The conventional wisdom is simple: when corporate insiders buy their own stock, follow them. They know the business better than anyone, and if they're putting their own money on the line, the stock must be undervalued.
We wanted to test that claim rigorously. Not with anecdotes or cherry-picked examples, but with every SEC Form 4 filing over a three-year period, proper benchmarking, and statistical significance testing.
The short answer: insider buying does predict positive returns — but only for about 10 days. After that, the signal doesn't just fade. It reverses. And the real edge isn't in following any single insider. It's in watching for a very specific pattern that most people overlook.
The Dataset
We analyzed 906,088 SEC Form 4 filings from January 2023 through March 2026, sourced directly from SEC EDGAR. This covers every insider transaction reported to the SEC during that period — purchases, sales, grants, awards, tax-related transactions, and option exercises.
For our headline analysis, we filtered down to the signals most commonly cited as actionable:
- Open market purchases only (transaction code P) — excluding grants, awards, and tax sales
- C-suite insiders — CEO, CFO, COO, Chairman, and President (excluding Vice Presidents and below)
- Minimum transaction value of $100,000 — filtering out trivial purchases
This filtering reduced 906,088 filings to 3,236 backtestable signals across 1,169 unique tickers.
Methodology
Getting the methodology right matters more than the results. Insider trading studies are riddled with lookahead bias, and we took specific steps to avoid it.
Entry price: For each signal, we used the opening price on the first trading day after the SEC filing date — not the transaction date. This distinction is critical. The insider may have traded days or weeks before the filing hits EDGAR. The public doesn't know about the trade until the filing appears, so the filing date is the earliest you could realistically act.
Exit price: Closing price at 5, 10, 30, 60, and 90 calendar days after entry. We included the short windows (5 and 10 days) because the most interesting question for active investors isn't "does this work over three months?" but "how quickly does the market react?"
Benchmark: S&P 500 (SPY) over the identical holding period. Both the stock and the benchmark use the same open-to-close methodology to ensure an apples-to-apples comparison.
Transaction costs: 10 basis points (0.10%) round-trip deducted from all returns.
Prices: All prices are split-adjusted via Yahoo Finance, ensuring stock splits don't distort return calculations.
Statistical testing: One-sample t-tests on excess returns versus zero, with 95% confidence intervals. We flag any result with fewer than 30 observations as having insufficient sample size.
Survivorship bias: Approximately 14% of signals were excluded because the underlying stock was delisted during the analysis period and price data was unavailable. Since delisted stocks tend to be losers, this likely biases our results slightly upward. We disclose this rather than pretend it doesn't exist.
Finding 1: The Signal Is Real — But It's a 10-Day Signal
This is the core finding. Insider buying generates statistically significant positive excess returns at 5 and 10 days. By 30 days, the signal is gone. By 60 and 90 days, it's actually negative.
| Holding Period | Mean Excess Return vs S&P 500 | Win Rate | p-value | Significant? |
|---|---|---|---|---|
| 5 days | +0.98% | 51.2% | <0.0001 | Yes *** |
| 10 days | +0.97% | 51.3% | <0.0001 | Yes *** |
| 30 days | +0.02% | 43.8% | 0.93 | No |
| 60 days | −1.56% | 40.6% | 0.0003 | Yes *** (negative) |
| 90 days | −1.59% | 38.2% | 0.003 | Yes ** (negative) |
This reframes the entire narrative around insider buying. It's not "insiders know the stock is undervalued and eventually the market catches up." It's a filing-reaction trade — the market adjusts to the information in the filing within about two weeks, and after that, there's no residual edge.
The negative returns at 60 and 90 days are particularly interesting. They're statistically significant, meaning this isn't just noise. One possible explanation: the types of stocks that insiders buy tend to be recent underperformers (that's why insiders consider them cheap), and the short-term pop from the filing news doesn't change the underlying trajectory.
The practical takeaway: if you're using insider buying signals, act quickly and don't overstay. The data supports a 5–10 day holding period, not a buy-and-hold approach.
Finding 2: Cluster Buys Are the Real Signal
This is the strongest finding in the entire study, and it's the one with the most practical value.
A "cluster buy" is when two or more distinct insiders make open market purchases of the same stock within five trading days of each other. One insider buying could mean anything — portfolio rebalancing, a board member making a token purchase for optics, contractual obligations. But when the CEO, CFO, and a director all independently reach into their own pockets in the same week? That's a different signal entirely.
