Retesting weighted indicator balances across horizons

From a live call to a statistical-weight

An indicator can be scored by checking each bullish or bearish call against whether the market later moved that way over a fixed horizon, then converting hits and misses into a period hit-rate.

That hit-rate's statistical significance can be mapped to a 0-to-9 statistical-weight, which is then placed in a bullish or bearish pan whenever the same-horizon indicator is live.

Four horizons in one composite-balance

The same weighting tally is run at 5-, 13-, 26-, and 52-week horizons and then merged into one composite-balance.

The composite-balance is meant to be read as an inverted risk gauge that rises near bottoms and falls near tops, so a turn down after a peak was treated as the buy-side condition and a turn up as the sell-side condition.

A short-horizon cluster and a dated risk map

A cluster of three or four 100% short-horizon readings inside three or four weeks was used as a discrete trigger after earlier episodes had been reviewed against subsequent upside breakouts within a month.

Averaging every horizon balance that already speaks to the same future date yields a dated risk map that updates as new readings arrive and does not state how high or low prices will go. That overlay is the multi-horizon-roadmap.

Refreshing the chi-square-screen

Robustness checking compared each indicator's buys and sells with later market direction at those four horizons, then applied a chi-square-screen to ask whether the hit-rate was distinguishable from chance.

Assigned weights change when the sample is refreshed. A long stretch in which about 75% of weeks were up made downside accuracy harder to establish, which is why a later window with a more even mix of up and down weeks was planned after market structure had changed.

Structural-decay and a repeatable relationship

Indicators lose weight or are removed when the activity they measure fades, is distorted by new trading mechanics, or is redefined by reporting changes, while accidental correlations not tied to market behavior are treated as unreliable. That loss of accuracy is structural-decay.

Ranking rules by total profit can favor one large winning episode amid many misses. A high hit-rate whose chi-square-screen result stays reasonably stable across reruns was preferred as evidence of a repeatable relationship.