How much is that doggy?
Big data provide new ways to gauge price rises
INFLATION is a simple concept, but price rises are surprisingly hard to measure. First, statisticians must work out what stuff people buy, and in what proportions (the “basket” of goods). Then they must track the prices of those goods over time. Finally they must decide how to account for new products, changing tastes and the fact that if the price of, say, apples rises, some people will buy another fruit instead rather than pay more.
Big data could make all of this easier. At the moment, calculating America’s consumer-price index (CPI) involves sending people into shops to note down prices. The basket is based on a survey of consumers which is updated only every three years or so. This looks increasingly cumbersome in a world where every online purchase is logged, somewhere, in a database. In theory online baskets and prices, at least, could be tracked digitally.
Adobe, a technology firm, is trying to do just that. The firm collects anonymised sales data from websites that use its software. The amount of data available is vast: according to the firm, it includes three-quarters of online spending at America’s top 500 retailers. It is using this ocean of information to compile a “digital price index” (DPI) to rival official measures of inflation.
Two economists, Pete Klenow of Stanford University and Austan Goolsbee of the University of Chicago, are helping the firm to crunch the numbers.
The DPI has several advantages over the conventional approach. It tracks 1.4m goods, compared with the CPI’s 80,000. It is based on actual purchases rather than advertised prices, increasing its accuracy. And the volume of data allows Messrs Klenow and Goolsbee to use fancier statistical methods to account for people changing what they buy as prices move.
The new index completely misses changes in offline prices and spending on things like petrol and rent. It will not replace the CPI any time soon. It does suggest, however, that official statistics may themselves be missing big price movements, especially for consumer technology.
The researchers found that the price of computers fell by 13.1% in the year to January, almost double the 7.1% fall recorded in the CPI. Televisions fell more in price than the CPI reports, too. The speed of innovation in technology might account for the difference. The researchers found that fully 80% of technology spending is on new products, which the more nimble DPI can incorporate quickly.
If this is a widespread phenomenon, and inflation is lower than officially recorded, that has implications for central bankers, borrowers, savers and anyone who strikes long-term contracts. It also means that GDP might be understated, says Mr Klenow. If overall spending is recorded accurately but inflation is exaggerated, output must be higher than thought.
Official statisticians are improving their methods. The CPI includes some prices that are collected automatically by “scraping” websites (something Britain’s statisticians are also experimenting with).
But if their take-up of big data is sluggish, official statistics could eventually face disruptive private-sector competition.