“Should This Loan be Approved or Denied?”: A Large Dataset with Class Assignment Guidelines

Figures & data

Table 1(a). Description of 27 variables in both datasets.

Variable name	Data type	Description of variable
LoanNr_ChkDgt	Text	Identifier – Primary key
Name	Text	Borrower name
City	Text	Borrower city
State	Text	Borrower state
Zip	Text	Borrower zip code
Bank	Text	Bank name
BankState	Text	Bank state
NAICS	Text	North American industry classification system code
ApprovalDate	Date/Time	Date SBA commitment issued
ApprovalFY	Text	Fiscal year of commitment
Term	Number	Loan term in months
NoEmp	Number	Number of business employees
NewExist	Text	1 = Existing business, 2 = New business
CreateJob	Number	Number of jobs created
RetainedJob	Number	Number of jobs retained
FranchiseCode	Text	Franchise code, (00000 or 00001) = No franchise
UrbanRural	Text	1 = Urban, 2 = rural, 0 = undefined
RevLineCr	Text	Revolving line of credit: Y = Yes, N = No
LowDoc	Text	LowDoc Loan Program: Y = Yes, N = No
ChgOffDate	Date/Time	The date when a loan is declared to be in default
DisbursementDate	Date/Time	Disbursement date
DisbursementGross	Currency	Amount disbursed
BalanceGross	Currency	Gross amount outstanding
MIS_Status	Text	Loan status charged off = CHGOFF, Paid in full = PIF
ChgOffPrinGr	Currency	Charged-off amount
GrAppv	Currency	Gross amount of loan approved by bank
SBA_Appv	Currency	SBA's guaranteed amount of approved loan

Table 1(b). Description of additional 8 variables in SBA case dataset.

Variable name	Data type	Description of variable
New	Number	=1 if NewExist=2 (New Business), =0 if NewExist=1 (Existing Business)
Portion	Number	Proportion of gross amount guaranteed by SBA
RealEstate	Number	=1 if loan is backed by real estate, =0 otherwise
Recession	Number	=1 if loan is active during Great Recession, =0 otherwise
Selected	Number	=1 if the data are selected as training data to build model for assignment, =0 if the data are selected as testing data to validate model
Default	Number	=1 if MIS_Status=CHGOFF, =0 if MIS_Status=PIF
daysterm	Number	Extra variable generated when creating “Recession” in Section 4.1.6
xx	Number	Extra variable generated when creating “Recession” in Section 4.1.6

Table 2. Description of the first two digits of NAICS.

Sector	Description
11	Agriculture, forestry, fishing and hunting
21	Mining, quarrying, and oil and gas extraction
22	Utilities
23	Construction
31–33	Manufacturing
42	Wholesale trade
44–45	Retail trade
48–49	Transportation and warehousing
51	Information
52	Finance and insurance
53	Real estate and rental and leasing
54	Professional, scientific, and technical services
55	Management of companies and enterprises
56	Administrative and support and waste management and remediation services
61	Educational services
62	Health care and social assistance
71	Arts, entertainment, and recreation
72	Accommodation and food services
81	Other services (except public administration)
92	Public administration

Figure 1. Heat map, default rates by state (Figure 1 was created using JMP).

Table 3. Industry default rates (first two digit NAICS codes).

2 digit code	Description	Default rate (%)
21	Mining, quarrying, and oil and gas extraction	8
11	Agriculture, forestry, fishing and hunting	9
55	Management of companies and enterprises	10
62	Health care and social assistance	10
22	Utilities	14
92	Public administration	15
54	Professional, scientific, and technical services	19
42	Wholesale trade	19
31–33	Manufacturing	19, 16, 14
81	Other services (except public administration)	20
71	Arts, entertainment, and recreation	21
72	Accommodation and food services	22
44–45	Retail trade	22, 23
23	Construction	23
56	Administrative/support & waste management/remediation Service	24
61	Educational services	24
51	Information	25
48–49	Transportation and warehousing	27, 23
52	Finance and insurance	28
53	Real estate and rental and leasing	29

Table 4. Quartiles of gross disbursement.

Quartiles	CHGOFF	PIF
100% maximum	$4,362,157	$11,446,325
75% quartile	$140,796	$255,000
50% median	$61,962.5	$100,000
25% quartile	$27,767	$49,034
Minimum	$4000	$4000

Table 5. Loans backed by real estate.

