The current debt crisis has shed light on the importance of accurate fiscal forecasts. In particular, the accuracy of revenue forecasts is central since they set the limit within which expenditure should remain in order to reach fiscal balance. Therefore, forecasting tax revenue accurately is a key step in the implementation of sound fiscal policies. The current paper contributes to the empirical literature on budget predictions by providing new evidence about Swiss cantons. Using data from 26 Swiss cantons over 1944-2010, we apply the method developed by Elliott et al. (2005) to test the rationality of direct tax revenue forecasts. We mainly find that 1) when considering the percent forecast error, loss functions are asymmetric in a majority of cantons, 2) allowing for asymmetric losses, results of rationality tests are substantially altered in the sense that more cantons turn out to produce rational forecasts 3) when considering forecasts of growth rates, almost no evidence of asymmetric loss function is found and finally 4) forecasts of tax revenue growth rate turn out to be rational in a higher number of cantons than forecasts of levels of tax revenue.