FNCE 5341 University of Connecticut Logit Model for Credit Default Paper

University of Connecticut, MSFRM Program
FNCE 5341 – Financial Risk Management III (Credit Risk)
Group Assignment #1
Jose V. Martinez
Fall 2022
This is group work, which you will perform in groups of up to five students. If for some
exceptional reason you would like to work in a larger group please talk to me first. Your
submission should consist of a Word document and an Excel spreadsheet. The Word
document should contain the names of all group participants, and the answers to the questions
together with an explanation of what you have done and how to navigate the spreadsheet. The
Excel spreadsheet will contain the calculations that you performed in order to answer the
questions.
Your submission must be in my mailbox (jose.v.martinez@uconn.edu) and that of Chi Zhang
(the TA – chi.13.zhang@uconn.edu) by 11pm, September 18th. The subject line should
contain the following information: “Assignment 1”. Late submissions will be penalized, so be
sure that you allow yourself sufficient time to safely email your group’s files. Pay special
attention to the format of your Word document: it needs to be presented in such a way that it
is clear to us how your analysis has been performed.
1) Using the data provided for this assignment in HuskyCT, estimate a Logit model of
individual default probabilities using all variables included in part A of the
spreadsheet. Use any non-linear transformations you consider appropriate. What is the
best model you can come up with? Compute factor coefficients and diagnostic
statistics. Do the factors included in the model help predict default?
2) Using the models you estimated in 1 compute scores and default probabilities for a
borrower defined by the following characteristics: Age=30; Gender=Male;
Employment=3; Housing=own; Saving=little; Checking=moderate; Credit
amount=5,000.
3) Consider now the judgement of a team of analysts who rate borrowers (into 5
categories). These ratings are included in part B of the spreadsheet.
a. Do these ratings have any predictive power over defaults when considered in
isolation?
b. Do they have any incremental predictive power when considered jointly with
the other ratios?
c. Are they sufficient or efficient predictors of defaults? By that, we mean that
the prediction obtained from using these ratings cannot be improved upon by
(adding to the model) any other available information (i.e., the other factors
used in this assignment).
4) When one of the factors being considered for inclusion in a Logit model is itself a
default probability (𝑝 ), it is best to use the logit of that factor (Λ 𝑝 in the model
rather than the raw factor (𝑝 ) itself. Clearly explain why this is the case.
5) Read: “Machine learning models and bankruptcy prediction” by Flavio Barboza,
Herbert Kimura, and Edward Altman, published in Expert Systems with Applications,
Volume 83, year 2017 and answer the following questions:
a. In addition to the five original factors in Altman’s model, which other
variables are considered in this study? Are those extra variables useful in
predicting corporate defaults?
b. Which machine learning models are used in the study? List them all.
c. Based on the results presented in the paper (you can concentrate on the
reported accuracy ratios and areas under the ROC curve if you wish), are you
confident that the machine learning techniques analysed outperform logit
models? How generalizable do you think the presented results are (to different
samples and different, more optimized, logit versions)? Explain (in one
paragraph).