Innovative ways of crunching big data are advancing our ability to make accurate predictions about key aspects of the economy, including GDP growth, employment, and interest-rate decisions. A new forecasting model from Wharton Nippon Life Professor in Finance Jules van Binsbergen and other researchers captures changing economic sentiment in the U.S. over 170 years to make these predictions. The researchers explain their model in the paper “(Almost) 200 Years of News-Based Economic Sentiment.”
The researchers customized a machine learning technique to analyze roughly one billion articles across 200 million pages of 13,000 U.S. local newspapers dating back to 1850, to measure economic sentiment granularly at the city, county, and state levels.
“If you want to better understand the role that sentiment has on economic activity, GDP growth, labor market decisions, and other fundamentals, it’s always better to have more granular and longer time series data,” says van Binsbergen. “We show this sentiment has predictive power for economic activity over and above the standard predictors.”
The AI model looks at the context in which a word or phrase is used in a given sentence. It classifies words and phrases related to the economy as positive or negative and measures the intensity of their sentiment.
Moving the Needle
Using this algorithm to capture sentiment in the future is possible, despite challenges in determining the right mix of media sources. Van Binsbergen is happy their study has advanced predictive techniques: “There are people that forecast GDP for a living. … It seems that our sentiment measure leads their forecasts, meaning that it seems to run more information than what professional forecasters seem to be using.”
Published as “Bringing More Predictive Power to Economic Forecasts” in the Fall/Winter 2023 issue of Wharton Magazine.