Over the past decades, many countries have exhibited mortality rates approximately following a piecewise linear law. This is visible in the form of steady improvements over multiple years, followed by a rather abrupt trend change, and then again stable improvements according to the new trend. We investigate trend change events across various countries and both genders in detail, revealing cross-country clusters and serial correlations in the trend change signs. This sheds light on how population mortalities may evolve with respect to each other, which can be helpful in setting correlation assumptions in internal models.
Our presentation will examine the challenges arising in setting mortality improvement assumptions, exposing known but under-explored vulnerabilities of current practices.
An entirely new framework is proposed, one built around the characteristics of short-, medium- and long-term outlooks, which puts greater weight on forward-looking (rather than extrapolative) approaches. We will also discuss herding and “group think” and why these might be problematic.