--- title: Using AI to Model FOMC Decisions description: Research perspective on the AI-based FOMC decision model developed for a recent academic paper co-authored by Hayden Helm and Andrew Dassori. image: https://blog.wavelengthcapital.com/hubfs/jonakoh-_-qY-XlCf9Zs4-unsplash-2.jpg --- [![Wavelength Insights & Ideas](https://blog.wavelengthcapital.com/hs-fs/hubfs/WavelengthCM_Logo_Horizontal_Blue.jpg?width=4200&height=400&name=WavelengthCM_Logo_Horizontal_Blue.jpg "Wavelength Insights & Ideas")](https://wavelengthcapital.webflow.io/research-insights) Open main menu Close main menu [Fixed Income](https://blog.wavelengthcapital.com/tag/fixed-income) [Interest Rates](https://blog.wavelengthcapital.com/tag/interest-rates) [Economy](https://blog.wavelengthcapital.com/tag/economy) [Markets](https://blog.wavelengthcapital.com/tag/markets) [Federal Reserve](https://blog.wavelengthcapital.com/tag/federal-reserve) [Monetary Policy](https://blog.wavelengthcapital.com/tag/monetary-policy) # Using AI to Model FOMC Decisions ![Andrew Dassori, Chief Investment Officer](https://blog.wavelengthcapital.com/hs-fs/hubfs/Andrew%20Dassori%20Headshot.jpg?width=45&name=Andrew%20Dassori%20Headshot.jpg) [Andrew Dassori, Chief Investment Officer](https://blog.wavelengthcapital.com/author/andrew-dassori) [Share this blog post on Twitter](https://twitter.com/intent/tweet?text=I+found+this+interesting+blog+post&url=https://blog.wavelengthcapital.com/using-ai-to-model-fomc-decisions) [Share this blog post on Facebook](http://www.facebook.com/share.php?u=https://blog.wavelengthcapital.com/using-ai-to-model-fomc-decisions) [Share this blog post on LinkedIn](http://www.linkedin.com/shareArticle?mini=true&url=https://blog.wavelengthcapital.com/using-ai-to-model-fomc-decisions) ![](https://blog.wavelengthcapital.com/hubfs/jonakoh-_-qY-XlCf9Zs4-unsplash-2.jpg) - While AI has been deployed widely by investors for automation, it has been used less to provide insights that are otherwise not humanly possible. - Knowing new tools can extract unique information from an institution with an outsized impact on markets – the FOMC – we built a framework using AI-based personas to model member decisions and combined this with economic modeling to forecast targets for interest rates. - The persona-based index we constructed correctly predicted the FOMC’s decision to hike, hold, or cut interest rates in 69% of the 32 out-of-sample meetings from 2022 through 2025, beating each of the model-based alternatives we tested. - Our approach shows the ability of language models to represent the policy stances of FOMC members, and its synthesis with economic modeling reveals new and useful insights into a critical rate for financial markets. The full paper, [A Persona-Based Rate Action Index](https://arxiv.org/pdf/2607.26545), can be viewed along with an interactive persona-based simulation at [www.federalreserve.ai](http://www.federalreserve.ai). ##### Background After facing early hurdles, AI has swept through the investment landscape with a flood of new tools that are expanding the capabilities of asset managers. This initial wave has led to widespread automation, and managing existing tasks agentically has improved productivity across the industry. With that said, in most cases, AI has yet to be used by investors to garner differentiated insights that would not be humanly possible, despite this being where the technology has the most transformative potential. Using new methods to extract and synthesize layers of information outside of human capacities can be impactful beyond automation and offer new dimensions of analysis for investors. Given this and the power of tools now available through recent technological developments, we built an AI-based framework that seeks to model the decision-making process for one of the most consequential institutions when it comes to financial markets: the Federal Open Market Committee (FOMC). From an investment perspective, few decisions are more important than those made by the FOMC when setting interest rates, and their statements have long been examined in word-for-word detail for clues about the future trajectory of rates. In [*A Persona-Based Rate Action Index*](https://arxiv.org/pdf/2607.26545), we establish a process for identifying, understanding, and aggregating FOMC member voting behavior through the lens of language models and then combine this with economic intuition to forecast the future path of interest rates. The synthesis of these models offers a logical approach to incorporating new technology in areas that have long been of interest to investors, while its leading nature demonstrates the potential for predictive power to provide useful and differentiated insight from existing approaches. ##### The Federal Reserve in Focus As described above, the Federal Reserve is among the most closely watched institutions for financial market participants with a storied history and deep set of data going back over a century. The branch responsible for monetary policy, the FOMC, sets the federal funds rate target at eight meetings each year, where twelve committee members cast votes on policy seeking to meet their dual mandate of maximum employment and price stability. ###### Figure 1: FOMC Membership & Voting Structure ![FOMC Membership & Voting - MODIFIED](https://blog.wavelengthcapital.com/hs-fs/hubfs/FOMC%20Membership%20%26%20Voting%20-%20MODIFIED.png?width=1600&height=796&name=FOMC%20Membership%20%26%20Voting%20-%20MODIFIED.png) Because the federal funds rate functions as the base rate from which the cost of other borrowing is priced in the US economy, investors go to great lengths to predict its forward path. Futures prices, economic conditions, and Fed statements are all used extensively to forecast the FOMC’s