quarterly · % YoY

Real GDP

Where this figure stands, where it is expected to go, and how far that expectation has been off before. Up to 4 countries at once; whatever you choose stays in the address, so the view is the link.

United States's economic output was 2.1% in Q2 2026, and a year out it is expected to be higher than that.

Q3 2026

1.7%

1.5% to 1.9%

Usually within 0.9 of the outcome at this range

Q2 2027

2.6%

2.1% to 3.1%

Usually within 1.6 of the outcome at this range

Even if nothing moved at all, the nearer figure would come out at 1.0%. A year-on-year rate compares with a month that is already history, so that much is arithmetic — what is still open is how much gets added on top.

The ranges are not a feeling about confidence. They are how far this method has actually been off before: half of past outcomes landed inside them, and half outside. The nearer one is tighter because the near future genuinely is more knowable than the far one.

% YoY. Sources: United States, India — OECD. Retrieved 2026-10-13. Everything after the dotted line is a forecast, not a record. The shaded band is where this method has landed half the time in the past — so half the time, outside it. No language model produces any of these numbers.

Take it with you:

How far off it has usually been

Every method above was replayed from each of the last three years' starting points, seeing only what was known at that point, and scored against what actually happened. That is the one number worth carrying away: how far this forecast has usually landed from the truth. The table breaks it down by method for anyone who wants to see which one earned its place.

The forecast drawn above has landed within 0.88 pp of the outcome on average a quarter ahead, and within 1.58 pp a year ahead.

How far each forecasting method has been off in the past, and how often its stated range held.
Method	Usually off bya quarter ahead	Usually off bya year ahead	Its range heldit claims 50 in 100	vs. the simple rulebelow 1.00 is better
Exponential smoothing	0.86 pp	1.58 pp	6 in 10	0.69
The middle of all methodsthe one drawn above	0.88 pp	1.58 pp	6 in 10	0.69
Theta	1.19 pp	2.29 pp	6 in 10	1.03
Automatic ARIMA	1.08 pp	2.48 pp	6 in 10	1.05
Assume last year repeats	3.06 pp	3.07 pp	5 in 10	1.38

United States, measured over the last 33 starting points. The last column compares each method against the simplest rule there is — assume this period repeats what it did a year ago. Below 1.00 means smaller errors than that rule, above it means worse. All of it is measured against today's revised figures rather than against what was known at the time, which flatters every method here a little.

Where the methods disagree

A year out, the methods stop agreeing. Here is where each one lands on its own, run on the same numbers, with the range each would not be surprised by. The spread is not a flaw to be hidden and it is not a reason to discount the forecast: it is the honest size of what nobody knows yet, and it is why the chart above draws the middle of them rather than picking a favourite.

United States, Q2 2027. The dot is what each method expects, the bar how far it would not be surprised to be wrong. They are run on the same numbers and still land 2.7% apart, which is the honest size of the disagreement — and the reason the chart above draws the middle of them rather than picking a favourite.

How this is measured

The stored quantity is the price index; the rate shown here is derived from it against the same month a year earlier. That direction matters: deriving the rate keeps base effects intact, where modelling the rate directly would treat them as noise.

Countries come from different providers with different publication lags, so the series do not all end in the same month. Each one is charted to its own last observation rather than trimmed to the shortest, and the source line names the provider.