Introduction

Portfolio carbon footprints—total CO2 emissions attributable to holdings—are increasingly important for institutional investors facing climate commitments. However, direct emissions data is sparse. Machine learning models estimate missing emissions data using available company information, enabling accurate portfolio carbon footprint calculation and reporting.

Emissions Data Gaps

Many companies, particularly small-caps, do not report Scope 1, 2, or 3 emissions. Scope 3 (supply chain) emissions are estimated inconsistently. AI models trained on companies with reported emissions can predict emissions for companies without, using features: sector, revenue, asset base, energy intensity proxies.

ML Model Development

Train gradient-boosting models to predict Scope 1, 2, 3 emissions separately. Features: sector classification, company size, energy consumption indicators, supply chain complexity. Generate probability distributions of emissions, not point estimates, capturing uncertainty. Aggregate to portfolio level with confidence intervals.

Conclusion

AI-enhanced emissions estimation enables accurate portfolio carbon footprint calculation, supporting climate-aligned investing.