Research Standards
Every number says where it came from.
Five labels
A figure we take from its publisher is never presented as our own work, and our own work never passes for an official statistic. Dataset charts and the tables we calculate from them carry one or more of these labels. An external label names its institution, and says “scenario” where the institution publishes a scenario rather than a forecast.
- Official statistic
- Figures as their publisher releases them: a statistical agency, central bank, international body or port authority. CK changes only units and dates.
- CK proprietary indicator
- An index or composite CK Economics designs and builds from published series, with its method published.
- CK estimate
- A figure CK Economics calculates from published data (a sum, ratio, share or statistic) or estimates where the source publishes none.
- CK forecast
- A CK Economics statistical projection: a conditional estimate from a stated model and the data to date, with an interval, not a prediction of policy or shocks.
- External forecast
- A projection published by another institution, such as the IMF, shown as that institution publishes it.
Required disclosures
Under each dataset chart, and each table we calculate from several datasets, “Sources & methods” opens these items, the forecast ones wherever there is a forecast. Where one isn’t recorded, it says so rather than disappearing.
01
Sources
The publisher of every input, linked to the page or endpoint we collect from.
02
Observation and publication dates
The first and latest period observed, and the date of the source’s figures the chart reflects.
03
Geographic and sector definitions
The area a series covers, in the source’s own terms (a country, a customs district, a port), and the sector it is filed under.
04
Transformations
Whether the source adjusts for seasons, and anything we change or calculate: a unit, a sum of parts, a ratio, a rolling statistic.
05
Forecast assumptions and method
The model, the data it was fitted to and what it assumes, or the institution whose forecast it is.
06
Backtesting
How far the model’s past forecasts missed, against a naive forecast scored on the same periods.
07
Intervals
The range around a forecast and its probability, such as an 80% range, or for scenarios, which scenarios the range spans.
08
Revisions
Whether the source has recently revised figures, and how its revisions reach the site.
09
Limitations
What the figures can’t tell you: preliminary periods, seasonal swings left in, assumptions a forecast can’t escape.
Forecasts are conditional estimates
A forecast states what a model expects if the patterns in the data so far continue. It is not a prediction of policy decisions, shocks or markets, and it misses in some periods by design: that is what the interval is for. Every CK forecast shows its method, the data it was fitted to and when, its horizon and its interval.
Each CK forecast is backtested before it is shown: refitted at past dates and scored against two naive forecasts (no change from the latest figure, and the same period a year earlier). It runs only as far ahead as it clearly beats the better of the two, and its interval is the spread of its own backtest errors. A series that fails at the shortest horizon (no clear skill, a range that doesn’t hold, or too little history) gets no forecast. The scores are in “Sources & methods” under each chart; a projection estimated before its backtest says so there until it is re-estimated. The methodology gives the thresholds. Where the IMF forecasts the same series, we will score against the IMF too.
A forecast published by another institution carries that institution’s name, such as “IMF forecast”, and is shown as published.
Scenarios are not forecasts
A scenario answers “what if”: a chokepoint closes, a port shuts, a tariff rises. It is labelled as a scenario in its sources and methods, with its assumptions, and is not scored as a forecast. Where a publisher’s scenarios frame a range, from low to high, the range is described as scenarios, not as a probability.
The trade simulator follows the same rule: every result is labelled a first-year comparison with no stockpiles, illustrative and not a forecast, and its methodology sets out the model, its sources, its validation against past disruptions and its limits. Its macro mode does the same: every path is the model’s change from the IMF baseline, added to that baseline and labelled an illustrative scenario, with its limits beside it and its checks against central banks’ published responses on the methodology page.
Independence and conflicts of interest
- Sponsors and advertisers never influence conclusions. Nobody who pays for, funds or commissions work can approve or change it before publication, and we carry no advertising.
- Funding is disclosed. Work commissioned or funded by someone else says so at the top, naming who.
- Conflicts of interest are documented. An author with a financial interest, job or advisory role that bears on a piece says so in its disclosure, printed at the top.
Everything here is general economic information. It is not investment advice, and nothing on the site is a recommendation to buy or sell anything.
Revisions and corrections
Sources revise their figures and we follow them; our own errors we correct openly. The methodology sets out how each works, and how to report an error. Questions about these standards go to econkling@ckeconomics.com.