ESG and Sustainability Signals

Two rating agencies score the same company's environmental, social, and governance practices and disagree by more than they'd disagree scoring its credit risk. One weighs carbon intensity heavily and gives the company a poor score; another weighs labor practices and board diversity, where the company does well, and rates it strongly. Both scores are published as a single letter grade, and both get used by asset owners to screen the same stock in or out. That disagreement is the starting point for how quant desks actually treat ESG data: not as a finished signal, but as noisy raw material that needs the same scrutiny as any other dataset before it goes into a portfolio.

What actually goes into the score

A typical ESG rating combines things as different as a company's board independence, its water usage per unit of revenue, its safety incident rate, and whether it discloses its supply chain — normalized, weighted, and averaged into one grade. Two consequences follow. First, because providers choose different weights and different raw inputs, correlation between major ESG rating agencies' scores on the same company is often only moderate, far below what you'd expect for two measures of the same thing. Second, because these ratings update slowly — often annually, from self-reported disclosures — an ESG score reacts to news changes with a lag, unlike a price-based signal that updates every tick.

This matters for return prediction in two separate, sometimes opposite, ways. A low-carbon tilt can behave like a crowded factor bet: as more capital chased sustainability mandates through the 2010s, stocks with better ESG scores got a persistent valuation bid that had little to do with their fundamentals, meaning the "signal" was partly picking up flow, not information. Separately, the component scores — particularly governance measures like board independence or executive pay structure — have shown more durable links to future returns and blow-up risk than the composite ESG grade, because governance failures are closer to a leading indicator of fraud or mismanagement than a carbon score is.

Worked example

A researcher regresses next-quarter stock returns against four variables: the composite ESG score, and its E, S, and G sub-scores separately, controlling for size, value, and momentum. The composite score shows almost no significant relationship to returns once those standard factors are controlled for — most of its apparent predictive power was really a value or size tilt in disguise. But the governance sub-score alone retains a small, statistically significant relationship: firms in the bottom decile of governance scores underperform by a noticeable margin over the next year, consistent with governance problems preceding earnings restatements and other value-destroying surprises. The researcher builds the eventual signal from the governance sub-score plus a short-interest overlay, and discards the environmental and social components as too noisy and too slow-moving to add value on their own.

What this means in practice

ESG data is treated by quant desks the way any alternative dataset is treated: check what it actually measures, check how correlated it is across providers, and test its components separately before trusting the headline number.