We are often asked about the statistical performance of our forecasting models. In this note, we present the performance of our euro area models for the month-on-month projections. As usual, for the euro area we evaluate model errors relative to the flash release. All statistics shown in this note can be downloaded here.
We began covering the euro area in mid-2023, so the sample remains somewhat limited. Nonetheless, several important signals have already emerged. Below we report the mean, the absolute mean, and the standard deviation of the forecast error. In both cases we show the six-month rolling window (i.e., a six-month moving average). We provide the statistics for both headline and core inflation.
Forecast Errors
Starting with the mean error, the results indicate that—for core inflation in particular—our forecasts are unbiased, with the mean centred around zero. By contrast, we are not yet fully satisfied with the performance for headline inflation, where a slight upward bias remains, roughly 4 basis points per month.
Figure 1. Forecast errors (MA(6))
The recent dynamics of model performance are quite clear: the models appear to have learned from their earlier errors. This is visible in the notable decline in both the absolute mean error and the standard deviation over the past two years. Today, the accuracy of our models has reached a particularly high level. For example, the current standard error for both headline and core inflation is approximately 2 basis points on average—the lowest level ever recorded for our framework, and a benchmark that will be difficult to surpass.
Figure 2. Absolute MoM forecast errors (MA(6))
Figure 2. Forecast standard errors (MA(6))
We will, of course, continue to monitor these statistics and work to further improve our forecasting performance. Looking ahead, we may also present the same set of metrics for our U.S. models.