MARLEY AI model: Carbon price cuts raise CO2, erode green investment
AI simulation of Italy's power market shows price relief backfires on emissions.
A team of researchers led by Javier Gonzalez-Ruiz and Massimo Tavoni published a new arXiv paper (2608.12363) using MARLEY, a multi-agent reinforcement learning framework, to model Italy's power sector under the controversial 2026 Decreto Bollette policy. The proposal would strip the carbon price equivalent from gas plant bids in wholesale electricity markets, aiming to lower energy costs amid geopolitical tensions. The study tests this across configurations with varying levels of green investment support, resource adequacy, and flexibility.
The results are stark: partial carbon price suppression delivers only short-term cost reductions, while total system costs barely change long-term—consumers ultimately repay the deferred emissions. CO2 emissions rise in most configurations because suppressing the price signal reduces incentives for renewable and storage investment. Only the most aggressive green-investment support prevents this, but that approach marginalizes wholesale prices, creating a hybrid market paradigm that contradicts the policy's rationale. The paper includes extensive sensitivity analysis and supplementary material.
- MARLEY simulation shows partial carbon price suppression yields minor long-term cost savings but higher CO2 emissions
- Consumers repay deferred emissions costs, while renewable and storage investment incentives erode
- Only extreme green-investment configurations avoid emissions rises, but they undermine wholesale market signals
Why It Matters
Informs EU policymakers that carbon price suppression delays decarbonization without real cost benefits.