The computation of Eurosystem staff projection ranges was changed in autumn of last year.
Instead of deriving them from past forecast errors, these ranges are now based on Bayesian
Vector Autoregressions (BVARs) that account for shock and estimation uncertainty. The
ranges are conditional on assumptions on several exogenous variables such as oil prices,
exchange rates and foreign demand.
In general, model-based projection ranges have two possible advantages over calculating
uncertainty based on past errors. If the sample of forecast errors is small, the uncertainty
about past forecast uncertainty can be large. Moreover, in the case of structural change, past
forecast uncertainty can be a misleading measure of future forecast uncertainty. But the
structural change can be incorporated into a model from which uncertainty is derived.
However, model-based projection ranges also have two major disadvantages. First, the
effects of model uncertainty are ignored. Second, information not contained in the model
variables, which is typically incorporated into the forecast by means of subjective
adjustments by the forecaster, is ignored.
How to best represent forecast uncertainty is still the subject of intensive discussions in the
Eurosystem, and a final decision has yet to be taken. Furthermore, it should be taken into
account that forecast uncertainty may vary markedly over time, so that neither past forecast
uncertainty nor the forecast uncertainty derived from models is always a good basis for
assessing current forecast uncertainty. This is one of the reasons why the forecast
uncertainty conveyed by the ranges of the Eurosystem staff projections may be
supplemented with a qualitative assessment by the ECB Governing Council. (For example, in
December 2008, when year-end uncertainty was rising, the Governing Council concluded
that in its “view …, the economic outlook remains surrounded by an exceptionally high
degree of uncertainty.”).
In addition to uncertainty, forecasts might be subject to upward or downward risks. That is,
realised data might be more or less likely to come in above the point forecast than below the
point forecast. Information about risks can be of great relevance to decision-makers. If risks
to price stability lie on the upside, this could, for example, call for a more restrictive policy
response than risks on the downside.