	Default	Paid in full
Loans Back by Real Estate (Term ≥240 months)	2472 (1.64%)	147,868 (98.36%)
Loans Not Backed by Real Estate (Term <240 months)	153,876 (21.16%)	573,212 (78.84%)

Figure 2. Status of the loans active or not active during the Great Recession.

Figure 3. SBA-guaranteed portions for paid-in-full and defaulted loans.

Table 6. California-based case study: Information for two loan applications.

Loan	Name	City	Date	Loan amount requested	SBA portion guaranteed	Secured by real estate?
1	Carmichael Realty	Carmichael, CA	Current (not recession)	$1,000,000	$750,000	Yes
2	SV Consulting	San Leandro, CA	Current (not recession)	$100,000	$40,000	No

Table 7(a). California case study: Initial logistic regression model with five explanatory variables.

Parameter	DF	Estimate	Standard error	Wald Chi-Square	Pr > ChiSq
Intercept	1	1.3537	0.3229	17.5729	<0.0001
New	1	−0.0772	0.2101	0.1349	0.7134
RealEstate	1	−2.0331	0.3636	31.2663	<0.0001
DisbursementGross	1	−3.37E-7	3.52E-7	0.9173	0.3382
Portion	1	−2.8298	0.5594	25.5909	<0.0001
Recession	1	0.4971	0.2413	4.2441	0.0394

Table 7(b). Type III analysis.

Source	DF	Chi-Square	Pr > ChiSq
New	1	0.14	0.7130
RealEstate	1	39.96	<0.0001
DisbursementGross	1	0.97	0.3258
Portion	1	27.41	<0.0001
Recession	1	4.27	0.0389

Table 8. California-based case study: Re-specified model with three explanatory variables.

Parameter	DF	Estimate	Standard error	Wald Chi-Square	Pr > ChiSq
Intercept	1	1.3931	0.3216	18.7670	<0.0001
RealEstate	1	−2.1282	0.3450	38.0529	<0.0001
Portion	1	−2.9875	0.5393	30.6898	<0.0001
Recession	1	0.5041	0.2412	4.3679	0.0366

Figure 4. Misclassification rate versus cutoff probability level.

Table 9. California-based scenario: Classification of loans.

	State of nature: Reality
Classification	Loans charged off	Loans paid in full	Total
Higher risk (more likely to be charged off)	31	14	45
Lower risk (more likely to be paid in full)	324	682	1006
Total	355	696	1051

Table 10. California-based scenario summary.

Loan	Name	Date	Loan amount requested	SBA portion guaranteed	Secured by real estate?	Estimated probability of default	Approve?
1	Carmichael Realty	Current (no recession)	$1,000,000	$750,000	Yes	0.05	Yes
2	SV Consulting	Current (no recession)	$100,000	$40,000	No	0.55	No

Table 11. Grading rubric for assignment.

Step (Weight)	Doesn't meet expectations	Approaches expectations	Meets expectations	Exceeds expectations
1) Identifying indicators of potential risk (30%)	Identifies indicators of potential risk that do not make sense	Identifies indicators of potential risk, but fails to provide reasonable justification	Identifies indicators of potential risk and provides reasonable justification as why or why not variables should be considered	Identifies indicators of potential risk, provides reasonable justification as why or why not variables should be considered, and provides supporting evidence (analyses)
2) Understanding the case study (10%)	Understands the case study, but provides an inaccurate and/or confusing synopsis of the business decision in question	Understands the case study by providing a synopsis of the business decision in question, but does not include discussion of dataset limitations	Understands the case study by providing a synopsis of the business decision in question and includes discussion of limitations related to either time frame or selection bias	Understands the case study by providing a synopsis of the business decision in question and includes discussion of limitations related to time frame and selection bias
3) Building the model, creating decision rules, and validating the logistic regression model (50%)	Builds a model that does not make sense	Builds a model that makes sense, but does not create an appropriate decision rule or adequately validate the model	Builds a model that makes sense, creates an appropriate decision rule, but does not adequately validate the model	Builds a model that makes sense, creates an appropriate decision rule, and adequately validates the model
4) Using the model to make decisions (10%)	Derives an inaccurate estimated probability of default for both loans and makes poor decisions for both loans, using their model	Derives the correct estimated probability of default for one or both loans, but makes poor decisions for both loans, using their model	Derives the correct estimated probability of default for both loans and makes a good decision for one loan (but not the other), using their model	Derives the correct estimated probability of default for both loans and makes a good decision for both loans, using their model