Health spending of 19 OECD countries is forecast for 1965-1987, with 1980-87 used to test ex ante forecasts made from models calibrated on the earlier period. The mean absolute error (MAE) of percentage growth rate forecasts was 3.0%. An exponential smooth with α=0.3 reduced MAE by 15%. Using last year’s international reduced MAE 16%. An econometric model forecasting health spending as a function of lagged GDP using the own-country time series reduced MAE 8%, while estimating internationally across all 19 countries reduced MAE 17%. A combination of all four forecasts was better than any one alone, reducing MAE 25%. Ajusting the combination weighting to reflect past accuracy did not improve performance. Applying the historical trend in share of GDP spent on health to GDP forecasts occasionally worked well, but usually produced less accurate forecasts. Box-Jenkins ARIMA methods worked well for only one country, France.
