Content

Data science meets actuarial science
- The data science landscape
- Predictive modeling tasks in insurance for life, non-life (property & casualty, P&C) and health insurance products
- Features of insurance data: the frequency-severity setting, and beyond
- Applicable frameworks (IT, regulation, ethics and society).

Part I: data science methods for risk classification

Risk classification for frequencies and severities
- Generalized Linear Models (GLMs): theory, use in pricing, estimation and inference
- Building tarification models with GLMs
- From GLMs to GAMs: flexible effects of continuous and spatial risk factors

Machine learning basics for actuarial work
- Taxonomy of machine learning methods for supervised (regression and classification) and unsupervised learning
- Vanilla loss functions vs actuarial loss functions, model accuracy, bias-variance tradeoff
- Tuning parameters, hyperparameters, model tuning
- Resampling methods: k-fold cross validation, stratified k-fold cross validation

Tree-based machine learning methods
- Decision trees for regression and classification
- Ensembles of trees: bagging, random forests and gradient boosting machine
- Model interpretation: variable importance, partial dependence plot, individual conditional expectation, hunting for interaction effects

Lasso, friends of Lasso and the actuary
- Shrinkage, regularization and sparsity
- Ridge, Lasso, group and fused Lasso
- SMuRF: Sparsity with Multi-type Regularized Feature modelling.

Demistify neural networks
- The basics of neural networks: ANNs, CNNs, RNNs
- Non-insurance applications of (deep) neural networks
- Neural networks for analyzing claim frequencies and severities, Combined Actuarial Neural Networks (CANNs)

Part II: experience rating with credibility models and Bonus-Malus scales

Credibility methods for experience rating
- Aims and challenges
- Classical credibility models: an actuarial cornerstone dating back to the 60s-70s
- Bayesian credibility models
- A statistical approach to credibility models (with mixed models)

Bonus-malus scales
- The scale, Markov chains, transition rules
- Relativities: connecting credibility models and Bonus Malus scales
- BM scales incorporating a priori risk classification

Part III: claims reserving
- Aims and challenges
- Deterministic methods for claims reserving: chain-ladder
- Stochastic claims reserving models: best estimate, standard error, predictive distributions (bootstrapping and Bayesian approach)
- Contemporary topics: Solvency II, reserve risk, micro-level loss reserving.

Course material

- Hastie, T., Tibshirani, R., and Friedman, J. H. (2009, 2nd edition). The elements of statistical learning: data mining, inference, and prediction. Springer series in statistics. Springer.
- James, G., Witten, D., Hastie, T., and Tibshirani, R. (2023, 2nd edition). An introduction to statistical learning. Springer, New York.
- Denuit, M., Maréchal, X., Pitrebois, S. and Walhin, J.F. (2007). Actuarial modelling of claim counts. Wiley.
- Wuthrich, M. and Merz, M. (2008). Stochastic claims reserving methods in insurance. Wiley.

Articles, lecture sheets and R markdowns and Google Colabs will be distributed via TOLEDO.

Format: more information

Weekly lectures, regular computer labs and tutorials.