Do Insider and Institutional Signals Actually Work? What Our Data Says

We computed the historical base rate of every signal we surface: thousands of institutional accumulation events since 2020, momentum and volatility signals since 2017, and the honest reason our insider rates are not published yet.

Every investing signal arrives wrapped in a story. Insiders are buying. Institutions are accumulating. Momentum is strong. The story is always plausible, which is exactly why it deserves a number: across every time this signal fired in history, what actually happened next?

We built that number for every signal we surface, and we publish it, wins and losses both, on our evidence pages. Here is what the data says, and just as important, what we refuse to say yet.

Institutional accumulation: real, modest, and honest

Our institutional accumulation signal fires when, among the 500 largest managers filing holdings with the SEC, the number holding a stock rose quarter over quarter while their net dollars in it also rose. Since 2020 that has produced thousands of scored events, measured from the day the filings became public, never from the quarter they describe.

The result is a lesson in what honest signal statistics look like. At the six-month horizon the median event slightly lagged the index, and fewer than half of events beat it. The average, though, was meaningfully positive at longer horizons, because the distribution has a fat right tail: most accumulation events fizzle, a minority run far. A signal like this is not a buy button. It is a shortlist generator whose hit rate you now actually know.

Price signals: nine years of base rates

Momentum, relative strength, low volatility and dividend yield each carry roughly nine years of history across a universe of more than two thousand names. The patterns match what the academic literature has long reported, including the uncomfortable parts: entering the top momentum decile looked brilliant in some regimes and painful in others, and the median entry lagged the index at the six-month mark even while the tails paid. The full distributions, year by year, are on the evidence pages, because a single summary number hides exactly the variation that matters.

Insider signals: why we do not show a rate yet

Here is where honesty costs us a marketing line. Our stored insider filing history begins in 2026. As of this writing, the clustered insider buying signal has closed only a handful of measurable events, far below the 30-event floor we set before any rate appears. So the evidence page for it currently says, in effect: too early, come back as history accrues.

We could have shown a rate computed on a dozen events. It would have been a coin flip wearing a percentage, and the next dozen events would have rewritten it. The floor is not a limitation we apologize for. It is the difference between publishing evidence and publishing decoration.

The combination the desk actually watches

The most interesting setup in our data is agreement between independent groups: company insiders buying in the open market during the same quarter that large institutions added the name. Insiders know the business; institutions move the money; neither reads the other's mail. Our conviction screen lists the names where both fired together, each with the base-rate context of its component signals attached.

A base rate is not a prediction. Past firings do not decide the next one, and every page that shows these numbers says so. What a base rate does is calibrate: it turns a story into odds, and odds are something you can actually reason with. Insight, not advice.