1. Risk Group Assignment Algorithm: Develop an algorithm that groups insured based on their observable characterisEcs such as age, accident history, vehicle type, and other relevant factors that emerge from your analysis. This algorithm is an automaEc decision rule, which sorts every new applicant with a certain set of characterisEcs into a single risk class. Referring back to the lecture notes for criteria for good risk classes, jusEfy your risk categorizaEon system.

2. Predic7ng Total Losses: EsEmate the distribuEon of total losses for the next year. This involves esEmaEng parameters for the frequency and severity distribuEon based on historical claims data and simulaEng the probability distribuEon for total losses next year. Using the distribuEons for frequency and severity seen in class, match the first two moments of the empirical distribuEon. Choose one distribuEon for severity and frequency, or get bonus points for checking which distribuEon has the best goodness-offit!

3. Premium Determina7on: Calculate premiums for each risk group by incorporaEng expected losses, while ensuring that the probability of observing claims larger than premiums is less than 0.5%.

4. (aCer you receive the second data set on Oct 31) Evalua7on: At the end of the next year, you will receive realized claims data. Assess the performance of your risk categorizaEon and premium determinaEon strategies by comparing them to realized losses which are recorded in “claim_data_groupX_2025”. These are new policyholders which you classified in January 2025 based on their characterisEcs and your algorithm developed in task #1. a. Calculate the actual premiums charged for each policy based on your algorithm and pricing strategy from steps 2 and 3 and compare them to claims. Don’t adjust the algorithm you already developed based on this new data! This is an out-of-sample test for your algorithm developed on the 2024 data. b. Compute the loss raEo for each risk category. Is the loss raEo stable across your different risk categories, or does it fluctuate significantly, i.e. are there specific risk categories that significantly outperform or underperform your predicEons? What does the stability (or lack thereof) in the loss raEo suggest about the quality of your risk categorizaEon and premium calculaEon?