Uncertainty and forecast precision

Abstract Macroeconomic forecasters typically report a single estimate per time period for each macroeconomic variable. But they rarely provide consumers of forecasts with information about the degree of confidence there is in the forecast or the likely range of dispersion of the actual outcome relative to the conditional forecast. Partly this is due to the non-linearity of the model and thus the cost of producing standard error bands. But there is also the problem that a mechanical stochastic simulation may misrepresent the degree of forecast uncertainty because forecasters use non-model information to produce a forecast which is more precise, at least for the immediate future, than the model alone. In this paper we propose a method for generating standard error bands which gives a truer reflection of the forecaster's uncertainty.