Executing with AI: Reverse Stress Testing using AI Example
Financial institutions are finding it difficult to insert AI into their financial risk products at scale. Financial institutions have been hiring data scientists, acquiring technology, doing POCs, in short, playing around with AI. While they understand it in theory, and may even put an AI project into production, getting it to run scale is another story. Financial firms understand AI’s power but the engineering effort to incorporate that technology into their existing platforms at scale requires a different effort.
Let me tell you how Microsoft has helped a investment management solutions company, Axioma, solve its engineering problem at scale. Axioma had developed a new mathematical way to think about reverse stress testing large financial portfolios. But the depth of AI required, and the technologies required were beyond the resources Axioma quantitative group had access to. They turned to Microsoft for help.
Quick background into Reserve Stress Testing. After the Great Financial Recession in 2007-2009, the US government began requiring stress testing as means of evaluating the solvency of a firm. A stress test identifies the economic scenarios that can put a company out of business. Normally stress test scenarios are chosen based on expert knowledge and historical evidence. The problem with the current way to do stress testing is it requires a huge computational budget and can take weeks to explore all the financial scenarios that could cause a company to fail. A reverse stress test explicitly identifies and assesses only those tail risk scenarios that most likely threaten a business, that is, cause the institution to fail and go into default. But the mathematically modeling to work backwards can be equally challenging, compute expensive and time consuming.
Now enter Axioma’s new mathematical and AI way of thinking about the problem. The challenge with reverse stress testing approach is it is a “counterfactual experiment.” In other words, it is first imagining catastrophic events affecting the financial markets, then working backwards to what the portfolio looks like after those events. What is needed is not cause-and-effect, but effect-and-cause reasoning. That’s not what AI does well (https://www.wsj.com/articles/ai-cant-reason-why-1526657442). AI typically works on the level of “association.” But Axioma developed a new approach to reverse stress testing but did not have the AI experience to develop their ideas into code.
Axioma reached out to Microsoft for help. Leveraging Azure Machine Learning Services, Microsoft helped create an end-to-end workflow with Axioma’s quant team and key data scientists from Microsoft to create a solution. That allowed the team to work seamlessly across organizational boundaries, enabling a faster time to solution
But how to run this at scaled.
The next critical decision was using the Open Neural Network Exchange (ONNX) standard. The ONNX standard is an open ecosystem that empowers AI developers to choose the right tools as their project evolves. ONNX provides an open source format for AI models, both deep learning and traditional ML. It defines an extensible computation graph model, as well as definitions of built-in operators and standard data types. More importantly, ONNX is widely supported and can be found in many frameworks, tools, and hardware. Onnx enables interoperability between different frameworks and streamlining the path from research to production. This helped Axioma increase the speed of innovation and enable embedding this solution into their existing solutions with minimum impact.
Outcomes – As a result of this work, Axioma has demonstrated reverse stress testing can be done in hours instead of several weeks of processing. Faster stress testing reduces capital requirements, enabling firms to invest more of their capital into better returns. Axioma’s Risk solutions can be found the Azure marketplace.
This work demonstrates
- Azure AI platform enables cross team and cross company collaboration to develop new AI solutions.
- The financial community can depend on and leverage Microsoft to develop new solutions based on advanced modeling and AI techniques to manage risk.
- With ONNX, financial firms can deploy models to data, instead of moving data to where the models run. Firms can scale development and testing on Microsoft Azure, and then deploy models to where the data is
Axioma, acquired by Qontigo, provides an integrated suite of front-to-back investment management solutions to a global client base, including asset managers, hedge funds, insurance companies, pension funds, wealth managers and investment banks. Axioma’s award-winning services are comprised of multi-asset enterprise risk management, portfolio construction, performance attribution, regulatory reporting and custom indexes.
Microsoft AI – Microsoft believes that, when designed with people at the center, AI can extend your capabilities, free you up for more creative and strategic endeavors, and help your organization achieve more. Microsoft’s work with organizations on the front lines of financial services gives Microsoft a unique perspective on the industry and where it is headed.
