Modeling based on data information is one of the most challenging research topics in actuarial science for loss reserving and risk valuation. Most of these analyses are based on aggregated data but nowadays it is clear that this approach does not tell the whole story about a claim and does not describe precisely its development. This talk will describe the rationale of the use of claim-level information in combination with statistical learning techniques to demonstrate why it is important and illustrate results using real data examples.
In our study, we compare traditional aggregated techniques (portfolio-level) with individual models (claim-level) using information about each of the payments made for each of the claims in the portfolio, as well as characteristics of the insured. Model performance is evaluated using out-of-sample mean squared error of prediction and mean absolute error based on a detailed dataset from a P & C insurance company. Predictions for future total loss and its distribution, as well as for standard risk measures such as Value-at-Risk and Tail Conditional Expectation, are evaluated and discussed using simulations.