Bayesian Modeling of Shock Lapse Rates Provides New Evidence for Emergency Fund Hypothesis

This article considers the problem of estimating shock lapse rates in term life products. Four models are estimated using Bayesian Multiple-Block Gibbs sampling. Goodness-of-fit was compared using weighted average lapse rate fitted error. A simulated data setting was employed to validate the algorithm. Among the methodological contributions to the literature we introduce Bayesian estimation for the lapse rate parameters permitting the identification of parameter distributions as opposed to point estimates. Also we introduce a flexible panel data model accommodating both mixed effects and cross-effects between explanatory variables. The data are weighted in the likelihood function according to their relevance as measured by policy counts. Finally, we utilize a large proprietary dataset of U.S. postlevel premium period term policies that enables superior inference over the parameter estimates. Building on the above improvements and recent data covering the 2008 crisis, we find strong evidence in favor of the Emergency Fund Hypothesis as a driver of shock lapses.