Machine learning measurement of carbon emission financial risks

Purpose and goal:

The aim has been to study how machine learning methods can be used to determine the cost of carbon dioxide emissions and the risk that the cost will increase. By improving the optimization models that measure the cost term structure of future emissions, the estimation of stochastic processes and determine optimal decisions, significant improvements have been obtained. It leads to a better understanding of the systematic risks, risk measurement and risk management for emission rights.

Expected results and effects:

The adaptation of the model to measure the term structure was expected to lead to more accurate measurements, better modeling of the risks in futures markets for emissions rights and improved risk management. When validating with historical data, significant improvements can be observed. Via performance attribution, the improvement can be traced to a more cost-effective hedge being identified where the risk exposure is limited.

Approach and implementation:

In collaboration with Handelsbanken, SEB and Swedbank, the models have been validated and developed to become more realistic. Optimization provides both the tool to identify improvements, but also to validate that the improvements are obtained when they are applied in practice. Through a systematic operations research approach, the uncertainty that exists in real problems can be managed, in order to identify optimization models that really work better in practice. Through the project, we have established a collaboration where we can improve the modelling.