Overview
Rob J. Hyndman is a statistician and forecasting methodologist at the Department of Econometrics and Business Statistics, Monash University (Australia). He co-developed the Hyndman-Ullah (HU) functional data approach to mortality and fertility forecasting, which extends the Lee-Carter model to handle multiple latent factors via principal component analysis of smooth functional representations of demographic schedules. He is also the creator of the demography R package, which provides implementations of the LC, LM, BMS, and HU methods.
Key Contributions / Features
- Hyndman-Ullah (HU) functional data method (Hyndman and Ullah 2005/2007): Extends LC by (1) modeling mortality as a smooth function of age using penalized regression splines; (2) using J principal components (pairs of k_{t,j}, b_j(x)) rather than a single factor, addressing LC's single-factor limitation; (3) forecasting the k_{t,j} time series with damped Holt's method via state-space exponential smoothing; (4) accommodating robust estimation for outlier years. Applicable to both mortality and fertility.
- Multi-country comparison (Booth, Hyndman, Tickle, and De Jong 2006): Co-authored the 10-country evaluation showing that HU achieves the lowest mean absolute error in log death rates (0.149 male/female average) of the five methods compared, but that there is no significant difference among LM/BMS/HU/DJT for life expectancy. The evaluation also showed that HU's advantage over simpler methods is partly attributable to the short fitting period rather than the functional data sophistication per se.
demography R package: Implemented the LC, LM, BMS, and HU methods in a publicly available R package, substantially lowering the adoption barrier for functional data mortality forecasting.
Related
Sources