Carbon Uncertainty and Asset Prices

We find that measurement uncertainty in firms’ carbon intensity commands a positive risk premium, distinct from and alongside a robust premium associated with carbon-intensity levels. Because most emissions are estimated rather than disclosed, our main contribu­tion is to construct an observation-level measure of carbon-data precision at scale: for each firm-month, we quantify the uncertainty of the underlying carbon signal. We conceptually separate estimation risk (statistical error from inferring emissions under sparse firm infor­mation) from model risk, proxied by cross-vendor disagreement. Using a replicable Multiple Imputation by Chained Equations (MICE) framework combined with Random Forests, we jointly estimate firm-month emissions, carbon intensity, and uncertainty for the full CRSP universe, overcoming the selection bias inherent in vendor-restricted samples. Training on self-disclosing firms, we show that vendor-estimated data are largely redundant: adding ven­dor observations provides little incremental information beyond disclosures, consistent with estimates being anchored to the same disclosure backbone. By contrast, imputations within vendor-covered universes can embed selection bias and materially alter asset-pricing infer­ences. Overall, investors price not only carbon exposure but also carbon opacity.