Corporate carbon disclosure is central to asset pricing and sustainable finance, yet firm-level emissions data remain incomplete and selectively reported. We model disclosure as a selection process in which reporting depends on latent emissions and implement a missing-not-at-random (MNAR) correction that recovers the full emissions distribution rather than relying on conventional missing at-random imputations. Using U.S. public firms data from 2012–2023, we show that higher-emitting firms are systematically less likely to disclose. Correcting for this selection increases average Scope 1 emissions by 15–19%, with substantial mass added to the upper tail, implying that standard approaches materially understate aggregate emissions and associated carbon damages. Importantly, imputing missing emissions with MNAR-adjusted values leaves return–emissions
relationships largely unchanged, indicating that the correction does not induce spurious pricing effects. Our results highlight selective disclosure as a first-order source of bias in corporate carbon data with direct implications for empirical asset pricing and climate policy.
