How each figure is measured, which signals each asset class is scored on, and where the data comes from.
How each figure is measured
Every figure is measured against the instrument's own history, not against fixed thresholds. A momentum or volatility reading is reported as its percentile within that instrument's own record of the same measure — where today's value falls among every prior observation — so “36th percentile” means below 64% of that instrument's own history, on its own scale. Percentile windows differ by measure: RSI and daily and multi-day returns are ranked over the instrument's full available price history (which begins at different dates — USD/JPY from 1996, most currency pairs from 2003, each stock from its listing); 52-week range position is measured over the trailing 252 trading days; CFTC positioning over its own trailing two years of weekly reports.
Which signals each asset class is scored on
Every instrument is scored on five shared signals: daily % change, 20-day return, RSI, 20-day volatility, and 52-week range position. Beyond those, two signals are conditional on data availability. CFTC positioning — leveraged-money net percent of open interest — requires a listed futures contract with a Commitments-of-Traders feed, which among these columns applies only to foreign exchange. Relative volume — today's volume against its trailing 20-day mean — is scored for stocks, crypto, and energy, where per-instrument volume is available. Precious metals carry neither conditional signal, and are scored on the five shared measures alone. Each instrument contributes only its single most extreme signal to the cross-asset ranking, whichever that is.
Definitions
Definitions, for readers who look: RSI is Wilder's 14-day; ATR is Wilder's 14-day; moving averages are 20-, 50- and 200-day; volatility is the 20-day rolling standard deviation of daily returns; multi-day returns cover 5, 20 and 60 trading days. Returns are arithmetic (simple), not logarithmic. Cross-asset correlation and beta — a pair against the dollar index, a stock against the S&P 500, a metal or stock against the real yield — are computed on weekly returns, because free daily feeds close their bars at different intraday times and daily correlations wash out to noise; weekly resampling removes that artefact. The currency-pair dollar sensitivity is reported over a 26-week window; the stock and real-yield readings over both 26- and 52-week windows, so a recent shift shows up in the shorter window before the longer one. Positioning is leveraged-money net (long minus short) as a percent of open interest, from the CFTC Commitments of Traders (Traders in Financial Futures) report; it is unavailable for the forint pairs, which have no listed contract.
Data sources and timing
Sources: prices from Yahoo Finance (currency pairs, split-adjusted stock closes, cryptocurrencies, energy futures, and precious-metal futures); the federal funds rate, the 2-year and 10-year Treasury yields, and the 10-year TIPS real yield from the Federal Reserve (FRED); the euro-area AAA 2-year and 10-year yields from the ECB Data Portal; positioning from the CFTC; the official precious-metal fixes from the LBMA, logged as an independent cross-check rather than as the reported price; and each central bank's own published schedule for its decision dates. Every line carries its own as-of date, because these datasets update on different cadences — prices each trading day, CFTC positioning weekly, some policy rates monthly — and the report is regenerated automatically every trading day. Historical values may change when a data provider revises or corrects a previously published figure; each stored series records the date it was last revised, and the report recomputes from the current underlying data on every run.
Prices, timing and conventions
Prices, timing and conventions: each figure is an end-of-day value, and the report is stamped and dated by its reference date — the newest fully settled trading session across the instruments covered — not by the wall-clock time the report happens to run; a partial, still-forming bar for the current calendar day is deliberately excluded, so the same underlying data always produces the same report. The daily close means a different thing for each asset class, and the report does not pretend otherwise. Spot foreign-exchange trades continuously 24 hours a day, five days a week, so its daily bar is a snapshot cut at a fixed daily boundary in Coordinated Universal Time (UTC) rather than at any single exchange's closing bell; an overnight change between one day's close and the next day's open is therefore a real cross-session move, not a data error. Precious metals are quoted from CME-group futures (COMEX and NYMEX), which settle at their own contract times; United States equities use the regular-session market close and are split-adjusted with dividends excluded, so a stock split shows a normal daily move rather than a false collapse. Interest-rate and yield series carry the publishing central bank or statistical agency's own reference time. Because these clocks differ, cross-asset relationships are measured on weekly rather than daily returns, and every value is shown with the date it is current as of, so a reader can reproduce any figure from the same public sources.
Common questions
Why percentile rather than a z-score?
A percentile is the empirical distribution function evaluated at today's value, so the real choice is rank-based (percentile) versus moment-based (z-score), and this report is rank-based on purpose. A z-score is only meaningful under a location-scale assumption — effectively normality — and financial series break that badly, with fat tails, skew, and volatility clustering. Under a normal a z-score of 3 reads as a 1-in-740 event, but in a fat-tailed series that magnitude is far more common, so the z-score misleads exactly in the tails, which is the only region an outlier report cares about. A percentile assumes nothing about shape: “99th percentile” always means higher than 99% of that instrument's own history. It is also robust — one freak past value inflates the standard deviation and shrinks every later z-score, but barely moves ranks — and it is bounded 0–100 and directly comparable, which is what lets instruments with very different distributions share one scale. The honest cost is that a percentile keeps rank and discards magnitude: it tells you how rare today's reading is for this instrument, not how large, and it does not adjust for the shape of the distribution.
Why score each instrument on its single most extreme signal?
Taking each instrument's most extreme signal gives one comparable number per instrument with no weighting scheme to defend — any average across RSI, return, volatility and the rest would need arbitrary weights. It answers the actual question, “is anything unusual here?”, with the most unusual thing, rather than diluting one genuine tail event among several mild readings. And it avoids double-counting correlated signals: RSI, 20-day return and momentum tend to move together, so summing them would reward redundancy and let an instrument that is mildly elevated on several correlated measures outrank one with a single real extreme. The tradeoff, stated plainly, is that this ignores confluence — an instrument at an extreme on five signals scores the same as one at the same extreme on a single signal — a deliberate choice for simplicity and defensibility over a composite score.
How exactly is a percentile computed?
For the distribution signals, the percentile is the fraction of the instrument's own history strictly below today's value, plus half the fraction exactly equal to it — a midpoint rule for ties, so a value that ties several past observations sits in the middle of that tied block rather than being counted as fully above or below it. It uses the instrument's whole recorded history for that signal, not a fixed rolling window, so a longer-lived instrument is judged against more data. Not every signal is a percentile, though: the range signals — 52-week price position and CFTC positioning — are scored instead by where today's value sits within its band, from 0 at the bottom to 100 at the top, a current-location reading rather than a historical percentile. Both are then mapped onto the same 0–100 extremeness axis — the score is twice the distance from the midpoint — which is what lets them share one ranking.
Why these particular signals?
The six signals are chosen to cover distinct, largely independent dimensions of “unusual” rather than several views of the same thing: daily % change (single-session shock), 20-day return (accumulated momentum), RSI (momentum extremity), 20-day volatility (dispersion regime), 52-week range position (where price sits in its yearly band), and — where the data exists — CFTC positioning or relative volume. Price-move, momentum, volatility, range and positioning are conceptually different axes, so an instrument can be unusual on one while ordinary on the others, which is what makes ranking by the single most extreme signal meaningful. The choice of these six also reflects what is derivable from free, public data with a clear definition; it is a considered set, not an exhaustive or uniquely correct one.