Generalized Ratemaker

What is "State of the Industry"?

For the last 30 years, Generalized Linear Models (GLM) have been a comprehensive approach in the insurance industry for achieving relative rate adequacy. GLMs are a flexible generalization of ordinary least squares regression and incorporate a number of different error distributions. While GLMs have been used in insurance ratemaking since 1972, they suffer from some flaws and limitations when used in conjunction with insurance data.

What are some of the Limitations of GLMs?

  • Reliance on an assumed error distribution
  • Reliance on traditional statistical measures for determining predictor significance on training data, instead of evaluating model predictiveness on validation data
  • An inability to identify key non-linear compound variables
  • They often require a significant amount of time to develop and implement

How to Solve Limitations?

Talon's machine learning technologies overcome these flaws in the GLM framework by using non-traditional statistics to efficiently search for powerful compound variables without relying on statistical distributions. Combining a GLM with the advantages of Talon allows for significant improvement in rate accuracy.

GRM Benefits

  • Fast Import of Data
  • Easy to Use
  • Simplifies the creation of GLMs
  • Corrects GLM flaws
  • Emphasizes Predictive Models
  • Utilizes unique Machine Learning

What is "State of the Art" ?

Talon's Generalized Ratemaking (GRM) is a process that allows the practitioner to develop a complete rating algorithm. It quickly determines an optimal initial GLM and then uses Talon's Complete Segmentation to correct for the known limitations of the model created. The Talon GRM process allows for separate modeling of frequency and severity, providing functions to combine these models into one set of relativities.

Generalized Ratemaking (GRM)

The GRM process combines an efficient method for determining GLMs with Talon's Complete Segmentation.

By incorporating Talon's unique machine learning methods, the GRM process ensures that the non-linear components missed by GLMs are captured in the rate relativities developed. The exact form of the GRM process depends on the business goals being pursued and will use some or all of the following functions along with Talon's Complete Segmentation:

  • Single Generalized Linear Model (GLM)

    This function provides the ability to run single GLMs with a specified model form. These models can be run on either claim frequency or severity. A full range of GLM statistics is produced.

  • Iterative GLM

    This function creates numerous models from the set of possible model forms. Complete model statistics are produced for each model allowing the practitioner to determine the most appropriate model form.

  • Aggregate GLM

    This function takes a specified frequency and severity model, combines these models to determine a predicted pure premium, and uses a GLM to determine the best set of rate relativities.

Generalized Ratemaker Overview