Stefano Pedrojetta, MPA, Florian Chatagny, MA, and Nils C. Soguel, Ph.D. Chair of Public Finance, Swiss Graduate School of Public Administration, University of Lausanne, Route de la Maladière 21, Chavannes-Lausanne, 1022, Switzerland
Forecasting revenue and expenditure accurately is a critical point of an efficient fiscal policy. Misforecasting one or the other may lead to strong fiscal disequilibrium. Forecasting of tax revenue is the starting point of the government budgeting process since it sets the available envelope for expenditure. Overestimating tax revenue ex-ante may contribute to create or increase fiscal deficit ex-post, while underestimating tax revenue ex-ante may contribute to create fiscal surplus ex-post. Both situations are inefficient. In Swiss cantons, for many years, measures of tax revenue forecasting not only indicate strong errors but also suggest that these errors tend to be due to underestimation rather than overestimation. This would mean that forecast errors are systematically biased and room to improvement in the forecast of tax revenue and hence in the budgeting process does exist. The aim of this paper is twofold. On the one hand we seek to show that tax revenue forecast errors effectively suffer from systematic underestimation. On the other hand, we try to show how it is possible to improve forecasting using univariate time series regression methods. We relied on new data about tax revenue forecasting errors in the Swiss cantons over the fiscal years 1945-2007. To assess systematic biasedness, we first perform a standard t-test on the mean relative forecast errors to check whether it is significantly different from zero. Then we run a Jarque-Bera test for normality in order to assess the reliability of the t-tests previously used. Our tests show that tax revenue forecasts of a majority of cantons are significantly and systematically underestimated. These results are robust for different time spans and different tax revenue aggregates. This first piece of evidence suggests possibilities of improvement and appeal for further analysis. Thus by fitting ARIMA model on the data, we expect to show that forecasts generated by a simple AR(1) model exhibit lower biasedness or no bias at all. Such a result would give support to the idea that prediction of tax revenue could be easily improved and thus budgeting process and fiscal policy in Swiss cantons become more efficient.
Web Page:
www.idheap.ch/idheap.nsf/go/D4CDB00D24A55AD1C125728A005C6DE5?OpenDocument&lng=en