SFI Professor Scaillet was always puzzled by the fact that papers published in respected finance journals used the carbon estimates of ESG data vendors (such as MSCI, Refinitiv, and Trucost). What stood out to him was the aspect of self-selection due to the voluntariness of carbon disclosure. As an econometrician, he knows that this non-random self-selection must be considered when creating carbon emissions estimates or else they will be biased.

"For the Swiss financial industry to keep its competitive advantage, it is necessary to continue to invest in knowledge—as that is the service provided by the industry. As a financial econometrician, I always try to come up with new econometrics techniques in order to answer a very specific point in finance. Because if you do not use the right statistical methodology, you run the risk of thinking that you know something when you do not."

The Impact of Adverse Selection and Overly Simplistic Extrapolations
SFI Professor Scaillet and coauthors observed that ESG data vendors were only using a handful of characteristics—such as size, revenue, and sector—to extrapolate carbon estimates of non-reporting firms. As an alternative, they created a much richer methodological framework—looking at 173 characteristics—that corrects for strategic nondisclosure. They were then able to quantify both the statistical and the economic significance of the adverse selection bias. Looking at one vendor's data for 3'444 US firms between 2010 and 2023, they found a significant underestimation to the tune of USD 2.6 billion in carbon tax revenue shortfall and implied economic costs of up to USD 525 billion.

Their analysis reveals that when reported carbon footprints are voluntary, reporting becomes skewed toward firms with lower incentives to hide. As big-polluting firms are aware that their carbon footprint estimates will be based on a simple comparison to firms that do report—using only a few factors, like size, revenue, and sector—they choose silence. This is the paradox of green silence, where polluters are rewarded.

It is worth noting that not only did the research find significant underestimation, but there is also very low similarity between the researchers' estimates and those of data vendors. That is to say, it is not just a matter of increasing all the vendors' scores in order to get an accurate estimate of carbon emissions; accurate estimates depend on using this methodological framework for rich data environments. The authors are now building on this methodology to look at estimating the implied carbon premium.

More Information

• Green Silence: Double Machine Learning Carbon Emissions under Sample Selection Bias, Swiss Finance Institute Research Paper No. 25-66 (with C. Chen and A. Lioui) 

SFI Prof. Olivier Scaillet (UNIGE)

Olivier Scaillet is Professor of Finance and Statistics at the University of Geneva, Head of the Geneva Finance Research Institute, and holds an SFI Senior Chair. He is a regular speaker at leading finance and financial econometrics conferences. His papers have been published in the top academic journals in these fields and he is an elected fellow of several leading academic societies in econometrics and statistics. 

"SFI helps to advance the finance knowledge needed to keep the financial industry competitive, through lectures, knowledge exchange activities, and SFI Master Classes. The topics of SFI Master Classes are highly relevant and precisely aligned with the current needs of the financial industry."