ㄱ Creating Synthetic ESG Scores for Financial Modeling

Environmental, Social, and Governance (ESG) metrics have become integral to investment decision-making, yet the lack of standardized scoring systems can make consistent analysis challenging. Synthetic ESG scores—constructed by aggregating data from unstructured sources—offer a solution. By leveraging techniques like natural language processing (NLP), machine learning, and data aggregation, analysts can create custom ESG scores tailored to their financial models and investment strategies.

This article outlines how to build synthetic ESG scores by combining unstructured data sources, providing a step-by-step guide for data collection, processing, and modeling.

---

Table of Contents

1. What Are ESG Scores?
2. Why Create Synthetic ESG Scores?
3. Key Components of a Synthetic ESG Score
4. Steps to Build Synthetic ESG Scores

• Data Collection
• Data Preprocessing
• Score Aggregation

5. Case Study: Building ESG Scores for Tech Companies
6. Challenges and Limitations
7. Future Directions in ESG Scoring
8. Conclusion

---

1. What Are ESG Scores?

ESG scores evaluate a company's performance across three dimensions:

1. Environmental (E): Impact on natural resources and climate.
2. Social (S): Employee well-being, diversity, and community relations.
3. Governance (G): Leadership quality, ethics, and shareholder rights.

These scores help investors assess a company’s sustainability and risk profile.

---

2. Why Create Synthetic ESG Scores?

Limitations of Existing ESG Scores

1. Lack of Standardization: Scores vary significantly across providers (e.g., MSCI, Sustainalytics).
2. Opaqueness: Proprietary methodologies make it difficult to understand score drivers.
3. Coverage Gaps: Small-cap and emerging market companies often lack ESG ratings.

Advantages of Synthetic Scores

• Tailored to specific investment strategies or industries.
• Transparent and adaptable to new data sources.
• Incorporates real-time insights from unstructured data like news, social media, and filings.

---

3. Key Components of a Synthetic ESG Score

Data Sources

• Unstructured Data:
• News articles (e.g., climate impact reports).
• Social media (e.g., employee reviews).
• Earnings call transcripts.
• Regulatory filings (e.g., 10-K reports).
• Structured Data:
• Carbon emission metrics.
• Board diversity statistics.
• Corporate social responsibility (CSR) disclosures.

Metrics and Weighting

Define metrics for each ESG pillar:

1. Environmental: Carbon footprint, renewable energy usage.
2. Social: Employee satisfaction, diversity metrics.
3. Governance: Board independence, audit transparency.

Assign weights based on sector relevance (e.g., higher environmental weighting for energy companies).

---

4. Steps to Build Synthetic ESG Scores

Step 1: Data Collection

Sources

• News APIs: Google News, Bing News, or LexisNexis.
• Social Media APIs: Twitter, Reddit, Glassdoor.
• Regulatory Data: SEC EDGAR for filings.
• Alternative Data Providers: RepRisk, ESG Book.

Automation

Use web scraping and APIs for continuous data ingestion.

Example using Python for news scraping:

from GoogleNews import GoogleNews

googlenews = GoogleNews()
googlenews.search("Company ESG performance")
articles = googlenews.result()

# Extract titles and links
for article in articles:
print(article["title"], article["link"])

---

Step 2: Data Preprocessing

Natural Language Processing (NLP)

1. Text Cleaning: Remove stop words, punctuation, and irrelevant text.
2. Sentiment Analysis: Use tools like VADER or TextBlob to quantify sentiment.
3. Topic Modeling: Apply Latent Dirichlet Allocation (LDA) to categorize content into ESG-related topics.

Example sentiment analysis using VADER:

from vaderSentiment.vaderSentiment import SentimentIntensityAnalyzer

analyzer = SentimentIntensityAnalyzer()
text = "This company is a leader in renewable energy innovation."
score = analyzer.polarity_scores(text)
print(score)

Structured Data Normalization

• Standardize metrics to a common scale (e.g., 0–1).
• Handle missing values with imputation or interpolation.

---

Step 3: Score Aggregation

Methodology

1. Scoring Model:

• Assign sentiment scores for unstructured data sources.
• Aggregate metrics using weighted averages.

2. Weighting System:

• Adjust weights based on industry relevance.
• Use machine learning techniques like ridge regression or principal component analysis (PCA) for optimal weighting.

Final Score Calculation

[
\text{ESG Score} = w_E \cdot S_E + w_S \cdot S_S + w_G \cdot S_G
]
where:

• ( w ): Weight for each pillar.
• ( S ): Score for each pillar.

---

5. Case Study: Building ESG Scores for Tech Companies

Objective

Construct synthetic ESG scores for five major tech firms using a combination of sentiment analysis and structured metrics.

Implementation

1. Data Sources:

• News articles from 2018–2023.
• Glassdoor reviews for social metrics.
• Board composition and diversity data from annual reports.

2. Weighting:

• Environmental (20%): Energy usage, carbon offset initiatives.
• Social (40%): Employee satisfaction, diversity policies.
• Governance (40%): CEO duality, board independence.

3. Results:

• Positive sentiment spikes aligned with announcements of renewable energy projects.
• Glassdoor reviews highlighted diversity gaps as a social risk.
• Governance metrics revealed strong correlation with stock performance.

---

6. Challenges and Limitations

1. Data Quality: Unstructured data can be noisy or biased.
2. Subjectivity in Weighting: Weight assignments may not fully capture market realities.
3. Coverage Gaps: Limited data for smaller companies or emerging markets.
4. Real-Time Scalability: Continuous updating requires robust infrastructure.

---

7. Future Directions in ESG Scoring

1. AI-Powered Contextual Analysis: Use transformer models like BERT to analyze ESG context more accurately.
2. Cross-Asset Applications: Extend synthetic scores to bonds and ETFs for ESG-aligned portfolio construction.
3. Blockchain Integration: Track ESG metrics transparently using blockchain for provenance and verification.
4. Real-Time Monitoring: Deploy streaming NLP models for instant updates on ESG events.

---

8. Conclusion

Synthetic ESG scores provide a flexible and transparent alternative to traditional ESG ratings, enabling analysts to integrate diverse data sources into financial modeling. By leveraging NLP, machine learning, and customizable weighting schemes, these scores can offer unique insights tailored to specific investment strategies. As ESG data ecosystems continue to evolve, synthetic approaches will play an increasingly vital role in sustainable finance.

---

Would you like to see Python code for aggregating ESG metrics or a guide to integrating synthetic scores into a portfolio model?