#CELRAK10 GITHUB PROJECT PORTFOLIO ON CREDIT METRICS
FOLLOW ME AT
https://github.com/celrak10/CreditMetrics
https://www.linkedin.com/in/ceballosalfonsoeng/
SIMULATION OF A REGIONAL BANK CREDIT PORTFOLIO:
The data set that contains the simulation of the portfolio is called IFRS_Simulation
Feel free to use it, modify and change it
Characteristics of the portfolio:
1. Currently the portfolio is simulated as a healthy 96% performing loans,
2. 20% simple, 80% revolving lines
3. Around 90% of the clients has a credit on the local currency, 10% in USD
4. The NPL performance is measure on Buckets of 0,1,2,3,4 where 0 current and 4 default
5. Activity and Sector is based on NAICS 6 digit and 2-3 digit respectively
6. Some credits have funding program
What columns does the dataset contains? Dictionary of the data set:
Date Date in yyyy/mm/dd format. Dataset spans Jan-2022 to Dec-2024.
DATE_F Numeric year–month yyyymm (month is zero-padded, e.g., 202201, 202212).
Id_client Unique client ID (total of 5,199 clients).
Activity Client’s economic activity, based on 6-digit NAICS code.
Sector Client’s economic sector, based on 2- or 3-digit NAICS code.
SIZE Company size: 1 = Micro, 2 = Small, 3 = Medium, 4 = Large.
Program Whether the client is in a financing program (details to be explained).
NPL Days-past-due bucket: 0 = Current, 1 = 1–30 days, 2 = 31–60, 3 = 61–90, 4 = Default (>90 DPD).
Internal Score Regional bank’s internal score. Only 59–100 appear because lower scores are not credit- approvable.
Currency Currency of the loan.
Loan_Amount_USD Loan amount granted in USD.
Loan_Amount_Local Loan amount granted in local currency (MXN).
PD_reg Regulatory Probability of Default (Based on Anex 22 of Mexican regulatory credit risk framework)
LGD Regulatory Loss Given Default (Based on Anex 22 of Mexican regulatory credit risk framework)
I_rate Annual interest rate of the loan.
Capital Loan principal for the period.
Interest Loan interest for the period.
EAI Total amount for the period = Principal (Capital) + Interest.
Credit_B Total amount of the client’s credit lines at origination.
F_Debt Client’s total financial debt at origination.
Total_Debt Client’s total debt from financial statements (or estimated if statements are unavailable).
Assets Company assets at origination.
Equity Equity = Assets − Liabilities.
RATING Empty; in a future project I will use this column as credit that has external rating.
Cred_Type Credit type: Rev = Revolving, Simple = Term loan.
Seed Loan origination date.
Where the code to run the analysis and what it runs?
The code that runs all the modeling and analysis is FullCode.R
The topics that cover:
1. Modeling
PD by Logit, Random Forest, Neural Networks (Future Update)
AUC-ROC, KS, Brier Score, calibration curve
LGD estimation by beta and Gradient Boosting
EAD linear estimation
Expected Losses for portfolio and client
2. Dependencies (Non expected losses)
Transitional Matrix
Initial Rating
Thresholds for Rating
Simulations for loss distributions UL, Var, ECAP
Gaussian Copula with PD/LGD
Future updates (Probably August-September):
3. Stress Testing
Shocks to interest rates, FX depreciation, Sectorial deterioration
Natural Shocks in Vulnerable Sectors
Recalculation of the EL by shock scenarios
4. Metrics and Dashboarding (Probably BI)
EL, UL, VAR/Es, Ecap
RAROC
Top exposures
Sectorial concentration
Maturity, currency etc
5. Sensitivity Analysis
Internal credit sensitivity analysis
#CELRAK10 GITHUB PROJECT PORTFOLIO ON CREDIT METRICS
FOLLOW ME AT
https://github.com/celrak10/CreditMetrics
https://www.linkedin.com/in/ceballosalfonsoeng/