A more realistic way to measure the financial risk of decarbonisation plans

Summary

• Traditional sensitivity analysis estimated a 43% chance that the industrial decarbonisation plan would perform worse financially than expected, while classical robustness analysis estimated 72%.

• The adaptive method produced a 57% estimate by fixing decisions that must be made early and allowing later choices to respond to new information.

• The approach was tested on a Dutch industrial cluster emitting about 9.5 million tonnes of CO2 a year, where investment decisions must align with a 55% emissions reduction by 2030 and net-zero emissions by 2050.

Why can long-term plans give a misleading picture of risk?

Infrastructure and decarbonisation plans depend on assumptions about future fuel prices, hydrogen costs, carbon taxes and technology. These values are uncertain, yet the way uncertainty is represented can strongly affect whether a plan appears financially credible.

Sensitivity analysis can make a plan look safer by assuming decision-makers know how the future develops and can respond perfectly. Classical robustness analysis can do the opposite by assuming that decisions made today cannot be adjusted later. Neither assumption matches how most long-term investment programmes are managed.

Why is this a prescriptive analytics problem?

Prescriptive analytics is concerned with what should be decided now and what can be decided later. In a multi-stage investment problem, timing matters: some infrastructure choices become difficult to reverse, while other decisions can change as prices, policies and technologies become clearer.

The research therefore moves beyond asking how sensitive one fixed plan is. It evaluates a decision policy: a sequence of choices that can be updated at defined points in time.

How did the team test a more realistic approach?

Researchers from the University of Amsterdam, Delft University of Technology and Quo Mare developed an adaptive robustness analysis that fixes early decisions but re-optimises later ones when new information becomes available.

They tested the method on a cluster of Dutch industrial sites in fertiliser, oil and chemicals that together emit about 9.5 million tonnes of CO2 a year. The team generated 1,000 future scenarios using historical data and expert input for factors including natural gas prices, hydrogen costs and carbon taxes.

They then compared the same decarbonisation plan under sensitivity analysis, classical robustness analysis and the new adaptive method.

What did the comparison show?

The three methods produced very different answers to the same financial-risk question. Sensitivity analysis estimated a 43% probability that the plan would perform worse than expected. Classical robustness analysis estimated 72%. The adaptive method placed the probability at 57%.

That range matters because the investments involved can translate to costs of billions of euros. A 43% risk estimate can make a plan appear relatively secure, while 72% can suggest a level of exposure that warrants a different strategy. The 57% estimate reflects a middle ground in which early commitments remain fixed and later decisions can adapt.

The analysis also indicated that early investment in shared infrastructure, including electricity grids and hydrogen pipelines, could reduce financial risk later. For workers, communities, companies and public authorities connected to these industrial regions, better risk estimates can shape whether transition plans are financed, delayed or redesigned.

What can the findings mean for other applications?

The method is not specific to industrial decarbonisation. It can be applied to other decisions that unfold in stages under uncertainty, including renewable-energy deployment, flood-protection investment and public-health infrastructure.

Because the research team made the approach computationally practical and published the code openly, future work can test it in other sectors and decision settings where planners need to commit capital today without pretending that tomorrow is already known.