Deep Dive

Deep Dive

Apr 30, 2025

Apr 30, 2025

Can AI Forecast Inflation Better Than Economists? The Federal Reserve of St Louis Put It to the Test

Can AI Forecast Inflation Better Than Economists? The Federal Reserve of St Louis Put It to the Test

Watch Video

Watch Video

Watch Video

Predicting inflation is one of the most challenging economic puzzles facing everyone—from families planning their household budgets to central banks setting national interest rates.

The variables are vast, the changes sudden, and the outcomes deeply impactful. Now, new research from the Federal Reserve Bank of St. Louis suggests that artificial intelligence—specifically, large language models (LLMs)—may provide a valuable new lens on this elusive target.

The Problem with Traditional Forecasting

Historically, inflation forecasting has relied on surveys, expert judgments, and complex mathematical models. Think of the Survey of Professional Forecasters (SPF) or the intricate DSGE (Dynamic Stochastic General Equilibrium) models. While valuable, these tools can be time-consuming, costly, and often miss the mark during periods of rapid economic change.

Enter the AI Forecaster

The research by Federal Reserve economists Miguel Faria e Castro and Fernando Leibovici set out to explore whether an LLM—in this case, Google's PaLM (specifically, the Bison 001 model)—could offer better inflation predictions. LLMs are trained on massive amounts of text data, learning to detect patterns and absorb contextual information that humans might miss.

Simulating the Past to Predict the Future

The researchers didn’t just ask the model, "What’s next for inflation?" Instead, they used conditional forecasts. They prompted PaLM to act as though it was sitting in a past moment—say, February 2020—and asked it to predict future inflation using only the information available at that time. This allowed them to simulate real-world forecasting conditions without giving the model an unfair advantage.

A Closer Look at the Results

PaLM provided detailed forecasts and often even included economic reasoning—discussing Fed policy, oil prices, unemployment, and other contextual signals. Importantly, when tested across various time periods and forecast horizons, PaLM consistently outperformed the SPF in most cases, particularly during the volatile years of the COVID-19 pandemic.

Differences in Forecast Behavior

Both human experts and PaLM showed a tendency to anticipate inflation reverting to the Fed’s 2% target, but PaLM did so more slowly. This slower reversion suggests the AI may weigh incoming economic signals differently or be less anchored by policy targets, offering a more cautious take.

Robustness Checks and Reliability

The study addressed common AI criticisms—like randomness and prompt sensitivity—by testing various prompts, adjusting the model’s randomness settings, and using multiple forecasts to establish reliable averages. The researchers also tested whether the model could distinguish between real-time forecasting and simply recalling known outcomes, and the results suggested that it could.

Implications for Economic Forecasting

This research hints at a seismic shift in how economic forecasts might be conducted. LLMs offer a potentially faster, cheaper, and more scalable method of forecasting, especially for variables where traditional data is limited. Imagine businesses, local governments, or even individuals accessing real-time forecasts generated from current media, policy updates, and economic commentary.

Final Thought

While AI isn't replacing human economists just yet, this study shows it can significantly complement traditional forecasting methods. As the models improve and become more accessible, they may play an essential role in how we understand and anticipate economic changes—possibly even democratizing access to sophisticated economic insights.

FAQs

  1. What is the main takeaway from the St. Louis Fed's AI inflation study?
    Google's PaLM model was often more accurate than human forecasters at predicting inflation, especially for medium-term forecasts.

  2. What are conditional forecasts?
    Forecasts that simulate a past date and restrict the AI to only use information available up to that time.

  3. How was the AI tested for reliability?
    Researchers varied prompts, controlled for randomness, and compared multiple forecast outputs to ensure consistency.

  4. Does this mean AI will replace traditional economists?
    Not yet, but AI shows strong potential to augment and improve forecasting tools used by experts.

  5. Could AI be used for forecasting other economic indicators?
    Yes, especially in areas where traditional data is scarce or costly to collect.

Hashtags

#InflationForecasting #AIinEconomics #FederalReserve #StLouisFed #ArtificialIntelligence #GooglePaLM #EconomicForecasts #LLMResearch #FutureOfForecasting #FintechInnovation

Subscribe to our Newsletter

Ready to unlock the power of AI for your organization?

Let's discuss how we can partner to achieve your vision.

Address:

Urb. Four Seasons, Los Flamingos Golf,

29679 Benahavís (Málaga), Spain

Contact:

NIF:

ESB44635621

© 2024 Los Flamingos Research & Advisory. All rights reserved

Ready to unlock the power of AI for your organization?

Let's discuss how we can partner to achieve your vision.

Address:

Urb. Four Seasons, Los Flamingos Golf,

29679 Benahavís (Málaga), Spain

Contact:

NIF:

ESB44635621

© 2024 Los Flamingos Research & Advisory. All rights reserved

Ready to unlock the power of AI for your organization?

Let's discuss how we can partner to achieve your vision.

Address:

Urb. Four Seasons, Los Flamingos Golf,

29679 Benahavís (Málaga), Spain

Contact:

NIF:

ESB44635621

© 2024 Los Flamingos Research & Advisory. All rights reserved