It's one of the most persistent claims in retail investing: when CEOs buy their own stock, the stock outperforms. The academic evidence supports it. But does it hold up in practice, with real filings, real scoring, and real exit rules? We ran the numbers.
Between December 16, 2025 and March 16, 2026 we backtested our full algorithm against every C-suite open-market purchase filed with the SEC. Here's what happened.
The backtest setup
We scanned every SEC Form 4 filing during the 3-month window, filtering for:
- C-suite officers only — CEO, CFO, COO, and CTO
- Open-market purchases only — transaction code "P" (no option exercises, grants, or plan-based buys)
- Multi-factor scoring — each filing scored on 6 weighted factors using AI with web search
- Mechanical exits — +10% target, -15% stop loss, no discretionary overrides
Every signal rated BUY or STRONG BUY received a $100 position. Simultaneously, we opened a $100 SPY position as a benchmark. When the signal position closed (target or stop), the paired SPY position closed too — same holding period, same market conditions.
Most insider buying studies measure raw returns. We benchmark each position against SPY over the identical holding period. This isolates the signal's alpha from broad market movement. A +10% gain during a +8% market rally is very different from a +10% gain during a -3% drawdown.
The headline numbers
| Metric | Value |
|---|---|
| Filings scanned | 210 |
| Filings analyzed | 208 |
| BUY/STRONG BUY signals | 28 |
| Positions opened | 27 |
| Win rate (closed positions) | 89% (8W / 1L) |
| Positions still open | 18 |
| Average signal return | +7.2% |
| S&P 500 (same periods) | -2.83% |
Of 210 C-suite filings scanned, our algorithm flagged 28 as BUY or STRONG BUY — a 13.3% hit rate. This selectivity is intentional. The algorithm rejects most filings because a CEO buying stock isn't automatically a good signal. The scoring filters for filings where fundamentals, financials, technicals, conviction size, insider history, and cluster activity all align.
The 6-factor scoring model
Each filing runs through a weighted scoring system:
The top three factors — fundamentals, financials, and technicals — account for 70% of the score. These are assessed using AI with live web search, grounding each analysis in current data rather than stale snapshots. The bottom three factors — conviction, insider history, and cluster activity — are calculated mechanically from the filing data and our database of prior purchases.
Why conviction size matters
A CEO buying $15,000 of stock is different from a CEO buying $1.5 million. The conviction factor scores purchases on a logarithmic scale from $10K to $5M+. Larger purchases receive higher conviction scores because executives risking significant personal capital are expressing stronger confidence in their company's future.
Why cluster activity matters
When multiple C-suite officers buy the same stock within 30 days, the signal strengthens. Our backtest validated this: every cluster signal in the test period hit its +10% target. BMNM had both CEO and CFO buy — both positions returned +37%. HLNE had three executives buy on the same day. RLI had CEO and COO buy after earnings.
The winners and the loser
Of the 9 positions that closed during the backtest period, 8 hit the +10% target and 1 hit the -15% stop loss.
| Ticker | Insider | Result | Exit |
|---|---|---|---|
| EDSA | CEO | +10% | Target hit |
| TECX | CEO | +10% | Target hit |
| TECX | CFO | +10% | Target hit |
| BMNM | CEO (cluster) | +37% | Target hit |
| BMNM | CFO (cluster) | +37% | Target hit |
| PRPO | CEO | +10% | Target hit |
| PRPO | CFO | +10% | Target hit |
| ADC | CFO | +10% | Target hit |
| TBBK | CFO | -16.7% | Stop loss |
The single loss — TBBK (The Bancorp) — was a CFO purchase that breached the -15% stop loss at -16.7%. Even accounting for this loss, the asymmetric exit rules (+10% target vs -15% stop) create a favorable risk/reward ratio across the portfolio. One winning position recovers 60% of one losing position, and the 89% win rate means winners far outnumber losers.
CEO-only vs C-suite: the improvement
We also ran the backtest with a CEO-only filter (our previous algorithm) over the same period:
| Metric | CEO only | C-suite | Delta |
|---|---|---|---|
| Filings scanned | 172 | 210 | +22% |
| Win rate | 82% | 89% | +7pp |
| Average return | +5.5% | +7.2% | +1.7pp |
| vs S&P 500 | +8.3% alpha | +10.0% alpha | +1.7pp |
Expanding from CEO-only to the full C-suite improved every metric. The key drivers:
- CFO signals added value. CFOs buying their own stock proved to be as predictive as CEO purchases — they have the deepest visibility into the company's financial health.
- Cluster detection became possible. When a CEO and CFO both buy, the signal is stronger than either alone. With CEO-only scanning, we missed these reinforcing patterns entirely.
- Insider history context improved scoring. Tracking prior purchase behavior across all C-suite roles gave the algorithm better context for distinguishing routine buys from conviction purchases.
What the academic research says
Our backtest results align with decades of academic evidence on insider trading returns:
The key insight from both the academic literature and our backtest: not all insider purchases are equal. The signal gets stronger as you move up the hierarchy (CEO > director), increase the purchase size, and find cluster patterns. Our scoring model is designed to capture exactly these gradients.
Limitations and caveats
We're transparent about what this backtest does and doesn't prove:
- 3-month window. This is a short backtest period. Longer backtests will provide higher statistical confidence. We'll continue tracking and publishing updated results.
- 18 positions still open. The 89% win rate is based on 9 closed positions. The final rate could shift as the remaining 18 positions resolve.
- Survivorship in signal selection. The algorithm rejected 86.7% of filings. The win rate applies to the filtered set, not to all insider purchases.
- Market regime dependency. The test period (Dec 2025 – Mar 2026) featured a declining S&P 500. The algorithm may behave differently in a strong bull market.
- Hypothetical positions. The $100 position size is for tracking purposes. Real-world execution would involve spreads, commissions, and liquidity constraints — particularly in small-cap names.
We publish full track record data on our track record page, updated daily. Every position, every entry, every exit — no cherry-picking, no hiding losses. If the algorithm stops performing, you'll see it in the data before we say a word about it.
What this means for investors
The backtest confirms what the academic research has been saying for 25 years: C-suite insider purchases — when properly filtered — generate statistically meaningful alpha over the S&P 500. The edge comes not from blindly following every insider purchase, but from applying multi-factor analysis to separate high-conviction signals from noise.
Three patterns stood out:
- Cluster buying is the strongest signal. Every cluster event in our backtest (BMNM, RLI, HLNE) produced a winner. When multiple executives buy with their own money within the same window, they're collectively expressing conviction that's hard to replicate through any other public data source.
- CFOs are as predictive as CEOs. The expansion from CEO-only to full C-suite improved both win rate and returns. CFOs have the most direct visibility into financial health — when they buy, they're often seeing something the market hasn't priced in.
- The filtering matters more than the following. 210 filings produced 28 signals and 27 positions. The algorithm's value isn't in finding insider purchases — anyone can do that on EDGAR. The value is in knowing which purchases to act on.
These results are from a 3-month window. We'll continue publishing updated performance data as more positions close and the sample size grows. The real test of any signal system isn't a single backtest — it's consistent performance across market conditions, published in real time, with full transparency.
That's what we're building.