Big data experiment tests central banking assumptions in Norway
CENTRAL BANKERS may do well to pay less attention to the bond market and their own forecasts than they do to newspaper articles.
That’s the somewhat heretical finding of a new algorithm-based index being tested at Norway’s central bank in Oslo. Researchers fed 26 years of news (or 459,745 news articles) from local business daily Dagens Naringsliv into a macroeconomic model to create a “newsy coincident index of business cycles” to help it gauge the state of the economy.
Leif-Anders Thorsrud, a senior researcher at the bank who started the project while getting his Ph.D. at the Norwegian Business School, says the “hypothesis is quite simple: the more that is written on a subject at a time, the more important the subject could be.”
He’s already working on a new paper ( yet to be published) showing it’s possible to make trades on the information. According to Mr. Thorsrud, the work is part of a broader “big data revolution.”
Big data and algorithms have become buzzwords for hedge funds and researchers looking for an analytical edge when reading economic and political trends. For central bankers, the research could provide precious input to help them steer policy through an unprecedented era of monetary stimulus, with history potentially a serving as a poor guide in predicting outcomes. At Norway’s central bank, researchers have found a close correlation between news and economic developments. Their index also gives a day- to- day picture of how the economy is performing, and do so earlier than lagging macroeconomic data.
But even more importantly, big data can be used to predict where the economy is heading, beating the central bank’s own forecasts by about 10%, according to Mr. Thorsrud. The index also showed it was a better predictor of the recession in the early 2000s than market indicators such as stocks or bonds.
The central bank has hired machines, which pore daily through articles from Dagens Naringsliv and divide current affairs into topics and into words with either positive or negative connotations. The data is then fed into a macroeconomic model employed by the central bank, which spits out a proxy of GDP.
Mr. Thorsrud says the results of the index are definitely “policy relevant,” though it’s up to the operative policy makers whether they will start using the information. Other central bank such as the Bank of England are looking at similar tools, he said.
While still in an experimental stage, the bank has set aside more resources to continue the research, Mr. Thorsrud said. “In time this could be a useful in the operative part of the bank.”