Seminar Details
| Date |
18-6-2009 |
| Time |
15:00 |
| Room/Location |
Aula Conferenze |
| Title |
Computational intelligence techniques applied to general insurance problems |
| Speaker |
Dott. Pietro Parodi |
| Affiliation |
Aon Benfield, London (UK) |
| Link |
https://www.disi.unige.it/index.php?eventsandseminars/seminars
|
| Abstract |
Understanding risk -- that is, assess quantitatively the likelihood and impact of adverse events -- is the core activity of financial institutions (banks, insurance companies, investment funds, regulators, etc) and an important aspect of all commercial and non-commercial enterprises. The usual approach to understanding risk is through classical statistics. However, this framework is known to be inadequate, owing to the ever-changing, "soft" aspects of risk.
This talk argues that understanding risk is an ecological, rather than mathematical, problem, and that significant insight can be gained by casting risk analysis into the framework of what can be loosely referred to as computational intelligence. We will show how some of the most common problems in non-life insurance (pricing, reserving, capital modelling) involve:
(1) Making predictions (learning) from data (typically addressed by generalised linar modelling, regularisation techniques, neural networks, etc).
Example problems: selecting rating factors in motor insurance, territory clustering for household insurance
(2) Dealing with uncertain/soft knowledge (typically addressed by fuzzy set theory, bayesian inference)
Example problems: deriving expected frequency/severity of claims based on incomplete and inaccurate data, using soft knowledge on how the risk has changed over the years
(3) Modelling the market and the interaction between players (multi-agent systems)
Example problems: the impact of regulations, the underwriting cycle
The talk will emphasise the comparison between different techniques (e.g., generalised linear modelling vs regularisation) and the degree to which these seem to be adequate to analyse non-life insurance risks. Given the number of topics involved, this will necessarily be a high-level, broad-brush talk. |
|
|
 |
