Actuarial Projections of Health Care Costs: Accuracy, Bias and Asymptotes
30 Pages Posted: 6 Jul 2011 Last revised: 29 Oct 2011
Date Written: June 1, 2010
Most causes and correlates of short-run (1 to 3 year) fluctuations in health care costs and insurance premiums are unrelated to long term trend or “drift,” and thus pose quite different problems with regard to accuracy and bias. Short-run variation is dominated by lagged inflation, wage growth and underwriting effects, with a substantial random component ( ± 2 percent). As the projection horizon lengthens, error is dominated by systematic components related mostly to underlying economic growth that are harder to observe, leaving a substantial residual (which is often termed “the technology effect.)” Long-term (10 year) projections have reliable correlations with wages, but not with health status or type of insurance. Analysis of the relatively few (less than 20) extant health expenditure projections that were published in advance and hence allow for reliable calculation of forecast error show non-random drift/trend elements almost as large as the annual random deviations. Ten-year projections show errors around 10%, with baseline, inflation and GDP errors usually larger than errors in the health care cost growth trend. EU, OECD and U.S.OACT projection errors for 1985-2005 are compared. Results of a survey of practicing actuaries (n=92) conducted in April 2008, show that a majority expect U.S. cost growth to slow within the next ten years, but still to surpass 25% of GDP before beginning to stabilize. Virtually none of these professional actuaries expect health care costs to decline relative to wages in the foreseeable future but when probed, about half put the probability at less that 1-in-100, but 18% put the possibility of a decline at more than 1-in-10. The “state of the art” in forecasting long run medical spending is assessed in models used by CMS, CBO, and the Society of Actuaries. Tracking medical expenditures by nominal dollar growth and real per capita spending are useful, yet focusing on the share (of wages, laborforce, or GDP) provides the perspective most immediately applicable to policy and capable of providing the most robust long run forecasts. Spending limits appear to be a variable result of politics and budgetary constraints more than morbidity and mortality. Although death and taxes may be certainties, “when” and “how much” are not.
Keywords: Forecasting, Macroeconomics of Health, Projections
Suggested Citation: Suggested Citation