Stat-628

A collection of Stat 628 assignments

Presentation

6 and a half minute presentation.

Notes for coding

1. Tasks:
   • predict PRSM with multiple regression model with CIs
   • [Executive summary] include factors drive substantial variation in PRSM (esp. a statistically discernible effect but not a practically relevant (significant) one)
   • [Executive summary] introduce a baseline potential borrower by selecting values of each predictor in your final model
   • [Executive summary] include the main drivers of PRSM and indicate which are associated with greater or lesser credit risk relative to the baseline
   • [Technical report] include the removal of any outliers or suspect observations
   • [Technical report] include all transformations and the construction of any new predictors
   • [Technical report] include the procedure used to select the final model
   • [Technical report] include diagnostics to assess the extent to which the usual multiple regression model assumptions hold
   • [Presentation] include drivers for credit risk
   • [Presentation] future development or improvement
2. Data
   • Use training dataset to fit the model (estimate model parameters, perform inference, and check relevant model diagnostics)
     • Split the training dataset into train and dev. train dataset for EDA, and fit the model; dev dataset for perform inference
   • Use the model to predict PRSM in evaluation set
3. Unit of each variable
   • "1 unit changes in some predictors may not be relevant; consider using more realistic or practically relevant changes"
   • normalization, standardization, etc.?
4. Keep track of any references used, and list them in the executive summary.

Dataset

Overall

1. There may be errors in some of its historical data. Pay particular attention to values outside the allowable range of certain variables.
2. Certain predictors affect PRSM in a non-linear fashion?
   • transforming some predictors or creating new predictors by squaring or cubing individual numerical predictors or taking ratios of existing ones?

Response Variable

1. PRSM (y)
   • 2*{amount repaid at 6 months}/{total amount owed}
   • Should >= 0.
   • Expected to be 1. > 1 indicates ahead of schedule, and <1 indicates behind.
   • “discretizing” numerical predictors?

Possible Predictor Variable

2. FICO
   • Ranges 300~850.
   • Poor (300~579), fair(580~669), good(670~739), very good(740~799), excellent(800~850).
   • Information contained in the FICO score may be relevant when dealing with certain borrowers but not others?
3. TotalAmtOwed
   • load + interest
4. Volume
   • Expected volume of credit card transactions per month
5. Stress
   • Ratio of the monthly garnishment to the expected volume of credit card transactions.
6. Num_Delinquent
   • Number of delinquent credit lines.
   • Delinquency occurs when a business is more than 30 days behind payment of a debt.
7. Num_CreditLines
   • Total number of credit lines, including both delinquent and nondelinquent lines.
8. WomanOwned
   • An indicator of whether the business is owned by a woman.
   • 1 if woman-owned and 0 otherwise.
   • woman-owned businesses more likely to pay off their loans on time?
9. CorpStructure
   • A categorical predictor, records whether the business is structured as a sole proprietorship, corporation, limited liability corporation (LLC), or a partnership
   • Corporations may be slower than other businesses at paying back their loans?
10. NAICS
    • 6-digit NAICS code. The North American Industry Classification System provides a 5- or 6-digit code that classifies different industries. For instance, the code for universities and colleges is 611310. You can look-up individual codes at this link.
    • https://www.census.gov/naics/ tells us that the first two digits indicates the industry of the biz.
11. Months
    • The number of months for which the business has been open.
    • Business open longer are more credit-worthy? After a certain point, an additional month of operation has a diminished predictive effect?