The Predictive Power of the Multinomial Distribution - 2 Practical Examples
In my last two presentations for the IAA (2019 in Capetown and 2021 during the Virtual IAA) I dealt with the premium adjustment process for the flatrated fleet model
"bonus-malus", especially with regard to heightening the profitability of the flatrated fleet portfolio. One open point remains: To predict the loss ratio at the end of the year (on the basis of June figures) which is crucial in predicting the top-line for automatically renewed fleets in this segment. Here, the multinomial distribution (with GLM) comes into play. I intend to showcase how well this approach functions to predict the premium volume next year and, at the same time, to demonstrate how the worsening pandemic in November/December 2020 in Germany almost derailed the predictive power of my model (still, it showed a pretty amazing robustness). This gave me the motivation to apply the same “philosophy” for another flatrated fleet model (new calc) where each year each fleet has to be calculated afresh.