This paper documents the presence of systematic bias in the real GDP and inflation forecasts of private sector forecasters in the G7 economies in the years 19902005. The data come from the monthly Consensus Economics forecasting service, and bias is measured and tested for significance using parametric fixed effect panel regressions and nonparametric tests on accuracy ranks. We examine patterns across countries and forecasters to establish whether the bias reflects the inefficient use of information, or whether it reflects a rational response to financial, reputational and other incentives operating for forecasters. In several G7 countries Japan, Italy, Germany and France there is evidence of a change in the trend growth rate. In these circumstances, standard tests for rationality are inappropriate, and a bias towards optimism in the consensus forecast is inevitable as rational forecasters learn about the new trend. In all countries there is evidence that individual forecasters converge on the consensus forecast too slowly. However, the persistent optimism of some forecasters, and the persistent pessimism of others, is not consistent with the predictions of models of rational bias that have become popular in the finance and economics literature.