How Events in the News Can Improve Commodity Price Predictions
Overview: In the paper “News Event-Driven Forecasting of Commodity Prices,” NYU Stern Professor Srikanth Jagabathula, along with co-authors Sunandan Chakraborty (Indiana University), Lakshminarayanan Subramanian (NYU Courant Institute), and Ashwin Venkataraman (UT Dallas), looks for a solution to more accurately forecast commodity prices given the volatility of real-world events that impact those prices.
Why study this now: Because of the volatility of global events, it has become increasingly difficult to price various commodities – which can range from precious and industrial metals to crops to natural gasoline. This volatility becomes a risk for individuals and businesses whose work deals with such commodities. Solely using historical pricing data to predict future prices is no longer useful because the external factors that impact prices are changing too rapidly. More accurate predictions can help people make better informed decisions.
What the authors found: By analyzing the daily prices of four crops in India (onions, potatoes, rice, and wheat), as well as event signals extracted from 1.6 million news articles over a 15-year period, price forecast accuracy increases by more than 16%. The authors propose what they call “an event-based methodology,” which automatically extracts relevant news signals from the text of the news articles and correlates them with price changes.
What does this change: These findings can provide a range of benefits, including:
- Help businesses optimize inventory and reduce costs
- Reduce supply chain risk from sudden price fluctuations
- Inform policy to reduce negative effects of potential price shocks
- Provide information about market trends to consumers who can use those findings to budget more effectively
The research also helps to understand why prices may change due to natural events, policy changes, elections, or other external risks.
Key insight: “Accurate commodity price forecasts can benefit individuals, businesses, governments, and the society as a whole,” say the authors. “Our work adds to the literature on developing data-driven methodologies to improve forecasting in the operations literature…and we believe our paper helps demonstrate how modern generative AI tools can help harness rich unstructured data to improve operational decisions.”
This research was published in Manufacturing & Service Operations Management.