Modern banks such as the Canadian "Big Six" have been "power
users" of modern methods in mathematics, statistics, and computing
in their financial markets arms for many decades now to the point where
at least Master's level training in a quantitative discipline has become
essential for many working in trading and risk management. But perhaps less
known is the trend towards highly quantitative work in the less glamorous
but highly profitable commercial and retail banking arms of banks.
Commercial bankers must decide to whom to lend, in what amount, and on
what terms, both quickly and accurately. Not everyone gets a mortgage from
a bank, and of those who do, not everyone pays the same mortgage rate. These
decisions are made with a tap of a few keys in a retail banker's office.
How are these decisions made? With Big Data and the big analytics that go
For example, every time you use your credit card, the bank records when,
where, how much, and to whom you are paying. Some of us have had the experience
of getting a call from a bank, concerned that our recent use of a card in
an unaccustomed way or in a remote place is evidence that it has been stolen.
What is less known is the use to which this huge set of credit card data
can be used in finding patterns that suggest that someone's spending is
calling into question their ability to repay their loans.
Even the marketing group of the bank's credit card division gets into the
action. Many consumers hit the mall armed with a small arsenal of credit
cards. What few of us know is the degree of concern that banks devote to
keeping the card they have issued "top of wallet": the first card
you reach for at the point of sale. Banks can assemble, query, and analyze
the Big Data in their databases and use it to decide both how to shed unprofitable
but also how to retain profitable customers, by tying it with various marketing
This 2-day program will consist of a 1.5-day workshop on data analytics
in banking, capped off by a 1-hour research talk. The workshop leader is
Professor Cristián Bravo, Department of Industrial Engineering, Universidad
de Talca, Chile. The workshop deals with modern credit scoring and credit
risk techniques and includes coverage of probability of default (PD), loss
given default (LGD) and exposure at default (EAD). Case studies may include
social network analysis for credit card fraud, micro-credit, and an application
of semi-supervised learning. Participants in the workshop should have a
basic knowledge of data mining and an understanding of how to create and
use statistical models. Previous knowledge of predictive analytics used
in this application area is not required. A 1-hour research talk entitled
"State-Dependent Correlations and PD-LGD Correlation" given by
Dr. Adam Metzler, Department of Mathematics, Wilfrid Laurier University,
will conclude the program.
Online registration has now closed. We regret that
we can not accept any on-site registrations as the workshop has reached
Workshop fees: $200 regular rate, $150 faculty rate, $75 student
and postdoctoral fellow rate.
These fees include workshop participation, lunch and break catering both
days, and reception on Wednesday evening (cash bar).
Funding support: Graduate Students and Post-doctoral Fellows interested
in being considered for a travel award should select and complete the "funding
application form" option when registering. Awards will be given post-workshop.
Cristián Bravo is Instructor Professor at the Universidad
de Talca, Chile. He is an Industrial Engineer, has a Master in Operations
Research and a PhD in Engineering Systems from the University of Chile.
He previously served as the Research Director of the Finance Center at the
Department of Industrial Engineering, Universidad de Chile, and worked as
a Research Fellow at KU Leuven, Belgium.
He has been published in several Data Mining and Operations Research journals,
and edited a special issue at Intelligent Data Analysis (2014). His research
interests cover Credit Risk, specially applied to micro, small, and medium
enterprises, and the development and application of data mining and big
data models in this area.
Wednesday May 20
||Reception at Fields
Thursday May 21
||Talk by Adam Metzler
Adam Metzler, Wilfrid Laurier University
State-Dependent Correlations and PD-LGD Correlation - Modeling and Computation
It is an empirical fact that (i) correlations tend to rise during adverse
economic scenarios and (ii) default rates are strongly positively correlated
with loss-given-default. Unfortunately many of the most popular models used
in risk management applications do not incorporate these phenomena, and
the result is overly optimistic predictions. This talk will consist of two
parts. In the first part we present an empirically motivated model that
allows for state-dependent correlations in linear factor models and contains
both the simple mixture and so-called Random Factor Loading models as special
cases. We derive a number of tractable asymptotic approximations and illustrate
that state-dependence in correlations effectively precludes moderate default
rates, exacerbating both good times and bad. In the second part we discuss
efficient Monte Carlo methods for computing risk measures in a PD-LGD correlation
model developed by Miu and Ozdemir (2006).
Adam Metzler received his M.Math (2004) and Ph.D. (2008) from
the Department of Statistics and Actuarial Science at the University
of Waterloo. He was an assistant professor in the Department of Applied
Mathematics at the University of Western Ontario from 2008-2012 before
accepting his current position as assistant professor in the Mathematics
Department at Wilfrid Laurier, where he is currently Co-ordinator of
Financial Mathematics Programs. He maintains an active research program
in quantitative finance, with an emphasis on problems related to credit
Back to top