How to Read a Base Rate (Without Fooling Yourself)

A base rate answers one question: across every time this setup appeared in history, what actually happened next? It is the difference between "momentum stocks do well" and "stocks entering the top momentum decile beat the index 46% of the time over the next six months, with a median result 2.5% behind it." The first is a story. The second is something you can reason with, including reasons to be unimpressed.

Moneyta's evidence pages publish these numbers for every signal we surface. This guide is about reading them well, because a base rate misread is worse than no base rate at all.

The three numbers on every page

• Times fired is the sample size, and it is the first number to check. A 60% hit rate over 40 events and one over 4,000 events are different objects entirely.
• Hit rateis how often the stock beat the index after the signal fired. Note the benchmark: beating zero is easy in a bull market, so everything is measured against the S&P 500 over the exact same days.
• Median excess is the typical outcome: line all the events up and take the middle one. It is deliberately the headline instead of the average, because averages hide what tails do.

Why the median and the average disagree

Many real signals show a negative median and a positive average at the same time. That is not a contradiction. It means the typical event slightly lost to the index while a minority of events won big: a fat right tail. Our institutional accumulation signal looks exactly like this. Read that shape as "this signal generates a shortlist in which a few big winners hide," not as "this signal picks winners." The full distribution and the year-by-year table on each evidence page exist precisely so you can see the shape rather than trust one number.

Year-by-year is where regimes hide

A signal's all-history rate blends bull years, bear years and sideways years. The year-by-year table un-blends them. Momentum entries hit 58% in 2020 and 41% in 2021 in our data. Same signal, different world. If a signal only worked in one regime, the table shows it, and you should weigh it accordingly rather than assuming the average regime is the one you are in.

The fine print that is not fine

• Survivorship:our percentile signals rank against today's research universe, measured backward. Companies that delisted along the way are not in the ranking. Every evidence page says this in its depth note.
• Observability: events are dated when the information became public, not when it happened. A signal based on quarterly holdings filings fires at the filing date plus one day, never at quarter end. Backdated signals are how studies lie politely.
• Overlap: a stock that stays in a top decile for a year is one entry event, not twelve. Counting membership monthly would manufacture sample size out of autocorrelation.

When we refuse to show a rate

Below 30 closed events, our pages show the event count and a note instead of a rate. A percentage computed on a dozen events would change completely with the next dozen, which makes it decoration, not evidence. Our insider signals currently sit below this floor because our stored filing history begins in 2026. The floor is the feature: when the rate does appear, it will mean something.

What a base rate is not

A base rate is not a prediction. Past firings do not decide the next one. It is calibration: if someone tells you a signal is powerful and its own history says the median event lagged the index, you now know how much of the story was costume. Use base rates to size your skepticism, not your positions. Educational analysis of published data. Insight, not advice.