15 - 05 Forecasting Health Expenditure : Methods and Applications to International Databases 2015

This paper examines a number of issues encountered when using standard health accounts data to forecast national health expenditures. In particular, it focuses on measurement issues, model specifications, and a comparison of performance indicators based on commonly used health accounts data from OECD. It assesses the performance of alternative forecasting methods based on three criteria — accuracy, precision, and certainty. Based on these criteria, it assesses the performance of model specifications including univariate (i.e., health spending) and multivariate (e.g., macroeconomic factors), static (e.g., fixed effect) and dynamic (e.g., dynamic panel), and single-equation models (e.g., ARIMA) and system of equations (e.g., VAR). It uses the better-performing models to forecast health expenditures for individual countries. This analysis makes three contributions to the literature on forecasting health expenditures. First, with longer data series, in contrast to some previous papers on health expenditure projections, we obtain a result that is more conventional in the forecasting field — econometric time series models and statistical smoothing models perform better than econometric panel data models. Second, a recent literature review of health expenditure forecasting suggests that with better computing power and more refined data, the future of forecasting is complicated micro models. But modeling and understanding the determinants of expenditure growth (whether using micro data or CGE macro models) require considerably more data and effort, and may still do worse with pure forecasting. This paper confirms this. At the same time, it contributes to the call for more rigorous methods of forecasting, for more transparency, and for better assessment of performance. This analysis can inform both research and policy debate on budgetary planning and fiscal sustainability of health expenditure.