GLM Basic Modeling: Avoiding Common Pitfalls

Starting in the 1990's many of the larger US personal lines carriers began to implement predictive modeling techniques in the form of generalized linear modeling (GLM). Because of the early success realized by those companies, the vast majority of companies are now rushing to employ these techniques too. In their haste to keep up with competitors, many companies are making mistakes and not getting the full benefit possible. The following are some of the most common mistakes made by companies beginning to build GLMs: • Failing to get full buy-in from key stakeholders. • Relying too heavily on pre-analysis. • Using loss ratio analysis. • Modeling raw pure premiums for all coverages directly rather than modeling at the component level. • Restricting analysis to variables and groupings in the current rating algorithm. • Misusing offsets. • Treating the predictive model as a "black box". • Limiting the use of GLMs to risk models. This paper will address each of these pitfalls in turn. By being aware of these pitfalls, companies can hopefully minimize the transition period and achieve the full benefits of multivariate pricing as quickly as possible.