Correlation Radar

Are two markets moving in lockstep right now, or against each other — and is that normal for this pair? The curve shows the correlation of daily moves in a rolling window, across the entire shared history. The longest series reach back to 1970.

Important: an analysis of historical price data, not a forecast and not investment advice. The page reports no significance. The peak of a rolling series is an extreme value drawn from hundreds of overlapping windows — one always exists somewhere. All that is stated is where the current value sits in its own distribution.

Y axis = correlation coefficient from −1 (exactly opposed) through 0 (no relationship) to +1 (exactly aligned). The window counts shared trading days, not calendar days: 63 is roughly three months, 252 roughly a year. The line is thinned for display; every figure is computed on the full series.

All pairs at a glance

Sorted by the strength of the current value in the 63-day window. Percentile says where that value sits in the pair's own history — 95 means it was higher only 5 % of the time. Click a row to show the pair in the chart.

Methodology — how this is computed

Computed on moves, not on prices

Two rising prices almost always correlate highly without that meaning anything — both rise over the years, so the relationship tends towards one. That is the classic spurious correlation of trending series. Only the daily moves answer the actual question: do the two markets move the same way on the same days? Price series enter as logarithmic returns.

Yields are not prices

The yield series are interest rates in percent. A percentage change of a percentage is not a meaningful quantity, and the series can reach zero — a logarithmic return on it would be useless. They therefore enter as the daily change in basis points. Important when reading: a rising yield means falling bond prices. Anyone holding the 10-year yield against a bond ETF should expect a negative sign — and finds one.

Nothing is filled in

Crypto trades seven days a week, the NYSE does not, and XETRA has different holidays again. The calculation uses only days on which both series have a close, and the moves are formed afterwards across consecutive shared dates. Filling missing days would compare a bitcoin Sunday with a Friday close.

Two windows, not one

63 trading days is the main view and corresponds to about three months. It reacts quickly but swings hard. 252 days — roughly a year — is the slower cross-check. A short window on its own is no basis for placing a peak: across hundreds of overlapping windows, a spike is almost bound to appear somewhere.

What the page does not say

  • No p-value and no significance class. You choose the pair — a significance claim about a self-selected series would mislead.
  • Nothing about cause. Two markets can both react to the same third thing without either moving the other.
  • Nothing about the future. Correlations break, and most readily when people rely on them.
  • Nothing about the size of the move. Two markets can be perfectly correlated while one swings by one percent and the other by ten.

Data basis

Daily closing prices from our own database, 21 series across equities, commodities, rates, crypto, currencies and sectors. The universe is deliberately small and fixed in advance: anyone combining freely from several hundred tickers is assembling coincidences. Futures and ETFs are labelled as such, because a roll method or a dividend adjustment produces a different series from the plain spot price.

Frequently asked questions

What does the value actually mean?

It measures how aligned the daily moves were within the chosen window. +1 would mean every day one rises, the other rises too. 0: no discernible relationship. −1: exactly opposed. Values around 0.9 are normal between close relatives — a technology ETF and the Nasdaq hold largely the same stocks.

Why is the percentile more important than the value itself?

Because every pair has its own normal level. A correlation of 0.85 is unremarkable between two equity indices and an exceptional state between gold and the Nasdaq. The percentile says where today's value sits in that pair's own history — and only that is comparable.

Why does the history only run a few years for some pairs?

Because the calculation runs only while both series exist. The 10-year yield reaches back to 1970, ether only to 2017 — so the pair of the two begins in 2017. The shared period is shown for every pair, and with a short history a percentile is worth correspondingly less.

Is an extreme value a signal?

No. The page shows an analysis of historical price data, not a forecast and not investment advice. A rolling window produces hundreds of overlapping observations; a peak among them is not an event but a property of the calculation. Past patterns guarantee no future results.