Forecasting leading economic indicators in the US from financial news using multi-task learning
Article
Ying, JJC, Liu, CC, Tseng, VS et al. (2025). Forecasting leading economic indicators in the US from financial news using multi-task learning
. SOFT COMPUTING, 29(7), 3641-3657. 10.1007/s00500-025-10629-5
Ying, JJC, Liu, CC, Tseng, VS et al. (2025). Forecasting leading economic indicators in the US from financial news using multi-task learning
. SOFT COMPUTING, 29(7), 3641-3657. 10.1007/s00500-025-10629-5
Leading economic indicators are crucial statistics that are based on the economy and have a significant impact on various factors such as policies, stock market trends, etc. However, due to the long release time interval of these indicators, it is challenging to accurately predict their trends in advance. In this paper, we propose an effective framework for forecasting leading economic indicators in the United States by utilizing leading indicators to predict each other and learn their mutual correlation through an attention mechanism. To achieve this, we present a hierarchical, multi-task learning approach that uses textual data to make predictions for four essential leading economic indicators. We first group news articles by topic and then summarize and extract information using a time-series method and attention. Finally, we concatenate news information with historical data to forecast the leading economic indicators. Our experimental results demonstrate that our approach successfully forecasts leading economic indicators, and we are able to predict four indicators one month beforehand. This framework has significant potential for use in the finance industry and can be used to inform important decisions related to investment strategies, policymaking, and other financial aspects.
