Switzerland. One of the outcomes of Basel II is to allow banks to move
away from stringent reserve requirements and rely more on the risk
actually assumed by the individual banks for their specific customers.
However banks must be able to prove to regulators that their risk esti-
mates are well grounded. Accurately calculating the “loss given default,”
that is, the amount of money the bank is likely to lose if a customer
defaults on his loan, for an individual bank's customers can result in
reduced reserve requirements, thereby freeing up capital for other
investment. In one aspect of the accord, banks must maintain 5 years of
what is called “customer default” data from which to build models that
produce a probability of default, where “default” refers to the cus-
tomer's inability to pay back a loan. The results of these models must be
available for auditing and to provide the required proof to regulators
that risk used is based on actual data [BIS 2004] [Wikipedia 2005].
Within insurance, beyond typical customer acquisition and reten-
tion, one goal is to increase the number of policies held by customers.
This can be achieved through the development of successful policy
bundles, as well as cross-sell and up-sell of policies as described in
Section 2.1. Regression techniques can be used to set rates for insur-
ance premiums using customer demographics and psychographics,
and claims history. Setting rates too high results in lost business,
setting rates too low can result in overexposure on claims. Insurance
claims fraud detection is another area where data mining plays an
important role [SAS 2002a].
Within capital markets, data mining can assist with bundling
stocks into a mutual fund portfolio by clustering stocks, yielding sets
of stocks with common characteristics. By assigning stocks to clus-
ters, each cluster can be the starting point for further analysis and
assessment of which stocks to include in a particular portfolio. Data
mining has also been used to perform trader profiling to understand
the type and styles of traders, as well as trader abuse through insider
trading monitoring [NASD 2006].
It is practically a cliché to comment on the skyrocketing costs of
healthcare. Costs are often attributed to inefficiencies in process,
errors, fraud, and generally a lack of knowledge of what treatments
are necessary or appropriate for a given patient. With increasing
momentum, healthcare institutions and health plans are turning to
data mining to solve such important problems and contain costs