Will this move keep going — one direction — and for how long?
A gauge for sustained, one-directional, multi-day drives, built entirely on a
pre-registered backtest (frozen before anything was tested). Most cells read
“no edge”. That is the finding, not a broken page: 663 tests, 17 survive multiple-testing
correction, and every one of the 17 points the opposite way to the folklore. Nothing here is a
forecast — the page describes what the history measured and stops.
reading the study…Source research/trend-predictabilitypre-registration frozen 2026-08-13 · market data through 2026-08-12 · primary instrument ^GSPC
The gauge — odds of a one-directional drive, by horizon
Each column is a horizon. The big number is the base rate: how often it has
actually happened, unconditionally, since 1970. ?
What the gauge is measuring
loading…
Every predictor, every horizon — where the measurement landed
One row per candidate, one column per horizon. Each cell is a dot on a fixed
scale with the null band drawn in grey: a dot inside the grey band is a measurement that
cannot be told apart from nothing. ?
OutcomeSort
What today's readings actually say
Plain-language state, chosen by the same percentiles the charts draw. Descriptive
only — no sentence here says what happens next.
How one-directional has the market actually been, over time?
Price on top, the study's own forward trend-efficiency series flush beneath it.
WindowHorizon
Indicator pane · forward trend efficiency, as the study computed it
Is the Hurst exponent ever actually high — against its own honest threshold?
Every candidate predictor, drawn over the full history, with the grey band showing
what the same estimator returns on a market with no memory at all.
Scalewindow: shared with the price chart above
Tape colouringwindow: shared with every other time chart on this page
Dealer gamma — not cleared, and not rejected: not testable yet
The honest state of the one candidate the estate most wanted to work.
Companion pane · how the sample is accumulating, and when it becomes answerable
Could this test have found an edge if one were there?
A known effect was injected at seven strengths and the pipeline recovered every
one. That is what makes the null result strong rather than weak.
The full horse race — every predictor, every instrument
Fitted on data to 2014, scored on 2015 onward. Sortable; extremes coloured, the
middle deliberately grey.
Instrument
Does anything survive inside a volatility regime, or in the modern era?
Method, provenance and what is deliberately missing
Every number on this page, and the file it came from