Technical Appendix to “The Green Asset Gold Rush Methodology”

By Li Ziyi and Kang Qiming

Youying Low-Carbon Environmental Protection Organization

1. Objective and Research Question

This project explores the empirical relationship between a company’s environmental performance—particularly carbon intensity and ESG (Environmental, Social, Governance) scores—and its financial returns. We aimed to answer the following questions:

•Do companies with better ESG ratings consistently outperform the market in terms of shareholder return?

•Can machine learning models identify nonlinear relationships between green indicators and investment returns?

•How can green data modeling support sustainable financing for grassroots environmental projects?

2. Data Sources and Processing

We constructed a multi-dimensional dataset from the following sources:

Processing Steps:

•Aligned time-series data to the fiscal year.

•Winsorized outliers at the 1st and 99th percentile.

•Normalized ESG scores to a 0–100 scale.

•Removed incomplete entries and companies with <3 years of return data.

3. Empirical Models

We employed two complementary methods:

(a) Multiple Linear Regression Model

Formula:

Returni=β0+β1⋅ESGi+β2⋅CarbonIntensityi+β3⋅MarketCapi+εi

•Dependent Variable: 5-year average shareholder return.

•Independent Variables: ESG Score, Carbon Intensity, Control variables (Industry, Size).

•Result:

◦ESG score: positive and statistically significant at 99% confidence.

◦Carbon intensity: negative coefficient, also significant.

(b) Random Forest Regression

•Features: ESG sub-scores, carbon metrics, financial ratios.

•Output: Shareholder Return.

•Cross-validated with 5-fold CV.

•Feature Importance Ranking:

1.Carbon Intensity

2.Environmental Subscore

3.Return on Equity (ROE)

4.Scope 3 Emissions

5.Governance Score

4. Key Findings

•The S&P 500 ESG index outperformed the traditional index by 15.1% (2019–2024).

•Top 500 European carbon-reducing firms yielded 15.5% average 5Y return vs. 14.8% for FTSE All-Share.

•ESG and low-carbon performance are predictive of long-term stability and return.

•Our model showed ~9% gain in predictive accuracy over baseline when incorporating environmental variables.

5. Limitations and Future Work

•ESG scoring methodologies vary across providers—standardization is a challenge.

•Scope 3 emissions data remains incomplete for many firms.

•Future work will integrate NLP models to analyze sustainability reports and improve real-time tracking of environmental disclosures.

6. Application and Impact

•Report submitted to green venture funds and local environmental foundations.

•Internal training module created to teach data-finance linkage to volunteers.

•Next phase: Build a dashboard integrating live ESG signals for retail investors.