Green Silence: Double Machine Learning Carbon Emissions Under Sample Selection Bias - new publication by Olivier Scaillet
A new paper by Olivier Scaillet, GFRI's Director and SFI Senior Chair, raises alarm bells about carbon emissions being underestimated.
The paper "Green Silence: Double Machine Learning Carbon Emissions Under Sample Selection Bias" is co-authored with C. Chen and A. Lioui.
It examines green silence: when high-emitting firms strategically withhold data, anticipating third-party estimates will understate their true impact. In such cases, voluntary disclosure regimes become arenas of "green silence," where self-censorship creates bias and reported carbon footprints become non-random samples, systematically skewed toward firms with lower incentives to hide.
The authors connect this economic problem with machine learning, proposing a Heckman-inspired three-step framework for high-dimensional settings that corrects for strategic non-disclosure and achieves consistent variable selection under sample selection bias.
This approach enables quantification of both the statistical and economic significance of the bias, as well as an empirical inference on the extent of green silence. The findings show that voluntary disclosure induces adverse selection: green silence rewards polluters and undermines decarbonization. For publicly listed U.S. firms, underestimation translates into an annual tax revenue shortfall of USD 190 million and a hidden social cost of carbon of USD 38 billion. By applying this correction methodology, researchers and policymakers can mitigate these biases. This, in turn, enables more accurate measures of firm pollution-control adoption, associated costs, and the true valuation of environmental externalities.
The paper was featured in allnews (in French).
Sep 30, 2025