Portfolio Risk Analysis in Python

This project implements a structured portfolio risk analysis workflow in Python.
It covers both static and conditional risk models for measuring portfolio Value-at-Risk (VaR) and Expected Shortfall (ES), together with backtesting, multi-day risk analysis, and stress testing.

Key findings

• Normal VaR produces the lowest risk estimates and tends to be the least conservative.
• Student-t and Historical Simulation generate higher VaR and ES estimates, reflecting heavier tails and realized extreme losses.
• VaR violations cluster strongly during stress periods, especially around 2020.
• Multi-day VaR shows that square-root-of-time scaling is a useful benchmark, but it becomes less accurate at longer horizons.
• Stress testing shows that portfolio downside risk is driven mainly by equity shocks.
• Conditional methods such as GARCH-CCC and FHS-EWMA provide a more dynamic view of risk under changing volatility conditions.

Example output

Project objective

The goal of this project is to analyze the downside risk of a diversified portfolio using multiple market risk methodologies and compare their outputs in a clean, reproducible workflow.

Portfolio components

The portfolio consists of six components:

• Microsoft (MSFT)
• Shell (SHEL)
• JPMorgan Chase (JPM)
• S&P 500 index (^GSPC)
• EUR/USD exchange rate (EURUSD=X)
• Loan component (LOAN)

Portfolio weights

The portfolio uses the following fixed weights:

• MSFT: 20%
• SHEL: 15%
• JPM: 15%
• S&P 500: 20%
• EUR/USD: 10%
• LOAN: 20%

Methods implemented

Static methods

• Normal variance-covariance VaR and ES
• Student-t variance-covariance VaR and ES
• Historical simulation

Conditional methods

• GARCH(1,1) with Constant Conditional Correlation (CCC)
• Filtered Historical Simulation with EWMA

Additional analysis

• Student-t QQ-plot comparison for degree-of-freedom selection
• VaR backtesting
• Multi-day VaR (1-day, 5-day, 10-day)
• Stress testing under equity, FX, and rate shocks
• Final comparison across all methods

Workflow

1. Build portfolio return and loss series
2. Estimate VaR and ES using static models
3. Select Student-t specification using QQ-plots
4. Backtest VaR models
5. Compare historical multi-day VaR with square-root-of-time scaling
6. Perform stress testing
7. Estimate GARCH-CCC VaR and ES
8. Estimate FHS-EWMA VaR and ES
9. Compare all methods in final tables and figures

Folder overview

• code/ → Python scripts
• data/ → raw input data and rebuilt working dataset
• figures/ → generated plots
• results/ → generated result tables

Main outputs

Figures

• Student-t QQ-plots
• Backtesting plots
• Yearly VaR violations
• Multi-day VaR comparison
• Stress testing VaR changes
• Final VaR and ES comparison plots

Results

• VaR and ES estimates across methods
• Backtesting tables
• Multi-day VaR tables
• Stress-testing outputs
• GARCH-CCC parameter and matrix outputs
• Final method comparison table

How to run

Install dependencies first:

pip install -r requirements.txt