Forecasting GNP with noisy data: a case study