Methodology and sources

Published statistics are shown exactly as the source issues them, unmodified. Anything we work out ourselves is labelled as our calculation on the panel where it appears.

The calculated figures are: yield spreads such as 10Y−2Y, 10Y−3M and 30Y−10Y; curve slopes and spreads between countries at matching maturities; the gap between a two-year yield and its central bank’s current policy rate; percentage changes over a stated window; and inverse and cross exchange rates. Each is simple arithmetic on the published series, with no smoothing, adjustment or modelling applied.

On the comparison page the calculated figures also include: rebasing each series to 100 at the start of a shared window; annualised volatility as the standard deviation of daily returns multiplied by the square root of 252; maximum drawdown as the largest peak-to-trough fall within the window; the largest single-day gain and fall; the share of sessions that closed higher; and Pearson correlation between each pair of daily return series.

Markets keep different calendars. Equities trade five days a week with their own national holidays, futures follow exchange sessions, and cryptoassets trade every day of the year. Comparing them by position in a list would quietly pair a Monday in one series with a Wednesday in another, so every series is restricted to the dates that all of them share. That is why the window is always shorter than the calendar period it spans, and why adding a market with an unusual calendar shortens it further.

Returns are price returns only. Dividends, coupons, storage and the cost of rolling futures contracts are excluded, which understates income-paying assets against those that pay nothing. Annualising by 252 trading days is the equity convention; applied to a cryptoasset that trades 365 days a year it is marginally conservative, which is preferable to running two conventions inside one table.

Correlation is measured on daily returns rather than on prices. Two series that both drift upward will correlate strongly as prices whether or not they have anything to do with each other, and it is the day-to-day co-movement that says whether holding both is diversification or duplication.

There is no single publisher of comparable government bond curves across these markets, so each country’s figures come from its own central bank or finance ministry. That keeps every number official, but it means the curves are constructed differently: the United States publishes a par yield curve, Germany a Svensson-fitted zero curve, the United Kingdom nominal par gilt yields, Japan compound JGB yields, Canada benchmark bond yields and the ECB an aggregate of AAA-rated euro area issuers.

Differences of a few basis points between markets can come from that construction alone. The comparison is close enough to be useful and not close enough to treat as exact, and the construction is stated on the page next to the figures.

Each government also borrows in its own currency. A higher yield is not necessarily a better return to a holder of a different currency, and is not a judgement on creditworthiness.

If a provider is unreachable or returns an error, the affected panel says so explicitly and shows nothing else. We do not substitute a cached figure for a live one without saying so, and we never estimate a value that a source did not publish.

Every panel carries the observation date of the data and the time we last verified it. Where those differ significantly, the data is older than it looks and the timestamps will show that.

It does not rank, score or recommend individual investments, and it does not tell you what to buy or sell. It publishes reference data and simple calculations on it.

Nothing here is a personal recommendation or advice. Figures may be delayed or incomplete, and should be verified against the original source before any decision that depends on them.