Predicting Government Tax Revenues and Analyzing Forecast Uncertainty

This study examines the sources of uncertainty in predicting government tax revenues in Israel. In the first stage, we estimate a model based on several real and financial macroeconomic variables and identify a significant, stable and highly accurate relation between these variables and tax receipts. Moreover, we find that, given these variables, current tax revenues do not improve the projection of revenues. In the second stage, we test the quality of the model's projections on the basis of available information at the time the budget is prepared; we find that the forecast error based is six times greater than the error based on ex-post projection. These results imply that the forecast error predominantly reflects inaccuracy in the prediction of the explanatory variables and not misidentification of the relations among the variables. In particular, we find that GDP projections tend to be overly pessimistic — especially when they are prepared at times of below-average growth. In the third stage, we ask whether limited versions of the model predict tax revenues better; we find that the removal of the financial variables and the indicator for new-dwelling sales does improve the projection. However, models that are even more limited — based only on lagged tax revenues and a GDP growth forecast — provide less-accurate projections, and the probability that they will lead to significant errors in the construction of the budget is greater than that of the broader models