Blog

Modeling Group Life Mortality Experience: A How-To Guide - Part 1: Introduction

By Jonathan Polon on 2/22/2018

Part 1: Introduction

Introduction

In October 2016, the Group Life Insurance Experience Committee of the Society of Actuaries published the results of the 2016 Group Term Life Experience Study. In June 2017, the Society of Actuaries made available the underlying data at a very granular level.


This blog series will provide a step-by-step walkthrough of how to use this data to create a group life rate manual. These same steps could also be applied to create a rate manual from other datasets, such as a carrier’s own group life experience.

Applications 

There are compelling reasons for a group life insurer to create a model from the SOA group life experience data.

  1. Compare experience to industry: A carrier can run its own exposure data through the model created from the SOA data. The output will be the expected number/amount of death claims, based on industry experience. The carrier can then compare its actual claims to expected. The Actual/Expected ratio can be calculated in aggregate as well as by various splits (age, gender, salary, area, industry, etc.). This will provide the carrier with insight as to how its mortality experience compares to industry.
  2. Rate manual validation: The SOA data can be used as an out-of-sample dataset to validate a rate manual that a carrier trains using its own experience. (Technically, the carrier should first remove data that it contributed to the industry dataset but in practice that will only be material for a carrier that has contributed a very large proportion of the industry data.) This can be especially helpful when considering area and industry factors. There is a trade-off between granularity and credibility. I.e., as we increase the number of area and industry groupings we are left with less exposures/claims in each group and there is concern about overfitting the data (quantifying noise rather than signal). However, if the risk factors in the SOA model are similar to those in a model trained on a carrier’s own experience then the carrier can feel more confident in its model. 
  3. Rate manual augmentation: Smaller carriers may not have sufficient volumes of data to create a credible rate manual from its own experience. A smaller carrier can blend its own data with the industry data (on a weighted basis) and use the combined dataset as a basis for its rate manual.

 

Upcoming Blog Posts 

My intention is to continue to post on a weekly basis until this series is complete. I anticipate the future posts to be: 

  • Exploring the Data
  • Creating the Modeling Plan
  • Developing Base Rate Tables
  • Creating Area, Industry and Salary Groupings
  • Determining Multiplicative Risk Factors
  • Two-way Interactions
  • Model Validation
  • Concluding Remarks 

If you would like to be notified as new posts come online please follow me on linkedin or contact me at the email address at the bottom of this post.

 

Working in R 

The data analysis will be performed, as much as possible, in R (an open-source language for statistical computing) using the R-Studio environment (which is also open-source). This will allow me to share my code and allow you to reproduce the analysis.

R is not my usual environment for data analysis and modeling so I apologize in advance should my code lack elegance and efficiency.

R can be downloaded here: https://www.r-project.org/

R-Studio can be downloaded here: https://www.rstudio.com/

 

Feedback 

I welcome your feedback. Please do not hesitate to send me any questions or comments.

I can be reached at jonathan.polon@seb-analytics.com.