Beyond Binary: A Probabilistic Bayesian Framework for Assessing Carbon Project Additionality

The carbon credit system is designed to allocate credits to projects that genuinely require financial support. Projects such as wind farms, which would proceed even without incentives, should not be able to claim emissions reductions as additional; this principle is known as additionality. Current methodologies rely on a binary test to determine additionality: a project's rate of return either meets a benchmark or it does not. However, the distinction between projects that need credits and those that don't is often significant, and a binary classification fails to capture this complexity. As a result, projects with returns of 7.9% and 8.1% receive opposite classifications despite the negligible difference between them.

This research proposes a probabilistic alternative. Instead of binary categorization, the framework assigns each project an additionality probability based on its financial status. Projects are grouped into four zones, each with an initial probability reflecting the likelihood that projects in similar financial situations require carbon credits. A parameter, Φ, accounts for companies that invest in unprofitable green projects for reasons unrelated to credits, such as brand image, legal obligations, or strategic considerations.

Two illustrative examples and a test portfolio of 1,000 synthetic projects yield scores ranging from 5% (a Christmas tree farm) to 99% (a solar mini-grid in rural Sub-Saharan Africa). When key assumptions vary by 20%, final scores shift by approximately 6 percentage points, indicating that the framework is reasonably stable. To the best of our knowledge, no previous study has developed a financially tiered Bayesian additionality framework with explicit implementation likelihoods and dynamic updating. We hypothesize that this approach (1) generates stable additionality scores, (2) effectively distinguishes projects that genuinely need carbon credits from those that don't, and (3) enables more transparent, auditable standard-setting than current binary methods.