target, but none of these variables incorporate how the committee is composed of members with different views that shape each critical interest rate decision. ##### Analyzing Data with Large Language Models To go along with its meaningful impact on markets, the FOMC features an expansive set of data that is structurally well-suited for building AI-based personas with large language models (LLMs). AI-based personas are created by training LLMs on personal data libraries, and members of the FOMC have this information readily available in transcripts of speeches, testimony, meetings, and press conferences. The resulting outputs can be tested for identity at the individual level and aggregated by voting for decisions at the committee level, providing a powerful tool to track the overall view as it shifts with the economy. Through this process, we gathered 24,333 publicly available excerpts for FOMC members from 2006 to 2026 and created personal data libraries for each member, then built personas from statements covering a range of economic topics. We tested the resulting personas on distinguishability from one another and whether they identify who they are meant to represent, and all passed the criteria – with the amount of data interestingly not translating to better representation. Based on how they are characterized by prior statements, the monetary policy views of personas can be measured with questions by scoring responses on a hawk-dove scale. Using this approach, we ranked each persona ordinally based on hawk-dove score, and these results were in close agreement with established reputational rankings of FOMC members. ##### Building a Real-World Rate Action Index To transform these monetary policy views into an active measure that is relevant for investors, we integrated time-varying market and economic information to track changing behavior and establish a persona-based index (PBI). Data on CPI inflation, core PCE inflation, the unemployment rate, and the Federal Funds Rate Target were incorporated to create the PBI, and responses from each persona reflect the time-varying impact of these factors. ###### Figure 2: Persona-Based Index vs. Fed Funds Target Model Results from 2018-2025 ![PBI vs Fed Funds Target](https://blog.wavelengthcapital.com/hs-fs/hubfs/PBI%20vs%20Fed%20Funds%20Target.png?width=624&height=175&name=PBI%20vs%20Fed%20Funds%20Target.png) The synthesis of persona-based and economic modeling used to form the PBI shows not only an ability to track voting behavior and aggregate policy decisions at the FOMC, but also that these methods may hold predictive value. As shown in Figure 2, the PBI regularly leads the Fed Funds Target Rate, demonstrating this potential. ##### Key Findings As highlighted above, the persona-based index was able to correctly predict the FOMC’s decision to hike, hold, or cut interest rates in 69% of the 32 out-of-sample FOMC meetings from 2022 through 2025. This performance beat each of the model-based alternatives we tested, and while it underperforms blindly repeating the prior policy decision, that approach will be wrong at every turning point, which is where these predictions can be most impactful. ###### Figure 3: Persona-Based Index Predicted 69% of FOMC Decisions 2022-2025 Share of 32 out-of-sample meetings where correctly predicted hike, hold or cut ![Share of out of sample meetings correct](https://blog.wavelengthcapital.com/hs-fs/hubfs/Share%20of%20out%20of%20sample%20meetings%20correct.png?width=580&height=294&name=Share%20of%20out%20of%20sample%20meetings%20correct.png) Of the model-based alternatives, the macro model comparison is most notable. Given the same inflation, unemployment, and rate figures, the PBI meaningfully outperformed this multi-factor approach, showing the ability of personas to add predictive power beyond what is available in the raw numbers. It is also worth noting that for the out-of-sample period tested, the PBI was more effective at predicting hikes than decisions to cut or hold the Fed Funds Target Rate steady, and the index tends to lead policy by turning hawkish or dovish months before the FOMC’s decision to make a change. The PBI effectively traces the Fed Funds Target Rate over the full cycle through both hawkish and dovish policy regimes from 2018-2025, and its potential for predictive power – particularly at turning points in the cycle – can offer a valuable and differentiated measure in a crowd of one-dimensional policy frameworks. ##### Takeaways for Investors Our approach shows how language models can be used in areas of increased interest for investors by representing the policy stances of FOMC members, and our synthesis with economic data demonstrates the ability to model the changing view of the committee over time. The potential to use this framework when thinking ahead about critical policy decisions may create real value for investors and provide a prototype for the use of LLMs in other areas that add new dimensions of analysis to investment processes. As new tools increasingly give machines advantages when it comes to synthesizing large amounts of disparate information to make decisions, understanding how to best utilize these tools will also become increasingly important for asset managers. In this context, we see further changes on the horizon driven by the ongoing paradigm shift in technology. Our full paper demonstrates the ability to capture time-varying group behavior with a collection of digital personas that are impactful for financial markets – in the form of the FOMC – and can be viewed along with an interactive persona-based simulation at [www.federalreserve.ai](http://www.federalreserve.ai). \*\*\* **DISCLAIMER:** Wavelength Capital Management, LLC (“Wavelength”) is an SEC-registered¹ investment adviser