Universe · liquid500 · returns stored 2005-01-03 to 2026-11-03
Return explorer changed 2026-09-29
Ask a question of ten years of stock prices. Pick which stocks, on which days — by the model's ranking, earnings reports, sector, price patterns or the calendar — and see what their returns looked like next, set against every other stock on the same days.
Looking is not testing. Every question you try is another draw, and among enough of them one will look good by chance — the page counts your looks and says how large a chance result would be. Something found here is a hypothesis to pre-register before it is run, not evidence.
Your question
The address bar holds the whole question, so a link reopens it exactly.
Static copy. It holds the platform’s own answers, computed on 2026-11-04 when the copy was exported, for the page’s six example questions and the view with no conditions. The page opens on the first of its examples, and “Start over” brings them all back. Any other question (a changed condition, holding period, start day, comparison or years) needs the live database.
Distribution
The stock-days in your question, and the return that followed each.
How to read this
Each of your stock-days has one return: the stored target read a set number of trading days after day 0. Held 1 day and starting 1 day after, it is the close of day 1 to the close of day 2. Against its sector is the stock's return minus its sector's median that day (the model's own target); raw is the plain return; against the market takes out the market in proportion to the stock's beta.
The unit is the day. Each day's average over your stock-days is set against the average of every other stock with the same return that day, and the figures average over days — a day with one matching stock counts as much as a day with fifty. The ± is a standard error from the platform's block bootstrap over those days (blocks of 5, or of the holding period if longer, because a 5-day return overlaps the next four days'). t is the average divided by its standard error: how many standard errors it sits from zero. The median, the share above zero and the histogram pool the stock-days and carry no standard error.
A window that starts at day 0's own close cannot be traded when the question uses something computed from that close — the model's rankings, a price pattern, the volume rank — and the page says so. Earnings dates are the reported ones; most are announced weeks ahead, but choosing on a report that had not happened yet is hindsight, and choosing on whether it beat is knowing the outcome.
backtest The frozen model scored on dates it never traded, after the fact. A description of the holdout, not evidence — only live days are evidence.
Over time
Your stock-days against everyone else, year by year and added up day by day — so an average carried by one year, or one month, shows as that.
Year by year difference per day ± 1 standard error
Added up cumulative daily difference
The stock-days
A sample of what your question caught, to check it catches what you meant. Click a bar in the distribution to list every stock-day inside it.
| Day 0 | Symbol | Sector | Return |
|---|---|---|---|
| — | |||