{"id":45288,"date":"2020-03-23T11:42:35","date_gmt":"2020-03-23T18:42:35","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/industry\/microsoft-in-business\/?p=45288"},"modified":"2023-06-16T08:47:19","modified_gmt":"2023-06-16T15:47:19","slug":"executing-with-ai-reverse-stress-testing-using-ai-example","status":"publish","type":"post","link":"https:\/\/www.microsoft.com\/en-us\/industry\/microsoft-in-business\/business-transformation\/2020\/03\/23\/executing-with-ai-reverse-stress-testing-using-ai-example\/","title":{"rendered":"Executing with AI: Reverse Stress Testing using AI Example"},"content":{"rendered":"

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.\u00a0 While they understand it in theory, and may even put an AI project into production, getting it to run scale is another story.\u00a0 Financial firms understand AI\u2019s power but the engineering effort to incorporate that technology into their existing platforms at scale requires a different effort.<\/p>\n

Let me tell you how Microsoft has helped a investment management solutions company, Axioma, solve its engineering problem at scale.\u00a0 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.\u00a0 They turned to Microsoft for help.<\/p>\n

Quick background into Reserve Stress Testing.\u00a0 After the Great Financial Recession in 2007-2009, the US government began requiring stress testing as means of evaluating the solvency of a firm.\u00a0\u00a0 A stress test identifies the economic scenarios that can put a company out of business.\u00a0 Normally stress test scenarios are chosen based on expert knowledge and historical evidence.\u00a0 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.\u00a0 But the mathematically modeling to work backwards can be equally challenging, compute expensive and time consuming.<\/p>\n

Now enter Axioma\u2019s new mathematical and AI way of thinking about the problem.\u00a0 The challenge with\u00a0 reverse stress testing approach is it is a \u201ccounterfactual experiment.\u201d\u00a0 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.\u00a0 What is needed is not cause-and-effect, but effect-and-cause reasoning.\u00a0 That\u2019s not what AI does well (https:\/\/www.wsj.com\/articles\/ai-cant-reason-why-1526657442<\/a>).\u00a0 AI typically works on the level of \u201cassociation.\u201d\u00a0 But Axioma developed a new approach to reverse stress testing but did not have the AI experience to develop their ideas into code.<\/p>\n

Axioma reached out to Microsoft for help.\u00a0 Leveraging Azure Machine Learning Services, Microsoft helped create an end-to-end workflow with Axioma\u2019s quant team and key data scientists from Microsoft to create a solution.\u00a0 That allowed the team to work seamlessly across organizational boundaries, enabling a faster time to solution<\/p>\n

But how to run this at scaled.<\/p>\n

The next critical decision was using the Open Neural Network Exchange (ONNX<\/a>) standard.\u00a0 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.\u00a0 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.\u00a0\u00a0 This helped Axioma increase the speed of innovation and enable embedding this solution into their existing solutions with minimum impact.<\/p>\n

Outcomes \u2013 As a result of this work, Axioma has demonstrated reverse stress testing can be done in hours instead of several weeks of processing.\u00a0 Faster stress testing reduces capital requirements, enabling firms to invest more of their capital into better returns.\u00a0 Axioma\u2019s Risk<\/a> solutions can be found the Azure marketplace.<\/p>\n

This work demonstrates<\/p>\n