Using News Events To Predict Food Price Fluctuations
Four researchers from New York University (NYU) are using real-world events, identified from the first paragraphs of news stories, to forecast volatility in socio-economic indicators like food prices. Identifying newsworthy events that cause fluctuations in food prices could have a meaningful application in the developing world.
Governments diligently track food price indices; not only for the domestic impact they have on trade but also because spikes in food prices have significant effects on the standard of living, particularly in the developing world (e.g. world food crisis in 2008). Accurately capturing what drives a price fluctuation for a specific commodity is what this model, published in August, aims to do. The team suggests the model could be extended to predict changes in any event-sensitive indicator.
According to the team’s research paper, incidents or events that can have a sudden impact on food prices were not accounted for in existing predictive models. To remedy this, their new model mines through large amounts of text (from news, blogs, or social media) and sorts “events” into common themes based on the action words found as descriptors in the news stories. It lets the data lead the way, which is a stark contrast to previous models that require researchers to pre-determine a fixed number of themes when classifying data.
The team qualitatively tested their model using six years of historical data on food prices in India and more than 700,000 articles published in the Times of India between 2006 and 2012. Adding their event-classification framework onto an existing predictive model successfully reduced the error in forecasting food prices by 22 percent, when compared with the existing model’s predictive abilities alone. Another test using their model showed an increased accuracy of 5 to 10 percent in predicting significant spikes in food prices.
Nicholas Phythian, a former Reuters journalist who now consults with journalists covering sustainable development, said that accurate forecasting of food prices can have systemic positive effects on small-scale agriculture in developing countries. Phythian cites the impact of groups such as the Ethiopian Commodity Exchange, which gives farmers access to domestic and global food prices. A new predictive model would take things one step further. For a small farmer, “telling him not just what prices are but where they will go” could have major spillover effects on “sustenance, education, investment,” Phythian said.
The researchers plan on creating a second iteration of their model to account for food price fluctuations due to typical seasonal factors, which had not been accounted for in the news stories. They are interested in using their model to predict disease outbreaks and stock prices through the same analysis of news events.
With high volatility in commodity prices over the past decade, this development serves as a reminder of the interconnectedness of our social and economic systems.
Next time you notice your grocery store increase its prices, take a quick look at the front page of the newspaper as you head for the checkout line.
To read the full research paper, published by the Association of Computing Machinery in August, please see the citation below:
Chakraborty, Sunandan, Ashwin Venkataraman, Srikanth Jagabathula, and Lakshminarayanan Subramanian. “Predicting Socio-Economic Indicators Using News Events.” Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining — KDD ’16, August 2016. doi:10.1145/2939672.2939817
