|---|---|
| Under $1 | -10.3% | 37.5% |
| $1 to $5 | -7.7% | 39.7% |
| $5 to $10 | -4.2% | 42.3% |
| $10 to $20 | -0.5% | 48.8% |
| $20 to $50 | +0.3% | 50.8% |
| Over $50 | +1.0% | 53.1% |
Stocks under $5 are about 29% of priced insider purchases, and a sample we checked showed them to be typically micro caps under $300 million. They are also the most expensive to trade, at about 2% round trip. Larger purchases did not help either: buys of $1 million to $5 million trailed the Russell 2000 by a median 3.8% and those over $5 million by 5.3%, and CEO and 10% owner buys were behind average too. Removing sub-$5 stocks was the only simple rule that improved the median in both test periods.
What this means: filtering and context beat copying one person
The data points one way. The act of an insider buying is a weak input, and who the insider is tells you almost nothing. What changes the odds is context: how many insiders agree, what kind of stock it is, and whether the purchase is real conviction. That is why ClusterSignal does not follow individuals. It looks for clusters, several insiders buying the same stock in a short window, and then scores each one across five dimensions (insider signal, technical analysis, financial health, analyst sentiment, market context). It also flags purchases that look like a disguised stock offering instead of conviction, and keeps stocks under $5 in a separate product, PennySignal.
This study does not prove that those filters work. It tested single OpenInsider purchases, not ClusterSignal's scored alerts. A raw cluster detector with no scoring was also run on the same history and did not beat single buys. That detector is not the product: scoring and filters are what ClusterSignal adds on top, and they need their own test. We ran that test on our scored alerts: Do insider buying alerts beat the market? The edge there was clear against the S&P 500, thinner against similar stocks, and not yet proven against plain insider buys, on a small sample.
Limits of this study
- Survivorship bias. 27.1% of all collected purchases (21.9% of the clean test-window trades) have no price history, mostly delisted, acquired, merged or renamed companies. Bankruptcies are missing, but so are buy-outs at a premium, so the direction of the bias is unclear. Coverage is worst in 2021 (62.8% of filings priced) and best in 2026 (97.4%).
- Many rules tested. We evaluated 594 distinct rules and buckets. At that number, a few "significant" results are expected by chance. We only credit results that repeated in untouched later data, and even those are post-hoc.
- Trades are not independent. Several insiders often buy the same company on the same day, and some insiders buy repeatedly, so real significance is lower than raw sample sizes suggest.
- Matched stocks are imperfect. The control stocks come from companies that had an insider purchase at some other point and that have price data, so they carry survivor bias too. Both sides share it.
- Costs are estimated. We used a simple three-tier cost model by liquidity (0.5%, 1% and 2% round trip), not real spreads.
- Two periods differ. 2021 to 2023 and 2024 to mid-2026 behaved differently, and several patterns held in only one.
Frequently asked questions
Should you copy a single insider's trades?
Not on this evidence. Across 51,240 priced insider purchases filed from January 2021 to June 2026, entered at the next open after the Form 4, the median 60-day return was +0.7%, which is 1.7% behind the Russell 2000 (IWM) and 3.0% behind the S&P 500. Only 46.3% of trades beat the Russell 2000, and typical round-trip trading costs on small caps are 0.5% to 2%.
Does an insider's past track record predict their next trade?
No. We ranked insiders by their earlier results using only trades that had already finished. The rank correlation with the next trade's 60-day excess return was 0.035 for past median excess and 0.018 for past hit rate, which is essentially zero. Trades by insiders in the top quintile of past performance had a median 60-day return 3.7% behind the Russell 2000, against 0.6% behind for everyone else.
Do insider purchases beat similar stocks with no insider buying?
By a thin margin. Compared with 5 matched stocks with no insider buying, bought the same day, the pooled median edge was +1.65% per 60 days (95% interval +1.24% to +2.08%), and the insider purchase did better in 52.7% of 242,280 comparisons. That is roughly the size of round-trip trading costs on small caps (0.5% to 2%), so it is not a standalone strategy.
Do insider buys near a 52-week high do better?
Not because of the insider. Buys within 10% of the 52-week high looked good in a first pass, but against matched stocks that were also near their highs the edge was -0.10% in 2021-2023 (interval -0.86% to +0.69%) and +1.45% in 2024 to mid-2026. Our pre-registered test of that idea was not confirmed, and the effect looks like momentum rather than an insider effect.
Do insider purchases of stocks under $5 work?
No. Purchases of stocks priced $1 to $5 had a median 60-day return 7.7% behind the Russell 2000, and stocks under $1 were 10.3% behind. These stocks are about 29% of priced insider purchases, and they cost about 2% round trip to trade. We read this as a property of cheap stocks in general rather than of insider buying, since the pattern follows price.
What are the limits of this study?
27.1% of all collected purchases had no price history (mostly delisted, acquired or renamed tickers), which is survivorship bias we could not fix. We also evaluated 594 distinct rules and buckets, so some apparent patterns are expected by chance, and trades are not fully independent because several insiders often buy the same stock on the same day.
Is this investment advice?
No. ClusterSignal is an information service. This analysis describes historical data and does not tell anyone to buy or sell a security. Past performance does not guarantee future results.
Related reading
What is a Form 4 filing? explains where the purchase data comes from, and Do insider buying alerts beat the market? tests our scored alerts against four benchmarks.
⚠ ClusterSignal is a financial information service. This analysis and the alerts it describes are for informational purposes only and do not constitute investment advice. Past performance does not guarantee future results. Consult a licensed financial advisor before making investment decisions.
See how scored alerts have done
Every resolved alert, including the losers, is public.
View the Results page →