Projecting Dynamic Life Tables Using Data Cloning

In this paper we introduce a hierarchical Lee-Carter model (LC) specification to forecast the death rates of a set of demographically related countries. We assume that the latent mortality factor of LC is common for all of them, given the linkage among them. On the other hand, hierarchical modeling is usually conducted by Bayesian approach, which has the disadvantage that assumptions on the prior distributions are needed, which are not usually known or obtainable, introducing thus subjectivity in the model when setting these prior distributions. An option to overcome this limitation is provided by Data Cloning, a novel technique raised in the Ecology field that allows approximating maximum likelihood estimates in hierarchical settings. Even though this technique works with MCMC algorithms, it constitutes a frequentist approach, and the results are invariant to the prior distributions. Finally, we apply the methodology to a set of linked countries, getting a very satisfactory forecasting, concluding that it can be used in both private insurance companies and public pensions systems in order to forecast mortality and mitigate longevity risk.