What has this stock done before, when it looked like this?
Enter a ticker. We take its last six months of shape, search its entire history for the closest matches, and show what happened over the three months that followed each one.
Before you read the chart
—
—
Scroll the chart sideways to see the full window.
| Match | Period | Similarity | Next 3 months |
|---|
Past price paths only. Not a forecast, not investment advice. The measured accuracy above is what this method achieved on 3,968 out-of-sample forecasts.
How this is measured
The temptation with a chart like this is to only ever look at it. So we tested it instead: walk forward through thirty years of daily prices, and at each date find the matches using only data available on that date — a match is admissible only if its own three-month outcome had already happened. Then compare what it implied against what actually occurred.
- 3,968 forecasts across 128 dates whose outcome windows never overlap, so no result is counted twice.
- Significance clustered by date, because every stock moves with the market and a hundred tickers on one day is nowhere near a hundred independent tests.
- A random control — the same pipeline with the matching switched off. It scored the same, which is how we know the pattern matching adds nothing.
- A known-effect check — 12-month momentum shows up on this same data at t = +2.3, so the test is capable of finding a real signal when one exists.
Code: api/_analog.py,
research/analog/validate.py.