Integrating ESG Data into Quant Models

The ESG data landscape for quant teams

Integrating ESG data into quantitative models is more complex than integrating financial data. ESG data has lower coverage, higher latency, inconsistent periodicity, multiple competing providers, and significant missing data rates — especially for smaller companies and emerging markets. Before ingesting ESG data, quant teams need to understand its structural characteristics.

Choosing ESG data providers

The choice of provider significantly affects model results. Given the low correlation between providers (~0.54), using a single provider introduces substantial provider-specific bias. Best practice for institutional quant strategies is to either aggregate multiple providers or understand clearly why one provider's methodology is superior for the specific use case.

• MSCI ESG Ratings: Industry-standard for institutional use; strong governance and social coverage; annual scores with controversy overlays
• Sustainalytics (Morningstar): Risk-score framework (Unmanaged ESG Risk); strong controversy monitoring; widely used for exclusion screening
• Refinitiv (LSEG): Large universe coverage; good for emerging markets; raw data available at disclosure level
• S&P Global CSA: Based on annual corporate sustainability assessment; sector-adjusted; strong for DJSI-tracking strategies
• OpenESG API: Real-time scores with AI controversy monitoring; framework-level breakdown; suited for ESG factor research
• CDP Scores: Best-in-class for climate and water data specifically; A–F scale; annually updated; used for climate transition risk models

Data ingestion patterns

Point-in-time data construction

The most critical technical requirement for backtesting is using point-in-time (PIT) data — the score as it existed at a specific historical date, not the score as restated or revised. ESG providers periodically revise historical scores. Using revised data in a backtest is a form of look-ahead bias.

import pandas as pd
import openesg

client = openesg.Client(api_key="sk-...")

# Fetch point-in-time ESG scores for a universe
# as_of parameter requests data as it existed at that date
scores = client.scores.history(
    tickers=["AAPL", "MSFT", "GOOGL", "TSLA"],
    fields=["composite_score", "e_score", "s_score", "g_score"],
    start="2020-01-01",
    end="2025-12-31",
    frequency="quarterly",
    as_of="point_in_time"  # avoid look-ahead bias
)

df = pd.DataFrame(scores)
print(df.head())

Handling the publication lag

ESG reports typically describe the prior year and are published 3–6 months after year-end. An ESG score updated in April 2026 based on 2025 data should only be used in models starting from the April 2026 publication date. Assuming the score was available on January 1 2026 introduces a 3–4 month look-ahead bias.

Normalisation across sectors

Raw ESG scores should not be compared directly across sectors because material ESG risks vary dramatically by industry. A technology company with a 70 environmental score and an oil company with a 70 environmental score are not environmentally equivalent — the oil company's score reflects much higher absolute emissions in a carbon-intensive sector.

Within-sector normalisation

The standard approach for cross-sectional ESG factor construction is sector-neutral normalisation: convert raw scores to z-scores within each GICS sector, then use the normalised score in portfolio construction.

import numpy as np

def sector_neutral_zscore(df, score_col="esg_score", sector_col="gics_sector"):
    """Convert ESG scores to within-sector z-scores."""
    df = df.copy()
    df["esg_z"] = df.groupby(sector_col)[score_col].transform(
        lambda x: (x - x.mean()) / x.std()
    )
    # Winsorise at ±3σ to limit outlier influence
    df["esg_z"] = df["esg_z"].clip(-3, 3)
    return df

# Apply to your holdings universe
scored = sector_neutral_zscore(holdings_df)

Missing data strategies

Missing data is the norm in ESG, not the exception. Your treatment of missing data should be explicit, documented, and consistent. The main approaches are:

Backtesting ESG factors

ESG factor backtests have historically shown mixed results, which is expected — ESG is not a single factor but a multi-dimensional risk overlay. The strongest documented ESG effects are in controversy monitoring (negative news predicts short-term negative returns) and governance quality (strong governance predicts long-term valuation multiples).

• Use long formation periods (quarterly rebalancing is standard for annual ESG data)
• Test on a value-weighted portfolio, not equal-weighted, to match realistic execution
• Decompose returns into E, S, and G components separately — the composite often masks different directional signals
• Test the turnover and transaction costs — ESG strategies tend to have low turnover but high rebalancing costs at threshold crossings
• Be explicit about whether the strategy is long-short (factor) or long-only with ESG tilt