How It Works
Every InsideStox signal starts as a legally required SEC disclosure and survives a gauntlet of filters before it earns a score. Here's the full methodology — nothing hidden.
The raw material: SEC Form 4
When an officer, director, or 10%+ owner of a U.S. public company trades their company's stock, they must file a Form 4 with the SEC within two business days. Every filing is public on SEC EDGAR. InsideStox scans the freshest batch every trading day, covering a rolling two-day window so late-posting filings don't slip through.
We only care about transaction code P — a genuine open-market purchase. Everything else (sales, option exercises, grants, gifts) is discarded automatically.
The anomaly score (0–100)
Every surviving purchase is run through a scoring model that weighs several dimensions of behavior — the exact weighting is proprietary, but here's what we look at and why:
| Factor | Why it matters |
|---|---|
| Purchase size | Bigger personal checks mean bigger conviction. We only consider purchases above a $50K materiality floor. |
| First-time buyer | An insider with no prior open-market purchases suddenly buying is a genuine behavioral reversal — though our own tracked results say it earns a smaller premium than the idea suggests, and it is weighted accordingly. |
| Rare buyer | An insider who almost never buys is a meaningfully different signal than one who buys routinely. |
| Cluster buying | Multiple insiders at the same company buying around the same time is rare, and rarer still to be coincidence. |
| Already spiked | If the stock has already run before we surface the signal, the opportunity is largely gone. In our own tracked results this cohort has shown no edge at any holding period, so it carries a heavy penalty — and the size of the run-up is shown on the signal. |
| Share price | Very low-priced stocks carry outsized downside risk and the least reliable market data. Sub-$5 names are penalized; sub-$1 names are excluded entirely (see below). |
Signals are ranked into three tiers — strong, notable, and watch-list — so you can prioritize at a glance without needing to know the underlying math.
The weights are not fixed. We periodically re-test every factor against our own tracked results and adjust what each one is worth. In the August 2026 review the first-time-buyer premium was cut sharply, the already-spiked penalty was more than doubled, and the largest purchases were weighted up further — each because the tracked outcomes said so, not because the theory changed. Scores are absolute, not curve-graded, so a recalibration can shift the whole distribution at once; what a tier means is defined by the cutoffs, which move with it.
What we exclude — even when other trackers don't
- Hedge fund / 10%-owner accumulation. A fund's systematic position-building isn't a surprise executive bet. Skipped unless an actual officer or director is buying alongside.
- Mislabeled sales. Aggregator headlines routinely report 10b5-1 plan sales as "buys." We verify the SEC transaction code on the actual filing, every time.
- Compensation, not conviction. RSU grants, option awards, and option exercises are pay, not purchases. Excluded.
- Sub-$50K routine buys. Below our materiality floor, even if it's a first-time purchase.
- Stocks under $1.00. Excluded outright since August 2026. Across our own tracked signals this is where nearly all of the worst outcomes have lived, and where third-party price data is least trustworthy (splits, stale quotes, wide spreads). One published disaster costs a subscriber more than several small winners return.
- IPO and offering allocations. Insiders buying at the offering price on debut day is participation, not open-market conviction. Included only with an explicit caveat.
- Serial buyers. An executive who buys every month isn't anomalous. Flagged only when there's a clear step-up — a buy several times their usual size.
- Activist campaigns. Buying during a public board fight may be strategic rather than informational. Included but weighted down and caveated.
Living signals, honest scoreboard
A flagged signal isn't a press release — it's a position we track. Every trading day, each active signal's chart and return-since-flag are refreshed with real market closes. After 30 days a signal retires to the archive with its final result frozen, and after 90 days it ages out entirely — a three-month-old result says little about today's market. Win rate, average return, and best signal are computed from the trailing 90-day record — archived losers included — and displayed on the scoreboard at all times.