Final Project (Module A)

Data Science for Insurance - a.y. 25- 26

The data for the project can be obtained from any suitable source (e.g. Yahoo Finance, Kaggle, or R packages) and should consist of a d-dimensional dataset (\(d\geq 3\)) of risk factors or risk-factor changes. Examples include financial return series from a portfolio of assets or insurance loss data.

For instance, the following R packages contain financial and insurance datasets:

data(package = "copula")
data(package = "qrmdata")
data(package = "evir")

The analysis should include:

(a) Marginal modelling

Model each component separately. For financial return series, fit appropriate ARMA-GARCH models and validate them through residual diagnostics. For insurance loss data, select suitable parametric distributions capable of capturing skewness and heavy tails. The model choice should be justified.

(b) Dependence modelling

Investigate the dependence structure by using copula functions. Specifically, the characterization of the dependence between the \(d\) components can be addressed by

• estimating association measures and/or tail dependence coefficients;

• fitting one or more parametric copula models;

• selecting the final copula model based on goodness-of-fit measures, graphical diagnostics, and/or statistical tests.

Final Report

Prepare a report of no more than 10 pages containing:

• a description of the dataset and its main characteristics;
• the code used for the analysis;
• relevant tables, graphs, and model output;
• model estimation and validation results;
• comments on and interpretation of the findings.