Skip to content

adityarohatgi11/Master-Thesis-Project

Repository files navigation

Carbon Price Shocks and ETF Market Dynamics

A Comprehensive Analysis of Green vs. Brown Investment Responses

Project Status Analysis Period Extended Analysis License


🎯 Executive Summary

This project provides groundbreaking insights into how carbon price shocks differentially impact green versus brown Exchange-Traded Funds (ETFs). Using advanced econometric methods including causal forest analysis, event studies, and machine learning approaches, we demonstrate that traditional energy ETFs are 3x more sensitive to carbon price volatility than clean energy ETFs.

🚀 Key Discoveries

  • Brown ETFs (XLE) exhibit 13.5% average treatment effects vs. 4.2% for Green ETFs (ICLN)
  • LSTM models outperform GARCH by 30% in volatility forecasting
  • EU carbon policy announcements create measurable, asymmetric market responses
  • Clear market differentiation exists between sustainable and traditional energy investments

📊 Research Overview

🎯 Primary Objective

Quantify and explain the heterogeneous responses of green and brown ETFs to carbon price shocks, providing empirical evidence for climate-finance market segmentation.

🏢 Assets Under Investigation

ETF Name Focus Holdings ESG Profile
ICLN iShares Global Clean Energy ETF Green/Renewable Energy First Solar, SSE, Iberdrola, Vestas Mixed (Low to High ESG)
XLE Energy Select Sector SPDR Traditional Energy ExxonMobil, Chevron, ConocoPhillips High to Severe ESG Risk

📈 Data Sources & Coverage

  • Carbon Prices: EU Emissions Allowances (EUA) futures
  • Market Data: Daily ETF prices and returns (2018-2024)
  • Control Variables: VIX, Brent oil, 10-year Treasury rates
  • ESG Data: Morningstar sustainability ratings
  • Policy Events: EU carbon legislation timeline

🔬 Methodology & Innovation

🧮 Advanced Econometric Approaches

1. Carbon Shock Identification

Shock Detection: |AR(5) residual| > 1.5σ threshold
Validation: Cross-referenced with EU policy announcements

2. Causal Forest Analysis

  • Heterogeneous Treatment Effects: Conditional Average Treatment Effects (CATE)
  • Robustness: Bootstrap confidence intervals and placebo tests
  • Innovation: Machine learning approach to traditional finance problems

3. Event Study Framework

Event Windows: [-10, +10] days around carbon shocks
Abnormal Returns: Risk-adjusted performance relative to market
Statistical Testing: Cross-sectional t-tests for significance

4. Volatility Modeling Comparison

  • Traditional: GARCH(1,1) with maximum likelihood estimation
  • Modern: LSTM neural networks with dropout and regularization
  • Performance: RMSE, MAE, and directional accuracy metrics

5. Regression Analysis

Model: ETF_return_t = α + β₁×carbon_shock_t + β₂×VIX_t + β₃×Brent_t + β₄×DGS10_t + ε_t
Estimation: OLS with heteroscedasticity-robust standard errors (HC1)

📈 Key Findings & Statistical Evidence

🎯 Core Results

Metric ICLN (Green) XLE (Brown) Interpretation
Mean CATE 4.19% 13.49% Brown ETFs 3x more sensitive
Regression Coefficient 0.0849*** 0.1757*** Highly significant effects
R-squared 3.8% 7.2% Better model fit for brown ETFs
Volatility RMSE 0.226 0.164 Lower forecasting errors

Note: *** indicates p < 0.001

📊 Economic Significance

  • $1 carbon price shock17.6 basis points increase in XLE returns
  • Market asymmetry: Brown assets more exposed to carbon transition risks
  • Policy transmission: EU regulations create measurable US market impacts
  • Investment implications: Clear risk differentiation for ESG portfolios

🔍 Robustness Across Time Periods

Main Period (2018-2024):

  • Post-financial crisis stability
  • COVID-19 period included for market stress testing
  • Recent policy impacts captured

Extended Period (2010-2024):

  • European debt crisis provides additional market stress
  • Longer-term trends validate findings
  • Consistent coefficient signs across periods

