Find out more at btsub.sg/btdeal
You are here
Mining news stories catching on in economic forecasting
[WASHINGTON] Government officials and investors are on a constant quest for more and better information about the economy to gauge what policy levers should be pulled, or just to help them make money.
So in the era of "big data" and algorithms scraping US President Donald Trump's Twitter feed for trading signals, it was perhaps inevitable that economists would try to mine the vast array of media and social media sources to help decipher the direction of the economy.
There are many big data projects underway, but some recent efforts involve scouring decades of news stories for key words indicating pessimism or worry to build a sentiment index to track where the economy is heading.
Economist George Stigler wrote in 1967 that "information costs are the costs of transportation from ignorance to omniscience, and seldom can a trader afford to take the entire trip." But the declining costs of computing power as well as ready availability of the raw material, are now allowing economists to build these non-traditional indicators.
And as it turns out, they are pretty accurate.
Using "machine learning" techniques to digest more than three decades of articles from 16 major US newspapers, experts at the Federal Reserve Bank of San Francisco have come up with an index that outperformed the highly-regarded measures of consumer sentiment used for years as market benchmarks.
"There is evidence that the news sentiment measures have significant predictive power," the creators of the new index - Adam Shapiro and Daniel Wilson of the San Francisco Fed and Moritz Sudhof, of analytics firm Kanjoya (now Ultimate Software) - said in a paper published last month.
They showed that the new economic indicator can predict key data points, such as consumption, employment and inflation as well as the S&P 500 stock index, a year into the future.
Whereas traditional consumer confidence indexes are based on consumer surveys, the technique they developed scans news stories for words like "worried," "anxious," "satisfied" or "confident" to gauge sentiment, and then tracks changes from month to month.
"We were surprised at how well it worked," Mr Shapiro told AFP. "Going into the project we had no idea how well it would work and we were pleasantly surprised it had forecasting power."
Another paper from February 2016, showed that using news sentiment indexes can help predict economic growth in the United States and other countries, and improved the accuracy of estimates by professional forecasters, especially during an economic downturn.
"Sentiment measures contain information which is not accounted for by professional forecasters," wrote author Samuel Fraiberger, of Harvard, Northeastern and New York Universities.
And as a result, "News-based sentiment measures lead to a 19 per cent average reduction in forecast error of GDP growth relative to consensus forecasts." "Reductions in forecast errors are also larger during bad times which indicates that forecasters tend to underreact to bad news."
Kevin Kliesen, business economist with the St Louis Federal Reserve Bank, said use of nontraditional economic indicators are "all the rage these days" and can "help paint a picture of economy."
The St Louis Fed is in the process of building a new labor market index using newspaper stories with keywords like "layoffs" and "hiring" to capture sentiment.
"It's a different way of baking the cake," Mr Kliesen said.
But even with the vast amounts of data available, including credit card transactions or Facebook and Twitter posts, processing the sheer volume of information makes it difficult and expensive to get something useful.
Mr Wilson at the San Francisco Fed said his team hopes to get to the point where the news sentiment index they created can be incorporated into the bank's forecasts and published. They also are looking into the possibility of doing a similar word analysis of speeches by Fed officials.
"This was kind of a proof of concept" that showed these techniques "can be useful for tracking the economy." Now "what we're hoping to do is to be able to refine the process and eventually be able to measure this kind of news sentiment in real time, and hopefully, eventually, make that a public product from the San Francisco Fed."
How long that will take depends on how soon they can speed up the computing process to create the index in real time.