located in Connecticut. Wavelength may only transact business in those states in which it is notice filed or qualifies for a corresponding exemption from such requirements. Wavelength's website is limited to the dissemination of general information regarding its investment advisory services to United States residents residing in states where providing such information is not prohibited by applicable law. Accordingly, the publication of Wavelength's website on the Internet should not be construed as Wavelength's solicitation to effect, or attempt to effect, transactions in securities or the rendering of personalized investment advice for compensation over the Internet. Investing involves the risk of loss and investors should be prepared to bear potential losses. Past performance may not be indicative of future results and may have been impacted by events and economic conditions that will not prevail in the future. No portion of this commentary is to be construed as a solicitation to buy or sell a security or the provision of personalized investment, tax or legal advice. Certain information contained in this report is derived from sources that Wavelength believes to be reliable; however, the Firm does not guarantee the accuracy or timeliness of such information and assumes no liability for any resulting damages. For information pertaining to the registration status of Wavelength, please view the United States Securities and Exchange Commission’s website at www.adviserinfo.sec.gov. A copy of Wavelength's current written disclosure statement discussing Wavelength's business operations, services and fees is available from Wavelength upon written request. Wavelength does not make any representations as to the accuracy, timeliness, suitability, completeness or relevance of any information prepared by any unaffiliated third party, whether linked to Wavelength's website or incorporated herein, and takes no responsibility therefore. All such information is provided solely for convenience purposes only and all users thereof should be guided accordingly. ¹ SEC registration does not indicate a certain level of skill or training. ![Andrew Dassori, Chief Investment Officer](https://blog.wavelengthcapital.com/hs-fs/hubfs/Andrew%20Dassori%20Headshot.jpg?width=150&name=Andrew%20Dassori%20Headshot.jpg) #### Andrew Dassori, Chief Investment Officer [Follow me on LinkedIn](https://www.linkedin.com/in/andrew-dassori-0112898/) Andrew founded Wavelength Capital Management in 2013 to make forward-thinking strategies available to all investors and is the firm’s Chief Investment Officer. In this role, he oversees all investment functions at the firm, including portfolio management, research and trading. He was previously a portfolio manager at Credit Suisse Asset Management focused on global macro investment strategies, and served as a member of the firm's Global Citizens Program. After growing up on the South Shore of Boston, he has lived in London, New York, and Nairobi, and now resides with his wife and two daughters in Connecticut. #### About - [Our Firm](http://wavelengthcapital.com/firm) - [Investment Philosophy](http://wavelengthcapital.com/philosophy) - [Fund Information](http://wavelengthcapital.com/funds) - [News](http://wavelengthcapital.com/news) [Follow us on LinkedIn](https://www.linkedin.com/company/wavelength-capital-management-llc) [Visit our website](https://blog.wavelengthcapital.com/www.wavelengthcapital.com) [Follow us on Twitter](https://twitter.com/wavelengthfunds) ### Follow our Insights DISCLAIMER: Wavelength Capital Management, LLC (“Wavelength”) is an SEC-registered¹ investment adviser located in Westport, Connecticut. Wavelength may only transact business in those states in which it is notice filed or qualifies for a corresponding exemption from such requirements. Wavelength's website is limited to the dissemination of general information regarding its investment advisory services to United States residents residing in states where providing such information is not prohibited by applicable law. Accordingly, the publication of Wavelength's website on the Internet should not be construed as Wavelength's solicitation to effect, or attempt to effect, transactions in securities or the rendering of personalized investment advice for compensation over the Internet. Investing involves the risk of loss and investors should be prepared to bear potential losses. Past performance may not be indicative of future results and may have been impacted by events and economic conditions that will not prevail in the future. No portion of this commentary is to be construed as a solicitation to buy or sell a security or the provision of personalized investment, tax or legal advice. Certain information contained in this report is derived from sources that Wavelength believes to be reliable; however, the Firm does not guarantee the accuracy or timeliness of such information and assumes no liability for any resulting damages. For information pertaining to the registration status of Wavelength, please view the United States Securities and Exchange Commission’s website at www.adviserinfo.sec.gov. A copy of Wavelength's current written disclosure statement discussing Wavelength's business operations, services and fees is available from Wavelength upon written request. Wavelength does not make any representations as to the accuracy, timeliness, suitability, completeness or relevance of any information prepared by any unaffiliated third party, whether linked to Wavelength's website or incorporated herein, and takes no responsibility therefore. All such information is provided solely for convenience purposes only and all users thereof should be guided accordingly. ¹ SEC registration does not indicate a certain level of skill or training.