Professional Documents
Culture Documents
QMBU 420/520
TERM PROJECT TOPIC PROPOSAL
Gopichand ATHUKURI
Ali Kemal AYDINLIOĞLU
Robert HENZEL
Ömer SARI
The topic of our project is going to be loan default prediction analysis. The aim of the project
is to assess the customers that are likely to default on their mortgage loans and reduce the
adverse selection problem particularly in our business model which is explained below.
The business problem that we target is the existence of risky borrowers who promote moral
hazard. Our goal is to minimize their negative impact on financial accounts. For this end, we
aim to deliver an automated platform which precisely predicts mortgage loan default
probabilities based on an online questionnaire. This concept is going to be applied for
mortgage loan crowd investment, where we provide a platform to mediate between mortgage
loan takers and a crowd of (private) investors, where our business model is commission and
high credibility for selection.
Workflow
Using our web service, potential lenders and borrowers apply to our mortgage loan pool. The
applicants get accepted or rejected based on the online questionnaire that we specifically
provide. This questionnaire involves basic variables such as age, gender, marriage status and
also variables like credit history, education, applicant income and so on… If the applicant is
accepted he has access to the listed loans that are available.
Through our mediating service, people are enabled to get mortgages from individual lenders
(investors) rather than institutions. This promotes a direct P2P scheme of borrowing where
people deal with other non-bank people. Having enabled an investment opportunity for
lenders, we get a predetermined share of the interest as the service charge.
Value for Us
The value for our side is that we claim a portion of the interest rate of the loans and receive
other commissions. Most importantly, through the use of our prediction model, we have few
operational costs which fosters the profitability.
Risks for Us
Our business model crucially depends on the algorithm which classifies borrowers with the
lowest probability of default. This algorithm should be as mistake-free as possible. Also,
accuracy and efficiency should be consistent. Even a little mistake would add unwanted
borrowers to the pool until we can fix it. If we cannot ensure consistent efficiency, platform’s
reputation would rapidly go down. In such case, number of the late loan payments might
exceed reasonable levels. This would create great risk for us.
Data Source
For the business problem which is the problem of adverse selection, we decided to use
external data sources that are generated for similar loan default prediction purposes. With
this, we found a dataset from Lending club. Lending Club opened in 2007 with one simple
mission: create a more efficient, transparent and customer-friendly alternative to the
traditional banking system that offers creditworthy borrowers lower interest rates and
investors better returns. We will use lending club real data from 1/1/2014 to 31/3/2014 which
is more than 100,000 records (rows). We have two files one is for accepted loans and the
other one is for rejected loans. We will combine both the files to make complete dataset.
REFERENCE
http://web.archive.org/web/20140706042617/https://www.lendingclub.com/info/download-
data.action