Structured finance is essential for banks and other financial institutions to access funding. It connects capital markets with the real economy and enables the transition to sustainable economies.
Although this process involves intangible exchanges of money, bonds, and mortgages, it requires significant legal fees pre-closure, and ongoing reporting and compliance costs post-closure, along with extensive use of spreadsheets.
Experts in this space are increasingly acknowledging that existing analytics could be improved with climate and ESG-relevant data to stress test the pool of mortgages invested in their funds, for example.
This insight, together with the recent ESMA consultation paper that highlights the low level of data usage in securitisation, inspired us to organise the Structured Finance Hackathon, focusing on three tracks:
- Legal Tech: Streamlining pre-closure legal documentation drafting and modelling
- Climate Risk: Future-proofing buy-side analytics with ESG and climate data
- Reg Tech: Automating post-closure deal compliance reporting
The SF Hackathon
We chose Barcelona and the 5th of June to coincide with Global ABS, which our own hackathon judges were attending before the pitches.
Taking into account that the local ecosystem is still budding, we enabled online participation. More than 50 participants signed up, and four teams made it to the finish line to pitch their solutions to the judges.
We shared the hackathon challenges and provided tools & datasets in advance to let participants work on their code ahead of the event, then dedicated the on-site session to pitch training, networking and showcasing the prototypes.
We also hosted three informative sessions for participants:
- Marc Gratacos (Managing Partner at TradeHeader) delivered an amazing walkthrough on the Common Domain Model (CDM), which I’d highly recommend to anyone who wants to start digging into it;
- Martina Guzman (Pitch Coach and Master Trainer of the Best3Minutes methodology) delivered an interactive session on how to build and deliver a perfect pitch, which was very well received by participants.
- Sam Griek (Founder at EagleAI), showcased how multi-agent AI with Human in the Middle (HitM) methodology can accurately interpret CLO tests and triggers.
Finalists and Winners
The winning team developed a CDM-based real-time reporting engine for Asset Backed Securities (RegTech Track).
In order to deliver the solution, the team expanded the CDM so that it could accommodate the ABS deal-specific features. Then the team used Large Language Model Meta AI 3 (LLaMA3) to detect the eligible regulations based on the deal’s jurisdiction factors (e.g. EMIR for the “European derivatives” incorporated in the ABS). The solution updates the report and notifies any changes (delta).
This team tackled the LegalTech Track by building a copilot that maps an ABS prospectus to the CDM.
The team used a RAG to embed legal documents and save them in a vector database, and the LLaMA3 to obtain information concerning legal agreements, deal features and underlying assets pool composition. Finally, the team used Python-CDM to map the retrieved data into CDM and generate a JSON output.
Tonic, a solo participant, used OpenBB agents and data combined with the data made available by the real estate data provider, Immobiliare Insights, to optimise residential investments in Italy.
The app, EasyRealEstate, is a multi-agent team with coding and data analysis capability to segment real estate deals by any means or objectives, then automatically plan and help with decision making.
The team tackled the RegTech Track by personalising a FinGPT model that accurately interprets the EMIR regulation.
The personalised FinGPT model is based on LLaMA 2 and it provides lineage so that the users can trace back to the actual source, such as laws, guidelines or Q&As related to EMIR. This new model is more reliable and performant than a generic LLM when it comes to EMIR interpretation.
Another team explored the potential for AI in aiding the development of Digital Regulatory Reporting (DRR) code based on CDM. Discussing and proposing a number of applications, including the opportunity for AI to aid the maintenance of lineage between regulatory text and the executable code which implements regulatory requirements. An app was devised to review the existing DRR code and regulatory references within and validate they are correct, it is hoped that this can be shared with the FINOS community in the coming weeks.
I personally see a significant opportunity to integrate Fannie Mae and Freddie Mac loan-level data, provided by RiskSpan via Snowflake, into OS-Climate Data Commons and Physirisk workstream to solve challenges related to the Climate Risk track. (We haven’t found a team yet interested in this topic, if you are interested, reach out!).
Credits
Regardless of the small group of attendants, we’ve witnessed incredible energy and commitment from participants, which resulted in creative solutions, well-prepared pitches and lots of interest from the members of the jury.
The first thank you goes to those who joined and dedicated time and effort to make this event a success.
A special shoutout goes to the TradeHeader team, who shared their CDM excellence with the rest of the room, walked out of the event with the winning prize, and joined (en-masse) as first and left as last.
Thanks to TAO Solutions for jumping on board from the get-go and joining the jury so helping us assess contributions.
At Algoritmica we’re thrilled to collaborate with FINOS and we’ll continue working to bring more innovation around Structured Finance.
Calls to Action
- For participants - Reach out to help@finos.org and work with the FINOS team to contribute your code to FINOS Labs, gaining exposure to +80 members and the most active open source community in finance and fintech.
- For decision makers in finance - visit the FINOS website, sign up to the newsletter and share with us your interest in this topic; we’d also love to collaborate with you to host our next coding event.
