Unemployment Forecasts: Room for Improvement?

During the crisis of 2009–2013 many institutes made large errors in their unemployment forecasts. This paper develops alternative short-run models to improve these forecasts. The models are applied to the Netherlands and compared to the unemployment forecasts of the CPB. A BVAR model performs significantly better than the CPB forecasts. The BVAR can also be used to predict the labor market transition probabilities used in the forecast functions of two-state and three-state stock-flow models of unemployment. This also results in significantly better forecasts. However the combination of these three best-performing models produces the largest gain in forecast accuracy.