Evaluation of the Variants of the Lee-Carter Method of Forecasting Mortality: A Multi-Country Comparison

Heather Booth, Australian National University
Leonie Tickle, Macquarie University
Len Smith, Australian National University

The Lee-Carter (LC) method of mortality forecasting is now widely used. Enhancements proposed by Lee and Miller (LM) and Booth, Maindonald and Smith (BMS) address the choice of fitting period, the method for the adjustment of the time parameter and the choice of jump-off rates. In this study, the three variants of the method are comprehensively evaluated based on single year data by sex for several countries including Australia, England and Wales, France, Japan and Sweden. Errors in point forecasts by age and time for 1986-2000 are compared across variants and countries. Findings indicate that LM and BMS are superior to LC. Whereas LC and BMS tend to overestimate mortality, LM often underestimates mortality. Uncertainty is similarly compared across variants and countries. Error is decomposed into jump-off bias versus other error, and uncertainty into fitting period versus adjustment method.