LaplaceFig. 4Loading the latest forecast
Live
—
HistoryForecast50, 80 and 95% bands
Calibrated macro & commodity forecasts, scored in the open
Every economy15,000+ series · 200+ countries. Every commodity400+ contracts in six groups. ForecastPr(yt+h ∈ [q0.1, q0.9]) ≈ 0.8, with its odds.
Search any economic series or commodity contract. Laplace Terminal forecasts it, sets bands at 50, 80 and 95% from the series’ own past errors, and scores every forecast against what happened.
One series. One forecast loop.
- 01SearchAny country or contract
- 02PullThe published history
- 03ForecastMedian and three bands
- 04ScoreOut of sample, against no-change
Cached for the next reader. Redone when new data lands.
Series
Try , or . Press / to search from anywhere.
- 15,000+economic series, forecast on request
- 200+countries and economic areas
- 400+commodity contracts in six groups
- —of outcomes inside the 80% band, out of sample
- —core forecasts beat a no-change forecast
Results
Graded on tests the model never saw.
Each model won an evolutionary search over statistical, machine-learning and deep-learning candidates, scored on held-out history. Every forecast since is tested the same way, out of sample.
—median error against a no-change forecast (MASE)
—held-out tests won against no-change
—core series beat no-change
—of outcomes inside the 80% band, on tests left out of its fit
Fig. 2Calibration, out of sample
Each point is the share of outcomes inside a band on tests left out of its fit, averaged over the core series. A band on the dashed line holds what it says.
| Series | MASE | Tests won | 80% held | Grade |
|---|---|---|---|---|
| Loading the scores… | ||||
Table 1. Every scored core series, best first. Errors (MASE) under 0.9 times a no-change forecast’s beat it, 0.9 to 1.1 tie it, and more lose to it. Open a row for its forecast.
Anatomy of a forecast
One series becomes
a forecast you can check.
A real series from the core set, built up one step at a time. Scroll to run it.
01
The history, as published
The series arrives as the statistics office publishes it, one point per period, with no smoothing.
02
The model
The model won an evolutionary search over statistical, machine-learning and deep-learning candidates. It reads the history and nothing else: no narrative, no analyst override.
03
The median path
It returns a median path for the next periods: the value it judges equally likely to be above or below.
04
Three bands
Around the median sit bands at 50, 80 and 95%, fitted on the series’ own past errors. They widen with the horizon, because the further out, the less anyone knows.
05
The score
Every core forecast is tested out of sample against a no-change forecast, which keeps the latest value. Beat it, or the grade says so.
Fig. 3Loading a series
Live data
History
Context read
Median
50, 80 and 95% bands
No-change forecast
Errors as a share of the no-change forecast’s (MASE)—
Outcomes inside the 80% band, out of sample—
Grade—
Live from the terminal
The current forecast, as the terminal serves it.
Four series from the core set, read from the same cache the app uses. Hover the chart for any period; open the app for the other fifteen thousand.
The board
The world at a glance.
Rates, growth and prices in the largest economies, and the energy, metal and crop contracts that move them. Open any row for its forecast.
Loading the board…
Table 2. Board values come from the stored history. A forecast is made the first time someone opens the series, then served from the cache.
Pricing
Pay for forecasts, not seats.
Every plan is a number of forecast requests a day. Reading a cached forecast costs nothing.
01
Free
$010 requests a dayThe core set, the board and any series you open.
02
Pro
$29a month · 100 requests a dayFor analysts who ask for forecasts daily, with exports.
03
Desk
$99a month · 500 requests a dayA desk's worth of series, bulk exports and early access.
04
Team
Pilota pooled allowanceOne budget for the team, shared watchlists, a named contact.
Start with one series
Open any series and read its forecast.
Free for ten forecast requests a day. Every cached forecast is free to read.