Test a market belief in one sentence
Every trader carries a set of beliefs. Buy the breakout. Sell in May. Fade the gap. Buy when the VIX is above 30. Most were picked up from a book, a mentor or one memorable trade, and most have never been tested, because testing one used to mean a data vendor, a notebook and an afternoon.
Explore › Thesis makes that a sentence. Write what you believe, for example "stocks that gap down 5% on earnings keep falling over the next month", and EquationDB does three things.
- It writes the test. The belief becomes one
OUTCOMESstatement: the condition, the universe, the start date and the horizons. A second statement splits the same claim by decade, sector or market. The engine's own parser checks both before anything runs, and a statement that fails is rewritten, not shown. - It runs the test on stored history, decades of it for most markets, against the same symbols' ordinary return over the same years.
- It grades the result with one fixed rule, the same for every thesis. The language model writes the test; it never grades it.
One rule for every verdict
The edge is the average return after the signal minus the ordinary return, at the horizon the thesis names. t is the edge divided by its standard error. A thesis is supported when the edge points the way it says with t of 2 or more, mixed between 1 and 2, and not supported otherwise, with the reason given: no reliable edge, the opposite happened, or too few occurrences (fewer than 15). A supported edge that holds in fewer than half the decades or sectors is downgraded to mixed.
Nothing in that rule depends on how persuasive the thesis sounds.
You always get the statement
The verdict comes with the statement that produced it. Open it in Query and change the universe, the start date or the horizon; the answer is a starting point, not a ruling. That is the difference from asking a chatbot: the evidence is a query you can rerun tomorrow, not a paragraph.
OUTCOMES AFTER gap_pct <= -5 IN top500 SINCE 2005-01-01 FWD 5d, 20d, 60d GROUP BY decade
259 theses to start from
The page also holds a catalog of 259 practical theses in 15 categories: momentum, mean reversion, trend, volatility, volume, gaps, the calendar, earnings, insiders, valuation, macro, rotation, chart patterns, rare events and weather. Each was run on production before it was added, and each runs again on today's data with one click.
Read the result with care
Universes are today's members applied to all of history, so companies that failed are missing and edges after crashes look better than an investor would have had. Valuation fields are point-in-time from the day they were first stored. And one test is one view of a belief: the statements are there so you can try the others.
Open Thesis · How the verdict works