How accurate are the Husfarm price forecasts?
Every Husfarm price forecast is tested twice: on the series’ own history before it is published, and on the prices that actually followed after it was issued. This page shows both, against the simplest alternative: “the price will not change”.
Record of issued forecasts
The first forecasts were archived in the week of 9 Nov 2026. Their 4-week results will appear here from the week of 14 Dec 2026, and 12-week results about two months later. Until then, the test on each series’ history below is the best available measure.
Test on each series’ own history
For every series, the method is run from each of the last 52 weeks as if that week were today, and the forecast is compared with the weekly average price that followed. Data of 13 Nov 2026.
| Horizon | Forecasts compared | Typical error | “No change” error | Closer than “no change” | Total error against “no change” |
|---|---|---|---|---|---|
| Published forecasts: 811 series | |||||
| 4 weeks ahead | 21,058 | 10.6% | 12.7% | 60% | 15.3% smaller |
| 12 weeks ahead | 19,494 | 15.0% | 22.1% | 70% | 28.7% smaller |
| All tested series: 1,156, including those without a forecast | |||||
| 4 weeks ahead | 31,618 | 8.6% | 9.4% | 51% | 9.9% smaller |
| 12 weeks ahead | 29,474 | 12.6% | 15.7% | 58% | 21.0% smaller |
The published series were chosen because they did well in this test, so their figures flatter the method; the row for all tested series and the record of issued forecasts above are the fairer measures.
What the measures mean
- Typical error
- The median of the differences between the forecast and the actual weekly average price, in percent of the actual price. Half of the forecasts were closer, half further away.
- “No change” forecast
- The latest price, repeated for every week ahead. It is hard to beat over a few weeks, which makes it the fair benchmark.
- Closer than “no change”
- The share of forecasts whose error was smaller than the “no change” error for the same week and series.
- Inside the 80% range
- The share of actual prices that fell inside the published range. If the ranges are honest, it should be close to 80%.