Assessing the economic value of probabilistic forecasts in the presence of an inflation target

Non-technical summary

Inflation targeting central banks devote considerable resources to modelling and assessing the future path of inflation, and to understanding previous forecast errors. Nearly all monetary policymakers also issue periodic ex ante inflation event warnings -- telling the public when the path of inflation is expected to move outside “normal” levels. In this paper, we show that strong forecast performance by conventional statistical metrics is not sufficient for a candidate forecasting method to be helpful when the bank wishes to warn the public about inflation events. We use two specific examples to illustrate the importance of this finding for both strict and flexible inflation targeting central banks. The first example considers the UK inflation forecasts published by the Bank of England, operating within a strict inflation targeting regime. The second example evaluates inflation forecasts produced from vector autoregressions, using a sample of New Zealand data, drawn from the period in which the Reserve Bank operated within a flexible inflation targeting framework. In both examples, a conventional statistical analysis would have misled a policymaker about the sensitivity of the forecast performance from an economic perspective. Specifically, for some parameters of the loss function used in our examples, the performance gains over the benchmark models are negligible.

Paper

ENDS

© Scoop Media