We identified 1,472 clusters in our dataset.
| 5 days | 10 days | 30 days | 60 days | 90 days | |
|---|---|---|---|---|---|
| Cluster buys (N=820) | +2.02%*** | +2.41%*** | +2.29%* | +0.58% | +0.19% |
| Single insider (N=1,997) | +0.62%** | +0.50%* | −0.20% | −2.22%*** | −2.29%*** |
| Difference significant? | p=0.0001 | p<0.0001 | p=0.016 | p=0.011 | p=0.065 |
Three things stand out:
The magnitude is much larger. Cluster buys show +2.41% excess return at 10 days versus +0.50% for single insider buys. That's nearly a 5× difference.
The signal persists longer. Single insider buys fade to zero by day 30 and turn significantly negative by day 60. Cluster buys remain positive through day 30 and don't turn significantly negative at any window.
The difference between cohorts is statistically significant. This isn't just "clusters happen to be higher." The t-test comparing cluster vs single returns shows the difference itself is significant at 5, 10, 30, and 60 days.
Why does this work? The most likely explanation is information convergence. When multiple insiders independently decide to buy at the same time, they're probably reacting to the same internal information — a strong quarter they've seen early, a pipeline development, a deal in progress. The redundancy of the signal filters out the noise of any individual insider's personal motivations.
Finding 3: Sector Matters — Healthcare Leads
Not all sectors produce equally strong insider buying signals. Healthcare, and biotechnology in particular, stands out dramatically.
| Sector | 5d Excess | 10d Excess | N | Notable |
|---|---|---|---|---|
| Healthcare | +3.03%*** | +2.28%** | 443 | Strongest overall signal |
| Consumer Cyclical | +1.27%* | +2.14%*** | 325 | Strong at 10 days |
| Financial Services | +0.49%* | +0.48% | 640 | Largest sample, modest signal |
| Technology | +0.81% | +1.40%* | 380 | Significant only at 10d |
| Real Estate | +0.62% | −0.79% | 291 | Turns negative quickly |
| Energy | −0.41% | +0.49% | 135 | No significant signal |
Within Healthcare, we broke down results by sub-industry. Biotechnology specifically drives the result, with +4.8% excess return at 5 days (N=152, p<0.001). This is the highest excess return of any sub-group in the study with a meaningful sample size.
This makes intuitive sense. Biotechnology has the highest information asymmetry of any sector — insiders know clinical trial data, FDA feedback, and pipeline developments long before the market does. When a biotech CEO makes a large open market purchase, they're sitting on information that could move the stock 50% in either direction once public.
Finding 4: The Best Filter Combinations
Individual findings are interesting. Combinations are actionable. We tested several intersections of our filters to find the highest-conviction setup.
| Filter Combination | 10-Day Excess Return | Sample Size |
|---|---|---|
| Cluster + Healthcare | +5.65%*** | 120 |
| Cluster + CEO/Chairman | +5.19%*** | 97 |
| Cluster + Position increase >50% | +4.90%*** | 117 |
| Cluster alone | +2.41%*** | 820 |
| No filter (C-suite ≥$100K baseline) | +0.97%*** | 3,236 |
Every high-performing combination has cluster buying as the foundation. No single-insider filter combination comes close to the cluster results. This reinforces the core finding: the number of insiders buying matters more than who they are, how much they spend, or what sector they're in.
A caveat on sample sizes: the combined filters produce smaller samples (97–120 signals). While the results are statistically significant, these should be treated as directional rather than definitive. We'd want to see these hold in out-of-sample data before building a portfolio around the specific combinations.
Finding 5: What Doesn't Matter (as Much as You'd Think)
Some commonly cited insider trading heuristics didn't hold up in the data.
Transaction Size
| Size Bucket | 5d Excess | 10d Excess | N |
|---|---|---|---|
| $100K – $500K | +0.50%*** | +0.46%*** | 9,066 |
| $500K – $1M | +0.47%* | +0.68%** | 2,051 |
| $1M – $5M | +0.78%*** | +1.17%*** | 2,809 |
| $5M+ | +0.59% | +0.92%* | 1,326 |
The t-tests comparing the smallest and largest buckets are all non-significant. A $5 million purchase doesn't predict better returns than a $200,000 purchase. This contradicts the popular intuition that "bigger buy = more conviction = better signal."
Position Conviction
Insiders who doubled their position (+100% increase) showed marginally better short-term returns than those adding 10%, but the difference wasn't dramatic. The signal exists at all conviction levels, suggesting that the act of buying matters more than the amount relative to existing holdings.
