Revolutionizing ESG Performance & Reporting with Extreme Forecasting and Synthetic Data

Financial institutions can now estimate mortgage portfolio emissions more accurately, overcome challenge of missing data in risk management processes, and take one step closer to becoming ESG ready.

The Business Context

Extreme weather events, record breaking temperatures and UN climate change summits, COP26 and COP27, have put climate change at the top of the global agenda. Amongst the international efforts underway to combat global warming and achieve net-zero, financial institutions are considered to have a particularly important role to play. In addition to decarbonising their own operations, they can wield their financing influence to promote sustainable projects and encourage businesses and consumers to be more environmentally friendly.

Financial institutions are therefore being pushed to make changes to how they conduct their business, to ensure that environmental, social and governance (ESG) considerations are brought to the fore.

Key changes include:

• new climate change disclosures reporting the greenhouse emissions of loan and investment portfolios,
• incorporating ESG factors into risk management processes, and
• developing green financial products to encourage the transition to net-zero.

However, financial institutions face a number of challenges in making these changes. One of the main hurdles cited by financial institutions is that of missing or limited data. The PRA in its 2021 Climate Change Adaptation Report also highlighted the issues arising from “a lack of granular data or limitations in modelling techniques to reflect climate variables”. A good example of this issue is found in the UK mortgage market, where lenders rely on Energy Performance Certificate (EPC) Ratings data for risk decisions and reporting, but there are issues with missing or outdated information.

 EPC data Case Study

Graeme McRitchie, Head of Prudential and Enterprise Risk at Leeds Building Society, said:

“Estimating missing energy performance data presented a number of technical challenges that Leeds Building Society wanted to look externally for support with. In addition to achieving the stated project goal of estimating missing EPC data with full auditability, 4-Xtra went above and beyond the original project scope. It put in place the processes to automate the data cleaning and reformatting of public data to ensure continuous provision of the latest data and also built a fully-customised solution to deliver the model outputs. This also incorporated advanced data analytics, which now provides us with the capability to manipulate large amounts of data.”

Introducing the 4-Xtra Platform

Leeds Building Society is the fifth largest building society in the UK and a top 15 mortgage lender. As part of its ESG efforts it had become a member of the Partnership for Carbon Accounting Financials (PCAF), which provides a framework for financial institutions to assess and disclose greenhouse gas emissions of loans and investments.

Leeds Building Society was faced with the challenge of missing energy performance data for properties in its mortgage portfolio. Where possible, the Society derived this data from EPCs which measure the energy efficiency of buildings. Valid for 10 years on issuance, they are required for all properties when put up for sale or rent. However, this means that many properties in the UK do not have a valid EPC. Furthermore, there are also noted issues with time-lags in the EPC database, with register data only updated every four to six months.

Leeds Building Society required the energy performance data to assist with the risk profiling of its mortgage loans (where this data was not already available) and to estimate carbon emissions of its mortgage portfolio.

The Proof of Concept

To address these challenges, Leeds Building Society was open to working with a third-party that could help it quickly model EPC data gaps using a range of alternative techniques. After researching the market, it engaged 4-Xtra Technologies to undertake a three-month Proof of Concept (PoC) to demonstrate that it could effectively fill in the missing data to a level that would satisfy both Leeds Building Society’s requirements and the expectations of the regulator.

A Leeds University deep-tech spin-off, 4-Xtra Technologies has built on academic work in the theory and predictability of extreme events and used machine learning and artificial intelligence methods to develop an extreme event forecasting risk modelling solution and synthetic data generator for a variety of use cases in the financial services sector. These capabilities were applied to solve the challenge of missing EPC ratings and to investigate the feasibility of generating secondary variables, such as potential EPC rating and CO2 emissions.

4-Xtra worked with Leeds Building Society to prepare two data sets for modelling purposes, namely the Society’s own internal database and the public EPC data. Significant data cleaning and formatting was required for the public data, with 4-Xtra undertaking tasks such as label normalisation, systematic error correction and language translation. Two versions of the model to predict EPC ratings were then configured on each of the data sets
separately, as were additional model configurations to generate the additional secondary variables, potential EPC rating and CO2 emissions. These configurations were designed to automate monitoring and processing of data to ensure that Leeds Building Society has access to the latest information and are delivered with a full model and data-audit trail providing complete transparency.

In addition to the key model deliverables, 4-Xtra also went beyond the core scope of the project to provide a custom modelling overlay and a fully-customised and scalable Software as a Service (SaaS)-based production ready application accessible via API or a user-friendly web application interface. The solution also provides data analytics capabilities to enable manipulation of large data sets.

The Outcome

The PoC was successful and Leeds Building Society subsequently signed to become a full client of 4-Xtra. The EPC rating data generated by the models developed by 4-Xtra provided appropriate estimates and achieved the goals of the project. The reports generated by the models had, amongst others, high AUROC (Area Under the Receiver Operating Characteristic Curve) Scores, which are used to describe the effectiveness of a model, with 1 being a perfect score.

This means that Leeds Building Society can readily incorporate EPC ratings into its risk management processes (for any missing portfolio data) and more accurately estimate the emissions of its mortgage portfolio.

The engine has been designed with flexibility in mind and can be adjusted for future requirements or any data changes. For example, 4-Xtra has designed a neighbourhood extraction feature that enhances private datasets with the information available from public, government-curated ones. When any public data sets are updated, the power and utility of the neighbourhood features also increases, thus enhancing internal datasets. Future updates that are also being investigated include further utilisation of the EPC recommendations, as well as using flood data to assess flood risk.

Andy Mellor, Chief Risk Officer at Leeds Building Society, said:

“Leeds Building Society wants to demonstrate that it is effectively incorporating climate-risk considerations into its risk management processes and credit decisions. Missing energy performance ratings was a particular obstacle in achieving these requirements for some aspects of our mortgage portfolio. The successful PoC undertaken with 4-Xtra Technologies through configuration of its SaaS solution has demonstrated that we can now instantly estimate any gaps, which in turn improves our credit decisioning processes and enables us to report our mortgage portfolio emissions more accurately than previously.”