Looking Back to 1991 Economic Forecasting: A Report to the Treasury and Civil Service Select Committee

Abstract:

Originally written in 1991 to advance the formal analysis of forecasting models and methods, the report considers alternative forecasting devices including guessing, extrapolating, leading indicators, surveys, time-series models and econometric systems. Conditional and unconditional forecasts are reviewed and the issue of data accuracy is discussed. The main focus is on macro-econometric model forecasts and their forecast errors, so forecast variances are described. Forecast comparisons using mean square errors across models are criticized, as are methods of pooling disparate forecasts, which violate forecast encompassing. The non-stationarity of economic data is discussed in terms of unit roots and stochastic trends, technical progress and regime shifts. The various sources of forecast error are delineated including: uncertainty about parameters and non-modelled variables; cumulative innovation errors in forecasting endogenous variables; lagged feedbacks onto exogenous variables, parameter non-constancy; incorrect initial values; and model mis-specifications. The report concludes with some recommendations.