ACCT3015 Sem2 2023 – Part B
Due date: Sunday 05th of Nov 2023 at 23:59
In accordance with university policy, the following penalties will be applied when the assignment is submitted
after 11:59 pm on the due date:
1. Deduction of 5% of the maximum mark for each calendar day after the due date.
2. After ten calendar days late, a mark of zero will be awarded.
Guidance for Assignment:
PART B (Estimating Bankruptcy Prediction Models) should be submitted as a word document
in electronic form only using the ‘Turnitin’.
No submissions will be accepted via email.
Format: word doc; 12 point font, Times New Roman, paragraphs formatted with 1.5 line
spacing, and with margins of 2.5cm. This assignment must be done in ‘Word’ or other text
format program.
Answers to Part B of this assignment should not exceed 1,500 words. The word count is based
on the word count information as presented in Turnitin and this will be checked. For guidance
on words in excess of the word limit, students are advised to refer to the Business School’s
Policies at: https://business.sydney.edu.au/students/policy. Please note the word count in a
Microsoft Word document and a submitted document through Turnitin can be different. It is
each student’s responsibility to ensure the submitted word document through Turnitin for Part
B does not exceed 1,500 words in total. As a suggestion, you can use the following procedure
to assist you to track the word count of the Microsoft Word version of your assignment:
Open your assignment Word document. Use Ctrl+Shift+G to open the Word Count dialog box
and insert a tick in the “Include textboxes, footnotes and endnotes”. This function is also
available under Review/Word Count.
Please ensure you submit well before the deadline in case there are problems. Turnitin in
Canvas does not automatically email a digital receipt. Once you have successfully submitted
your assignment, a time and date stamp will appear next to your submission. Should submission
problems arise, you should contact the University’s ICT Service Desk.
Your assignment must be appropriately and properly referenced using the American
Psychological Association referencing style (or APA as it is more commonly known). The
University of Sydney library provides helpful guidance on using APA style referencing,
https://libguides.library.usyd.edu.au/citation/apa7
The filename of your submitted assignment should also be your SID.
For Part B, for example, “123456789.doc”.
As a student of the University, you are responsible for taking part in your education in an honest
and authentic manner. It is therefore expected that you take extra care to ensure that there are
no breaches of academic honesty. Your assignments will be manually and electronically
checked for plagiarism (copying). Any perceived breaches of academic honesty will be referred
to the Office of Educational Integrity for further investigation and penalised if verified. You
can read more information on what plagiarism is and how to avoid plagiarism from the
University link: https://sydney.edu.au/students/academic-dishonesty.html
All sources of support for the assignment must be acknowledged. Failure to acknowledge such
support may potentially breach the University’s academic honesty requirements.
If necessary, students are advised to refer to the special consideration advice contained in the
Administration Manual for Students on the Faculty website:
https://business.sydney.edu.au/students/policy
Grade Descriptors:
Due to the specific nature of this assignment, students are advised generic submissions will not
score high marks as the assignment requirements will not be met. The following can at best be
considered as a general indication, assuming consistency in quality of component parts.
High Distinction The attempted solution substantially exceeds minimal requirements, and is
consistent and coherent. It excels in substantive content and demonstrates professional
communication skills. It reflects consideration of your own well developed perspectives and is
framed in your own words.
Distinction The attempted solution substantially exceeds minimal requirements across most
tasks and through effective communication demonstrates a strong understanding of the
accounting issues and relevant accounting requirements. Good development of your own
perspectives.
Credit The attempted solution exceeds minimal requirements across most tasks, but is less
well developed, less framing in your own words (e.g. dull reproduction of slabs of text from
others), some errors or omissions in application.
Pass A minimal standard is reached across various aspects of the task.
Fail The task is not completed. The solution does not reflect an appropriate understanding
of the question/assigned task. Poorly developed responses.
Questions about the Assignment:
Students are required to post all assignment queries to the Assignment Forum on Canvas or
in ED. This is the required form of communication as it provides an opportunity for all
students to have access to the same information.
PART B: Predicting bankruptcy.
Note, that the 25 marks we award in this part of the assignment, will be rescaled down to 12.5
marks. Refer to the general instructions for further information on word count and submission
requirements.
INSTRUCTIONS:
1. Consider the corporate bankruptcy prediction problem, and the list of variables in
Appendix A. These variables are currently used to predict the probability of
corporate failure for a firm, using annual financial and market data. Recommend
three key additional variables, which are likely to be associated with the prediction
of corporate bankruptcy and hence likely to improve the predictive power of the
machine learning models. You should outline new variables that can be created
from the existing variables (such as financial ratios), as well as external data sources
you can collect from different Big Data sources. Justify why you believe they would
be appropriate explanatory variables. (5 marks)
2. Recommend and discuss methods of exploratory data analysis that you would
undertake to understand and visualize the data. Explain and motivate your
choices. (5 marks)
3. Provide a summary analysis of the predictive performance of the CART model
outlined in Appendix B, and an analysis of which variables contributed most to the
model’s predictive power. Interpret the CART decision tree. (5 marks)
4. Compare results of the gradient boosting model outlined in Appendix C.
Describe and explain any difference in model performance. Identify the role and
impact of predictor variables in the CART and gradient boosting models. (6
marks)
Presentation and Referencing (additional 4 marks)
• Additional marks are awarded for the presentation of your report, e.g., cohesiveness, quality of
discussion, references (APA 7th style). Your reference list will not be included in the word
count.
