The average insurer typically utilizes some form of territory ratemaking in its algorithm; thus, in constructing a GLM, one of the major issues revolves around how to reflect location in the statistical solution. The problem arises because there are too many territory categories to directly include in the statistical model. This issue can be resolved by altering the perception of the location dimension from a categorical rating variable to a continuous one.
This paper presents an alternative approach to incorporating the location dimension in the GLM analysis of the rating algorithm. The procedure develops the indicated relativities and boundaries in a statistical multidimensional framework thus removing the distributional effects of other rating variables and measuring the geographic risk alone. Furthermore, the territory procedure is based on the principle of locality, i.e., the expected loss experience at location L is similar to the loss experience around L.
The indicated relativities of each geographic unit are determined by modeling polynomial functions of latitude and longitude in the GLM statistical framework. By expressing the indication in terms of a polynomial the analyst can include location in the statistical model without having to worry about too man), additional parameters.
