Skip to content

Content Library

Back to Articles

Article April 4, 2025

Banks Are Gambling on Climate Risk Models with Flawed Data

Banks Are Gambling on Climate Risk Models with Flawed Data

Ying Poikonen, Head of Modelling Group, SMBC

Jing Zhang, Vice President, Sumitomo Mitsui Banking Corporation

At CeFPro’s recent Sustainable Finance conference, SMBC's modeling team warned that banks face severe challenges in sourcing usable data for climate risk modeling. AI may offer game-changing tools, but without strong governance, it introduces its own risk
  • Banks struggle to model climate risk due to poor quality and inconsistent data. 
  • SMBC experts explained how AI could automate and validate ESG data. They also emphasized governance and auditability.
  • CSRD compliance demands robust, real-time, and verified data strategies, not just vendor solutions.
  • Internal teams and external consultants are key to building trustworthy AI frameworks and climate models

For years, banks have talked about integrating climate risk into financial modeling, but it is only now, in the face of pressing regulatory deadlines under rules such as those included within the Corporate Sustainability Reporting Directive (CSRD) that the conversation is reaching a critical point.  

However according to risk leaders within the EMEA division of the Sumitomo Mitsui Banking Corporation (SMBC), most banks still want for the one thing that, they say, is essential to success: good data. 

"As financial regulations like CSRD demand greater transparency, one of the biggest hurdles we face is not only to understand climate risk, but to find good data to model it effectively," said Ying Poikonen, SMBC’s Head of Modelling, during CeFPro's recent Sustainable Finance conference. 

Her colleague, Jing Zhang, who leads climate risk modeling, also used the session on the challenges of finding quality data to outline how the bank’s journey began in 2019 with Prudential Regulation Authority (PRA) guidance.  

She explained that early scorecard models leaned heavily on expert judgment due to data scarcity, but went on to describe how, over time, a more structured approach emerged—identifying risks, mapping exposures, and quantifying financial losses.  

Each step brought new data requirements: from carbon pricing and decarbonization pathways to hazard-specific damage ratios and climate-adjusted macro forecasts. 

But the reality is that in real-world modeling, finding that data remains a massive challenge. Public datasets like NGFS or IPCC are free and credible but require extensive transformation.  

"It can be very time consuming and technically demanding," Zhang warned. Commercial vendors offer ready-to-use metrics but bring cost, quality, and consistency concerns – especially for smaller institutions. 

The session also highlighted additional issues around the application and interpretation of data, with many organizations in the financial services industry struggling to convert climate indicators into financial impact.  

Translating rising temperatures or policy shifts into monetary losses requires complex functions and assumptions, and as Zhang noted, banks are often forced to use damage ratios as ‘haircuts’ to asset values – a workaround that, while necessary, remains imprecise. 

This, Poikonen argues, is where artificial intelligence enters the equation as a solution that, she believes,  could dramatically improve ESG data quality.  

"AI can scan and extract a huge amount of data," she said, citing sources ranging from IoT devices to social media. More importantly, she argues, AI can validate ESG data, detect anomalies, benchmark disclosures, and power predictive analytics.  

Tools like RepRisk are already being used by SMBC to scan negative news and inform internal ESG ratings that influence credit decisions, but Poikonen remains quick to stress caution.  

"AI is wonderful and beautiful. It helps everyone,” she told delegates, “but it is not without risk."  

She cited SMBC’s recent rollout of Microsoft Copilot, which prompted major reviews of data governance, legal constraints, and model risk. "Before you embark on the AI journey, make sure that you do have a strong AI governance framework," she warned. That includes validating models before deployment and conducting self-assessments before facing internal audit. 

Events you may be interested in

View all of our upcoming events

Sign Up to Our Newsletter

Sign up

Connect Plus

Unlock more with Connect Plus membership

This content is available to Connect Plus members.