Top Insider Methodology

What the “Top Insider” signal means

A filer earns a place on the whitelist when their entire Form 4 trading record — backtested against SPY across the 90-day, 1-year, and 2-year horizons — clears strict statistical thresholds at every horizon.

This is a descriptive backtest, not a forward prediction. We measure how an insider's historical trades compared to the broad market and surface only the filers whose track record is robust across short, medium, and long holding periods.

Important caveat.Beating SPY isn't the same as having market-timing skill. A filer who buys a single stock that runs up will beat SPY even with random timing — that's thesis validation, not skill. To stay honest about this distinction, we classify every whitelisted filer into one of five archetypes, and only one of them earns the green “Top Insider” laurel.

The six archetypes

Each whitelisted filer is classified by two dimensions: issuer concentration (how many distinct tickers they trade) and timing edge (their mean excess return at 730d minus the mean excess return of all otherfilers trading the same top issuer — a buy-and-hold counterfactual). The result is one of six buckets:

• “Top Insider”(diversified timer) — trades many distinct issuers, with a measurable timing edge over buy-and-hold (≥ +10% at 730d). The rarest and most signal-rich archetype.
• Insider Timer— 70%+ concentration in a single ticker, but with a real timing edge over hold-the-issuer. Typically CEOs / officers who know one company deeply (e.g., Colette Kress on NVDA, Jennifer Newstead on META). Their specific timing on their own stock measurably outperforms holding the same stock long-term.
• Concentrated + Timed— one or two issuers dominate (40-70% of trades), and the filer's timing on those issuers measurably outperforms hold-the-issuer.
• Diversified Outperformer— broad exposure across many issuers that has beat SPY, but no measured timing edge. Closer to a smart sector tilt than insider skill.
• Concentrated Thesis— one or two issuers are 40-70% of trades. Beats SPY because the underlying thesis worked, not because the timing was skillful.
• Single-Issuer Thesis— 70%+ of trades concentrate in one ticker, AND timing edge is modest. The backtest mostly measures how that ticker did vs SPY (e.g., Horizon Kinetics on TPL).

The first three buckets (with a positive timing edge) get the green laurel — they represent genuine timing skill. The last three (where alpha comes mostly from exposure, not skill) get an amber-tinted badge, and the trade-card annotation explicitly shows the buy-and-hold baseline so you can judge for yourself whether the filer's edge is skill or thesis.

The criteria

To be on the whitelist, a filer must meet all of the following at every horizon in {90d, 365d, 730d}:

• At least 1,000backtested trades (deduped per economic event) — large enough to make summary statistics meaningful.
• Mean excess return over SPY ≥ +5%— the typical backtest beats the broad market by a meaningful margin.
• Median excess return over SPY > 0— ensures the typical trade is a winner, not an outlier-driven average.
• t-statistic ≥ 5— the mean is highly unlikely to be random noise even after accounting for trade-level variance.

Co-filer pair deduplication

Some economic events are reported by multiple legal entities (a fund manager and their corporate vehicle, for example). We collapse identical(n, mean_excess, win_rate)signatures into a single “anchor” row, surfacing the co-filers as a chip. This prevents the same trades from being double-counted on the public surface.

How often this updates

The underlying backtest aggregate refreshes every night at 05:30 UTC. New insider filings flow in as the SEC publishes them — typically the same day. A filer's “Top Insider” status can therefore change at most once per 24 hours.

How “excess return” is computed

For each Form 4 buy, we record the closing price on the trade date, then compute the percentage change over the next 90 / 365 / 730 trading days. We subtract the SPY return over the same window to get the excess return. The mean and median across all of a filer's trades become the headline numbers.

A small minority of trades (less than 1% in current data) fall on weekends or holidays; we use the closest prior market close. Trades whose horizon extends beyond today's date are excluded until their horizon resolves.

Why “win rate” is not in the criteria

Count-based win rates can mislead. A filer who makes 100 trades and wins 99 of them by a small amount while losing the 100th by a catastrophic margin can have a 99% win rate and a strongly negative mean excess return. We use the mean and median together (the latter rejecting outlier-driven means) instead, which is a more reliable signal of skill.

How the backtests work

Every insider and congress trade we ingest is backtested the same way. We compute its forward return at eight horizonsspanning a 2-year window — from 7 days out to 730 days — and subtract the SPY return over the identical window to get the excess return. The league tables then rank filers by mean excess return per horizon, and every cell carries its win rate and sample size so you can judge how much weight a number deserves before you trust it.

Returns are computed from daily close prices. Trades without sufficient price history — a ticker we can't resolve, or a window that extends past the data we hold — are excluded from the statistics rather than estimated.

Honest caveats

Backtests mislead when their limitations are hidden. These are ours, stated plainly:

• Survivorship bias.The price panel covers companies we track today. Delisted and acquired names can drop out of long-horizon statistics, which flatters averages. A filer's worst trades may be precisely the ones the panel can no longer see.
• Small samples. Many filers have only a handful of trades. A win rate computed on a tiny nis noise, not signal — that's why the sample size is printed next to every number we show.
• Transaction-anchored timing. Backtests anchor at the trade or filing date. A real trader fills at the next open, at a different price. Backtested levels are a ceiling, not an expectation of what a follower would have earned.
• Disclosure lag (congress). The STOCK Act allows up to 45 days between a trade and its public disclosure. Returns measured from the disclosure date are what a follower could have gotten; returns measured from the trade date are what the member got. We label which is which.
• Estimated earnings dates.Earnings dates shown alongside trades are inferred from each company's historical filing cadence. They are estimates, not confirmed dates.

What we don't do

• No consensus estimates— every number is computed from filings and market prices, not analyst forecasts.
• No recommendations— we describe what filers did, never what you should do.
• No performance guarantees— a backtest describes the past, full stop.

MarketPeel is a research tool, not investment advice.