It’s the data, stupid: a foundation for robust AI models

In the aftermath of the U.S. Presidential election, there are plenty of questions about the interpretation of polling data. Yet the ‘known unknowns’ like voter turnout means the ability of this data to predict a specific result will likely remain limited.

It’s a very different data story in financial markets. In fixed income, an explosion in available data means market participants are challenged to analyze vast amounts of information in near real time. It’s a task that exceeds human capabilities, where tools like artificial intelligence could be transformative – picture huge advances in algorithmic trading, risk management, or automating best trade execution. Whether applications like this are realized remains to be seen. But we know one thing for certain: AI financial models need a ton of data to learn.

For many market participants, ICE’s evaluated pricing and reference data is appealing not just for its quality and coverage, but because we are an independent source. Importantly, our approach combines both system and human analyses, so we never rely on black box models. Our fixed income offering includes analytics that add supplementary information around the price of a security, providing color to help support fixed income trading through various market conditions. These analytics include Liquidity Indicators, Size-Adjusted Pricing, Best Execution, Transaction Cost Analysis, Continuous Market Depth Indicators and Market Sentiment – tools which help frame out the market and characteristics of a bond, so users can be more precise with how orders are executed, and better identify trading risks and opportunities.

In addition to supporting client AI models, we’re using AI to advance fixed income markets. Here, we’re focused on three areas: improving data precision, timeliness, and helping to scale efficiencies. For example, using AI to extract data directly from documents presents huge gains in all these areas. Already, the attributes of new issue bonds and mortgage documents can be extracted with straight through processing and delivered to client workflows. In another use case, we apply machine learning to deliver spatial precision in assessing physical climate risk for any patch of dirt across the U.S. and globally. In each of these scenarios, the fat finger phenomenon of human error is removed, and the speed of work increases significantly.

AI presents huge potential for financial markets. But it remains a nascent technology, one which needs to be vetted to deliver customer value, tested, and built upon using a sound foundation of data. At ICE, we’re making significant investments to help clients build that foundation – and I’m excited to watch the innovation that follows.

Read more in the latest Fixed Income Monthly: https://www.ice.com/fixed-income-data-services/fixed-income/ice-fixed-income-monthly-report