The use of large-language models may be allowing economists to turn narratives into data. That’s especially helpful when parsing the Federal Reserve’s Beige Book, a report issued eight times a year that compiles anecdotes about the economy from across the Fed’s 12 districts and provides a national summary of the economy.
Their conclusion: The sum of the parts is more than the whole.
Four economic researchers applied such technology to determine if the Beige Book could help identify or predict recessions. Specifically, they used FinBERT, a variant of BERT, a deep learning model using natural language processing that was developed by Google researchers in 2018.
The model is more useful than word counts or simpler language processing models. “By using surrounding text, rather than simply reading from left to right, BERT aims to establish the context of the text and thus infer the meaning of language that might otherwise be ambiguous,” the authors wrote.
“Our focus is establishing (reduced-form) statistical facts about district-level and national sentiment in the Beige Book and examining their predictive power for U.S. recessions,” they wrote in an Economic Commentary published by the Cleveland Fed. “In so doing, our sentiment measure is an aggregated measure in the sense that it quantifies sentiment across different areas of the economy, such as overall economic activity, labor markets, prices, and different sectors.”
The authors — which include two Chicago Fed economists and a visiting assistant professor at the Washington University in St. Louis, and a postdoctoral research associate in finance at the same university — were particularly interested in how well the overall national sentiment, as conveyed in the Beige Book, compared with a consensus sentiment derived from the 12 districts.
They found that “the national economic sentiment does not always lie in the middle of the district-level estimates, as one might expect.” And since 2021, the national sentiment has been more positive than in most of the individual districts, they said.
The national sentiment, it appears, weighted some districts heavier than others. And their analysis indicated that the national measure underweighted the Cleveland, Minneapolis, Philadelphia, and St. Louis districts.
By contrast, the economists found that sentiment in Chicago, Minneapolis, Philadelphia, Richmond, and San Francisco was “most helpful in predicting current U.S. business cycle turning points.” The data showed that as sentiment improved in each district, there was an increased chance of the U.S. economy being in an expansionary phase.
And Boston’s district proved most useful in predicting three and six months ahead, they said.
From the mid-1980s until the onset of the COVID-19 pandemic in March 2020, the authors said they observed no instances of false alarms, “that is, of recession probabilities spiking above around 40% without a recession actually taking place.”
As of March 2024, the analysis suggested that the “probability of a national recession was low, conditioning on the national and district-level sentiment data as extracted from the March 2024 Beige Book. But this probability has bounced around considerably since the pandemic-induced recession of early 2020.”
“In summary, the observed heterogeneity in district-level economic sentiment can be used, over and above the information contained in national economic sentiment, to better forecast U.S. recessions,” they said.