Filing Speed
Insiders who file within 0–5 days of the transaction show similar returns regardless of the exact timing. One notable exception: insiders who take 6 or more days to file show −15.5% at 60 days. Late filing appears to be a red flag — possibly indicating less sophisticated or less confident insiders.
Finding 6: Market Conditions
A fair question: does this only work in a bull market? Our analysis period (2023–2026) was broadly favorable for equities, so we tested across different market regimes.
| Market Regime | 5d Excess | 10d Excess | N |
|---|---|---|---|
| Bull (SPY 3-month return >+5%) | +1.29%*** | +1.55%*** | 1,368 |
| Flat (SPY ±5%) | +0.77%*** | +0.47% | 1,452 |
| Bear (SPY 3-month return <−5%) | +1.68%* | +2.15%* | 205 |
The signal is present in all three regimes at the 5-day window. The bear market sample is small (N=205), so that result should be treated with caution, but the signal isn't purely a bull market artifact.
We also split by calendar year:
| Year | 5d Excess | 10d Excess | N |
|---|---|---|---|
| 2023 | +0.31% | −0.00% | 907 |
| 2024 | +0.92%** | +0.78%* | 839 |
| 2025 | +1.70%*** | +2.06%*** | 1,253 |
| 2026 (partial) | −0.13% | −0.41% | 236 |
The signal was strongest in 2025 and weakest in 2023. The 2026 partial year (January through March) coincided with significant market volatility from tariff uncertainty, which may explain the negative results on a small sample.
Finding 7: Market Capitalization
If the signal only exists in micro-cap stocks, it's not practically useful for most investors — those stocks are illiquid and expensive to trade. We tested across the market cap spectrum.
| Market Cap | 5d Excess | 10d Excess | N |
|---|---|---|---|
| Micro (<$300M) | +1.74%*** | +1.20%** | ~800 |
| Small ($300M–$2B) | +1.05%*** | +1.31%*** | ~1,100 |
| Mid ($2B–$10B) | +0.39% | +0.52% | ~800 |
| Large ($10B+) | +0.58% | +0.73%* | ~500 |
The signal is strongest in micro and small caps, as expected — these stocks have less analyst coverage and more information asymmetry. But it's also present in large caps at the 10-day window, which is important: this isn't a micro-cap-only phenomenon that disappears once you account for liquidity constraints.
Distribution of Returns
Individual signal results vary widely. The mean excess return of +0.98% at 5 days doesn't mean every trade makes about 1%. Some signals produce +30% returns; others produce −20%. The distribution is wide and right-skewed.
This is why the statistical approach matters. Any individual insider buy could go either way. The edge exists in the aggregate, across many signals, and it's most reliable when combined with the cluster filter.
Limitations
We believe in transparent research. Here's what could be wrong with this analysis:
Time period: Three years of data covering a predominantly bullish market environment. The signal's behavior in a prolonged bear market or recession remains untested.
Survivorship bias: 14–19% of signals were excluded due to delisted tickers. If those stocks performed poorly (likely), our reported returns are somewhat inflated.
No factor adjustment: We benchmark against SPY, not against a Fama-French factor model. The insider buy signal may partially capture small-cap premium, value premium, or sector momentum rather than pure insider information alpha.
Multiple comparisons: We tested many sub-groups across multiple time windows. Some results that appear significant at p<0.05 would lose significance under Bonferroni correction. The cluster finding survives correction easily; some sector-specific results might not.
Market impact: For smaller, less liquid stocks, the act of entering a position after a filing could itself move the price, reducing the realized return below what our backtest shows.
Forward-looking returns: The 2026 data is incomplete and coincides with unusual market conditions. As more data accumulates, the specific numbers will shift.
Implications
For investors who track insider activity, the data supports a specific approach:
Watch for clusters, not individual buys. A single insider buying is a weak signal. Two or more insiders buying the same stock in the same week is a substantially stronger signal that persists longer.
Act quickly. The edge is in the first 5–10 days. By 30 days, it's gone. If you're not set up to act on filings within 24 hours, you're probably too late for the strongest part of the signal.
Sector matters. Healthcare and biotechnology insiders produce the strongest signals, likely due to the high information asymmetry in those industries.
Don't hold too long. The data shows that holding insider-buy signals for 60–90 days actually underperforms the market. This contradicts the popular "follow the insiders for long-term value" narrative.
Avoid late filers. Insiders who take 6+ days to file their Form 4 show significantly negative long-term returns. Treat late filing as a warning sign, not a delayed opportunity.