Below we show the model parameters resulting from both the Algorithmic Pricing Approach (left) and the Traditional Approach (right). Parameters estimates are on the linear predictor (log relativity) scale therefore a parameter of zero indicates no effect. To compare the two approaches to model building we compare the left and right charts. Each column from 1:10 represents an iteration of the test. For example in the first test iteration the Algorithmic Approach found Rating Area to be significant but the Traditional Approach did not.
The Algorithmic Approach was able to detect much smaller effects than the Traditional Approach. This is observed by comparing the amount of blue cells in both charts. The Traditional Approach found fewer significant relationships, but the relationships identified were quite large. The Algorithmic Approach detected the same effects as the Traditional Approach, but shrunk their effect towards zero