🚀 Quick Start Guide

🔧 Prerequisites

# Required Python packages
pandas >= 1.3.0          # Data manipulation
numpy >= 1.21.0          # Numerical computing
scikit-learn >= 1.0.0    # Machine learning (Causal Forest)
tensorflow >= 2.6.0      # LSTM models
statsmodels >= 0.13.0    # Econometric analysis
matplotlib >= 3.4.0      # Visualization
seaborn >= 0.11.0        # Statistical plotting

📚 Execution Sequence

  1. Data Preparation: 03_Code/Main_Analysis/data_agg.ipynb
  2. Main Analysis: 03_Code/Main_Analysis/Main_Code_Project.ipynb
  3. Robustness Checks: 03_Code/Main_Analysis/Main_Code_Project_14Yr.ipynb
  4. Visualization: 03_Code/Visualization/presentation_plots.ipynb

🎯 Key Results Location

  • Summary Statistics: 04_Results/Statistical_Analysis/Summary_Results/
  • Regression Outputs: 04_Results/Statistical_Analysis/Regression_Results/
  • Visualizations: 04_Results/Figures/Main_Figures/

💡 Research Contribution & Impact

🎓 Academic Contributions

  • First comprehensive study comparing green/brown ETF responses to carbon shocks
  • Novel application of causal forest methods to climate finance
  • Machine learning integration with traditional econometric approaches
  • Cross-Atlantic policy transmission evidence (EU → US markets)

🏢 Practical Applications

  • Portfolio Risk Management: Quantified climate transition risks
  • ESG Investment Strategy: Data-driven green/brown asset allocation
  • Policy Impact Assessment: Market response measurement tools
  • Volatility Forecasting: Superior LSTM-based prediction models

🌍 Policy Implications

  • EU carbon policy creates measurable global market effects
  • Market-based mechanisms effectively transmit climate signals
  • Asset differentiation supports sustainable finance transition
  • Risk disclosure needs differ across green/brown investments

📊 Methodological Innovations

🤖 Machine Learning Integration

# Causal Forest Implementation
from sklearn.ensemble import RandomForestRegressor
from causalml.inference.tree import CausalTreeRegressor

# LSTM Volatility Forecasting
model = Sequential([
    LSTM(50, return_sequences=True, dropout=0.2),
    LSTM(50, dropout=0.2),
    Dense(1, activation='linear')
])

📈 Advanced Econometrics

  • Heteroscedasticity-robust standard errors (HC1)
  • Cross-sectional dependence testing
  • Structural break analysis across time periods
  • Placebo testing for causal identification

🔍 Robustness Framework

  • Multiple time horizons (6-year vs 14-year samples)
  • Alternative shock thresholds (1.0σ, 1.5σ, 2.0σ)
  • Different event windows [-5,+5] to [-15,+15] days
  • Subsample stability testing

📚 Documentation & Reproducibility

📖 Complete Documentation

🔄 Reproducibility Standards

  • Version-controlled analysis pipeline
  • Relative file paths for cross-platform compatibility
  • Documented dependencies and environment specifications
  • Complete data provenance tracking

Quality Assurance

  • Statistical validation of all model assumptions
  • Cross-validation for machine learning components
  • Sensitivity analysis for key parameters
  • Independent verification of results

🎯 Future Research Directions

🔬 Methodological Extensions

  • Deep learning approaches to volatility clustering
  • Network analysis of cross-asset contagion effects
  • High-frequency intraday shock transmission
  • International markets comparison (EU, Asia, US)

🌍 Policy Applications

  • Central bank climate stress testing
  • Regulatory impact assessment frameworks
  • Transition pathway modeling for different sectors
  • Carbon border adjustment market effects

💼 Industry Applications

  • Real-time risk monitoring systems
  • Dynamic hedging strategies for climate risks
  • ESG factor integration in portfolio optimization
  • Climate scenario analysis for institutional investors

👥 Contact

📧 Contact

📧 Email: aditya.rohatgi11@gmail.com

💼 LinkedIn: https://linkedin.com/in/adityarohatgi

🔗 GitHub: https://github.com/adityarohatgi11

For questions regarding methodology, data access, or collaboration opportunities, please feel free to reach out.

About

Master's thesis I completed at Cal Poly analyzing the impact of carbon price shocks on different types of investment funds. Used machine learning and econometric methods to show how traditional energy ETFs react much stronger than clean energy ones. This project combines my interests in climate policy, finance, and AI/ML.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors