| Network
of Excellence |
CARiMan |
The
Internet Portal for Computer Aided Risk Management |
Responsible: ITC,
WST
& WUT
& BUTE
Assistance: PRO, NOM,
all
partners
Our objectives in this
project is to research patterns in financial transactions to aid in the
assessment of financial risk and detection of fraud in the following areas:
- Credit card fraud
- Debit card fraud
- Money laundering
The
goal of any pattern recognition systems is to make good predictions for new
data, they have to have enough adaptability and generalisation to be able to
deal with new patterns. It is
essential in the use of pattern detection systems, to have an understanding of
data that creates the patterns, this is particularly important where linear
techniques are used e.g. a rule-based expert system. But it is also very
important for artificial neural networks, although they allow for fewer
restrictions because of their general-purpose nature.
All fraud detection systems
are essentially pattern recognition systems; they are designed to find
fraudulent behaviour using only the fraction of the data that is available and
accessible to the system.
This research is an
ambiguous attempt to build a reference for people involved in financial risk
management, and to encourage research it this area. It is the purpose of this
research to describe and analyse available data with the intention of using it
to form rules or features for patterns in fraud detection systems. In the
process, it will describe the environment of fraud detection.
Moreover, in this JPA emphasis will
be put on Portfolio Management on one hand, Insurance on the other hand.
A key element of innovation we would like to introduce into the project are
beyond current state-of-the-art portfolio management and risk management
strategies. Existing mathematical or stochastical models usually do only compare
different assets of one class or type. What is required here is an extension,
enabling the comparison of different assets belong to different asset classes in
precise mathematical terms. This is to be described as Homogeneous
Portfolio Model. If the overall amount of investments made by a client is
regarded as a portfolio, the following questions are usually raised:
1)
Is the customer's portfolio really tailored according
to his real risk tolerance?
2)
What strategies have to be applied in order to optimise
the portfolio?
3)
Is the portfolio's diversification high enough in order
to avoid certain (so-called non-systematic) risks?
Theories and methods to be applied here can be derived
from Mathematical and Stochastical Portfolio Theory and Risk Management Theory.
Special Programming Techniques include Agent Technology, Monte-Carlo Simulation
and Genetic Algorithms. The goal is here to develop models enabling the
measurement of the real risk tolerance of a client. Such models will have to
rely on recent contributions to economic theory, as made by Kahneman, recent
Nobel Prize winner in Economics, and others. Again such techniques are beyond
current state-of-the-art.