JPMorgan Quantitative Research

This repository contains quantitative models and analysis tools developed during the Forage JPMorgan Chase Job Simulation. The project focuses on two primary financial domains: credit risk modeling for retail banking and commodity pricing for natural gas storage contracts.

Project Modules

1. Credit Risk & Expected Loss Modeling

This module predicts the probability of default for personal loan borrowers to estimate potential financial losses.

• Core Script: credit_risk_model.py builds and evaluates multiple machine learning models, including Logistic Regression, Random Forest, and Gradient Boosting, to identify high risk profiles.
• FICO Quantization: fico_quantization.py implements optimal binning of FICO scores using Mean Squared Error (MSE) minimization and Log-Likelihood maximization through dynamic programming to maximize information about default probability.
• Rating Assignment: fico_rating_assignment.py provides a utility to map FICO scores to discrete credit ratings (1 to 5) and risk profiles, ranging from "Excellent" to "Very Poor" based on historical default rates.
• Feature Engineering: Includes custom logic for Debt-to-Income (DTI) ratios, FICO score binning, and employment stability indicators.
• Calculator: expected_loss_calculator.py provides a production-ready interface to calculate Expected Loss (EL) using the standard formula: EL = PD x EAD x LGD.
• Analytics: Generates comprehensive risk distributions, feature importance plots, and ROC curves to compare model performance.

2. Natural Gas Analysis & Storage Pricing

This module analyzes historical natural gas price trends and prices complex storage contracts using seasonal arbitrage strategies.

• Price Forecasting: nat_gas_analysis.py fits a seasonal model using linear trends and sinusoidal patterns to estimate prices for any future date.
• Storage Valuation: pricing-model.py calculates the value of storage contracts by modeling the revenue from sales minus costs for storage rental, injection, withdrawal, and transportation.
• Trading Interface: quick_pricer.py serves as a simplified script for a trading desk to evaluate specific contract scenarios and return a recommendation.

Tech Stack

• Language: Python 3.x
• Data Science: Pandas, NumPy, Scikit-learn
• Optimization: SciPy (Curve Fitting)
• Visualization: Matplotlib, Seaborn

Project Structure

• credit_risk_model.py: Main model training and evaluation script.
• fico_quantization.py: Module for finding optimal FICO score bucket boundaries.
• fico_rating_assignment.py: Helper script for assigning credit ratings to individual scores.
• expected_loss_calculator.py: Interface for individual and portfolio loss estimation.
• nat_gas_analysis.py: Historical price modeling and forecasting.
• pricing-model.py: Core logic for storage contract valuation.
• quick_pricer.py: Simplified tool for rapid contract pricing.

Getting Started

1. Ensure Loan_Data.csv and Nat_Gas.csv are available in the project directory.
2. Run credit_risk_model.py to train the risk model and export the serialized calculator.
3. Run fico_quantization.py to determine optimal FICO score buckets for the portfolio.
4. Execute nat_gas_analysis.py to view price trends and seasonal extrapolations.
5. Use quick_pricer.py to value a sample natural gas storage contract with custom parameters.

Visualizations

The models generate automated charts to provide financial insights:

• Credit Analysis: Default rates segmented by FICO and income levels.

• Quantization Analysis: Comparison of MSE and Log-Likelihood binning methods for FICO scores.

• Commodity Trends: Historical prices versus fitted seasonal models with one year extrapolations.

---

Developed by Alex Lin as part of the Forage JPMorgan Chase Quantitative Research program.