Forecasting in macroeconomics: A practitioner's view

Statistical criteria for forecast quality in practice have limited relevance. They can be valid only in the context of a set of untested assumptions which are usually unknown and certainly difficult to judge for the client. Three non-statistical criteria are stressed: logical coherence, economic coherence, and stability. These criteria are best served by the use of structural models. As loss functions are usually unknown and certainty equivalence is unlikely to prevail, a forecaster must enable his client to form his opinion on the uncertainty associated with the forecast. To this end, uncertainty variants and alternative scenarios appear adequate. The robustness and flexibility of policy choices should be tested against different scenarios.