Appendix A: Covariates included in models
# Variable Name Variable Type Description
1 act Numeric Current Assets - Total
2 am Numeric Amortixation of Intangibles
3 ap Numeric Accounts Payable - Trade
4 at Numeric Total Assets
5 bkvlps Numeric Book value per share
6 capx Numeric Capital Expenditures
7 ceq Numeric Common/Ordinary Equity
8 ch Numeric Cash
9 che Numeric Cash and short-term investments
10 chech Numeric Cash and cash equivalents Increase/Decrease
11 csho Numeric Common shares outstanding
12 dlc Numeric Debt in current liabilities
13 dltt Numeric Long Term Debt - Total
14 dt Numeric Total Debt including current
15 dvc Numeric Dividends Common/Ordinary
16 dvt Numeric Dividends - Total
17 ebit Numeric Earnings Before Interest and Taxes
18 ebitda Numeric Earnings Before Interest and Taxes, Depreciation and amortization
19 fincf Numeric Financing activities net cash flow
20 gdwlam Numeric Goodwill amortization
21 ivncf Numeric Investing activities net cash flow
22 lct Numeric Current liabilities - total
23 lt Numeric Liabilities - total
24 mkvalt Numeric Market value - total
25 ni Numeric Net Income (Loss)
26 oancf Numeric Operating Activities Net Cash Flow
27 optvol Numeric Volatility - Assumption (%)
28 prcc_f Numeric Price Close - Annual - Fiscal
29 prch_f Numeric Price High - Annual - Fiscal
30 prcl_f Numeric Price Low - Annual - Fiscal
31 revt Numeric Revenue - Total
32 sale Numeric Sales/Turnover (Net)
33 spce Numeric S&P Core Earnings
34 spcsrc Categorical S&P Quality Ranking - Current
35 teq Numeric Stockholders Equity - Total
36 tfva Numeric Total fair value assets (TFVA)
37 tfvl Numeric Total fair value liabilities (TFVL)
38 txp Numeric Income taxes payable
39 wcap Numeric Working capital (balance sheet)
40 xint Numeric Interest and Related Expense - Total
41 xintd Numeric Interest Expense - Long-term Debt
42 xrd Numeric Research and Development Expense
43 xsga Numeric Selling, General and Administrative Expense
Appendix B: CART Model Results
Best Model over 5-fold CV
Decision Tree Parameter Value
Class Weight Balanced
Splitting Criterion Gini
Max Depth 4
Min Samples Leaf 2
Min Samples Split 4
Decision Tree Plot
Please zoom in for further details, to interpret the tree.
Variable Importance
Confusion Matrix – Test Set
Predicted Class
0 1
Observed 0 482 86 568
Class 1 6 4 10
488 90 578
Model Performance Measures – Training and Test Set
ROC Curve
Appendix C: Gradient Boosted Tress Model Results
Model Summary: Model error measures
Name Learn Test
Average LogLikelihood (Negative) 0.04574 0.07497
ROC (Area Under Curve) 0.97532 0.84721
Variance of ROC (Area Under Curve) 0.00006 0.00293
Lower Confidence Limit ROC 0.96001 0.74106
Upper Confidence Limit ROC 0.99062 0.95335
Lift 8.80952 7.00000
K-S Stat 0.83655 0.61623
Misclass Rate Overall (Raw) 0.01649 0.01715
Balanced Error Rate (Simple Average over classes) 0.10310 0.21108
Class. Accuracy (Baseline threshold) 0.88932 0.87479
Fraction Data Used after influence trimming 0.35286 n/a
Variable Importance
Variable Score
CEQ 100.00 ||||||||||||||||||||||||||||||||||||||||||
MKVALT 97.94 |||||||||||||||||||||||||||||||||||||||||
SALE 95.39 ||||||||||||||||||||||||||||||||||||||||
NI 82.95 |||||||||||||||||||||||||||||||||||
EBIT 78.19 |||||||||||||||||||||||||||||||||
XSGA 58.68 ||||||||||||||||||||||||
CAPX 50.34 |||||||||||||||||||||
DVT 46.20 |||||||||||||||||||
OANCF 40.27 ||||||||||||||||
Summary for 89 Trees - Gains Chart - ROC, Sample: Full sample, Target class: 1
Confusion Matrix - Test
Actual
Class
Total
Class
Percent
Correct
Predicted Classes
0
N = 1085
1
N = 200
0 573 74.52% 427 146
1 10 70.00% 3 7
Total: 583
Average: 72.26%
Overall %
Correct:
74.44%
Specificity 74.52%
Sensitivity/
Recall
70.00%
Precision 4.58%
F1 statistic 8.59%
Note: Bankrupt (class) was coded as 1 and non-bankrupt as 0.