Sensing the Fed’s Direction with the Help of AI

As Federal Reserve policy has a significant impact on markets and the economy, some pundits will go to extremes to predict what the Fed will do next—including speculating about Fed Chair Jerome Powell’s tie and whether he wears a particular color at press conferences as a signal.

Morgan Stanley technologists have found a better way. Several of them worked together in order to create a model that uses artificial intelligence (AI) and deep learning to interpret Fed sentiment, a model they were then able to patent with the help of our in-house Patent Accelerator Program.

Morgan Stanley launched the Patent Accelerator Program in 2018 to help identify promising inventions originating within the firm and then fast-track them through the patent process. The program, a joint initiative between the firm’s Intellectual Property Legal group and Firmwide Innovation, has been responsible for pursuing and obtaining hundreds of patents for inventions generated by Morgan Stanley employees across the firm, and the momentum is growing: In 2022, the number of patents granted through the initiative reached a record 41. “We give our innovators access to an extensive lab environment and the tools and technical experts needed to accelerate innovation,” says Larry Bromberg, Morgan Stanley’s Global Head of Intellectual Property Legal, who co-founded the Patent Accelerator. “This removes barriers and ensures experimentation happens in a safe and compliant environment. The idea is to make innovation accessible, fun and easy to do.”

Making Sense of the Fed

The story of the Fed sentiment model began back in 2019, when Ken Zhang, Head of Research Tech Data Science, and Qingyi Huang, Managing Director and Head of AlphaWise Quant Research, used AI to analyze sentiment in 41,758 of the firm’s research reports from January 2013 to May 2019. “Rather than isolating keywords as existing sentiment models do, we used a deep-learning approach to analyze complete sentences,” Huang wrote in a note to clients.

Next, they refined the analysis, including only reports with price target adjustments. Using this lens, the model showed 9.6% outperformance between those it identified as top-rated stocks and the lowest group. “That was our first model that was trained and validated using our research content, and that used a big data, deep-learning driven approach,” explains Zhang. “And it was very successful.” Indeed, the strategy showed a solid performance over the time period measured by the model, with a positive return in 2018 despite the market being down by 5%.

That result led Zhang and Huang to develop a similar AI-powered model aimed at analyzing Fed sentiment. “Some of our colleagues in Fixed Income Research told us that there hasn't been any deep-learning-driven sentiment model that could automatically quantify Fed Open Market Committee (FOMC) files,” explains Zhang. He notes that most sentiment analysis of central banks’ communications has focused on dictionary-based approaches, which makes them too rigid and difficult to keep up to date.

Their alternative, patented in 2022, gathers FOMC statements, with a special emphasis on monetary policy statements, and assesses the degree of hawkishness/dovishness/neutrality of words and phrases they contain, tracking the sentiment trend over time. Given the increased focus recently on Federal Reserve communications and its use as a policy tool and driver of asset prices, the sentiment scores, which are published as an index on the Bloomberg terminal under the ticker “MNLPFEDS,” fills a particularly timely need. “The central bank's statements are being watched closely by markets globally,” says Jasper Lin, Executive Director, AlphaWise Quant Research. “Everybody's looking at them. The equity market, the bond market. They are very important pieces of communication. And this machine provides a consistent, systematic way of reading them, and to quantify and provide absolute scale of sentiment.”

The model could potentially be used to analyze communications from other major central banks, as well as to devise a trading strategy that drives alpha based on the model’s predictive value. It’s an approach that Zhang acknowledges is more difficult than the initial model based on Morgan Stanley Research, since FOMC statements come out less frequently than research reports, meaning there’s less data to crunch.

Nevertheless, a recent report found there was correlation between the Fed sentiment score and other financial metrics, and the model has shown to be a ~1 year leading indicator of monetary policy actions. According to a note issued by a team of Morgan Stanley researchers, the model exhibits a clear and substantial lead over monetary policy action, adding forecasting power to the level of interest rates, the shape of the yield curve and the direction of the U.S. dollar.

A Culture of Innovation

The success of the Fed sentiment model is nothing new to the people across the firm who have used the Patent Accelerator Program to bring their inventions to life. “Our Patent Accelerator is part of a broader, comprehensive framework that we have established to drive innovation in a structured way across the firm. It’s a concerted effort across business and technology to deliver high-quality, scalable solutions for our clients and business,” says Peter Akwaboah, Chief Operating Officer for Technology & Global Head of Innovation.

“We support our innovators through all stages of the process and make sure the experience runs smoothly,” says Bromberg. But just as important is how the Patent Accelerator Program encourages and acknowledges the employees behind those innovations. Says Bromberg, “We help our innovators receive the recognition they deserve both internally and externally. From a human point of view, there’s nothing better than that."

Summary: Current sentiment tools can be imprecise and clumsy. Morgan Stanley technologists have created a model that uses artificial intelligence (AI) and deep learning to interpret Fed sentiment, a model they were then able to patent with the help of our in-house Patent Accelerator Program.