BAN427 Insurance Analytics
Autumn 2019Spring 2019
The topics can be indicated as follows:
- Introduction to Non Life Insurance. Adverse selection and moral hazard - and how to measure this empirically. Background literature: Einav and Finkelstein "Selection in Insurance Markets" and maybe Aarbu (2017) - "Asymmetric Information in Home Insurance".
- Big data in insurance/finance. How to establish a Customer TimeLine, distinguish between static characteristics and triggers/incidence, how to use incidence data in real life predictions. Interesting article: Zhang, Y., Bradlow, E. T., & Small, D. S. (2013). New Measures of Clumpiness for Incidence Data. Journal of Applied Statistics, 40 (11).
- Prediction methods with applications to insurance. Standard "ML tool set" that will be used (logit regression, regression trees, random forest, ensemble methods, etc).
- Prediction versus causation. Causal models, combining ML and causal methods. Papers from Varian (2014) and Athey (several papers) will be relevant.
- How to use randomized experiments in order to improve business processes.
- Reinforcement learning and how this can be applied. Using reinforcement methods can replace the need for randomized experiments and is of this reason very interesting to use.
- How to deploy and maintain many prediction models. Short intro to cloud-based real time datawarehouse platforms, deployment of analytical services etc..
After completing the course students:
- Know how big data and machine learning techniques is used in the insurance industry
- Know how to build, deploy and test models and treatments using randomized experiments
- Can bring insurance problem into a statistical model
- Can analyze and predict important insurance outcomes using machine learning techniques
- Have general knowledge about measuring adverse selection and moral hazard from insurance data
- Know how domain knowledge can be used to extract "causal" knowledge from observational data
- Have knowledge about correlation vs causality - and how to empirically address causality using domain knowledge
7 lectures of 2 x 45 minutes. Anonymous data will be provided for applications of ML methods in the insurance business.
Econometrics - for example ECN402, BUS444 or BAN431.
Requirements for course approval
Mandatory participation in all lectures.
Assignment (group work, 2-3 students in each group).
Pass / Fail.
Python combined Jupyter Notebook, R is optional.
Some suggestions are made in the "Topics" section. A detailed list will be provided in Canvas.
- ECTS Credits
- Teaching language
Autumn. Last week of the autumn semester. Offered autumn 2019.
Adjunct Associate Professor Karl Ove Aarbu, Department of Business and Management Science.