Central banks have found no shortage of potential uses for big data, from signalling financial trouble to analysing the readability of their documents. Most uses, however, remain largely within the research department, and are only gradually flowing into the central bank’s broader work.
In one area there have been big breakthroughs, however, which are helping to reshape the way many central bank policymakers think about the economy. The work of Massachusetts Institute of Technology economists Alberto Cavallo and Roberto Rigobon for the Billion Prices Project and PriceStats (its business end) gives central bankers access to high-frequency indicators of inflation – published daily, rather than monthly, and with a lag of three days, rather than several weeks. It gives central bankers much richer information to work with, and in the past two years, PriceStats has expanded its offering to include purchasing power parity indexes.
The work started out as a research project by Cavallo and Rigobon, both academics at MIT’s Sloan School of Management. In 2007, Cavallo began developing daily inflation statistics for a set of Latin American countries using ‘scraped’ online prices as part of his PhD thesis at Harvard. Rigobon joined a year later, and helped expand the roster of countries to 50, forming the Billion Prices Project.
The scale of the number crunching soon demanded a team of economists and technology specialists, and PriceStats came into being in 2010 as a means of funding the work, as well as a viable business in its own right, led by chief executive Pilar Iglesias. “From the beginning, the company was profitable, and that made things easy as an entrepreneur,” says Iglesias, who studied for her master’s degree in business administration alongside Cavallo, to whom she is now married. “The main challenge was to recruit talent with the right skills since the concept of big data was just starting to get recognition.”
The daily inflation data is sold through a partnership with Massachusetts-based financial services provider State Street to private clients across the financial system, as well as many central banks and other public bodies. Where researchers want access to long time series with less need for immediacy, the firm tends to give the data out for free, and it has worked on public-sector projects with bodies including the United Nations and the International Monetary Fund. Some data is also available to download via the Harvard/MIT database.
Cavallo and Rigobon remain focused on the research side. One recent paper sets out the benefits of the online approach to price-gathering, relative to more traditional methods.1 The cost is low and the frequency high, they note. The samples are much larger, representing all products sold online by the retailers they observe. The basket also updates automatically as new products enter and old ones are discontinued. The data is comparable across countries, and permits remote collection – which is useful in Argentina, for example, where, for several years, the authorities wanted to prevent independent data collection, they say.
The paper does acknowledge some of the drawbacks of their approach, as well. The prices currently cover a much smaller set of retailers and product categories than those captured by official methods. Prices for services, in particular, are often not available online. Furthermore, online datasets lack information on quantities sold – this could be captured by alternative data sources, such as scanner data, at least for goods such as groceries.
Another paper, authored by Cavallo, presents the results of a large experiment into how online and offline prices compare.2 The research involved hiring 323 workers through crowdsourcing platforms to scan barcodes and collect prices of goods that are available both on- and offline. The data covers several countries, focusing on the top 20 companies in each that sell both on- and offline. The effort yielded 38,000 observations.
“The main finding is that online and offline price levels are identical about 72% of the time, with significant heterogeneity at the country, sector and retailer level,” writes Cavallo. “These percentages range from 42% in Brazil to 91% in Canada and the UK.” The research also implies that price changes have similar frequencies and sizes in the online and offline data, although the changes are not synchronous most of the time.
Putting the data to work
The Central Bank of Argentina has been one major user of PriceStats data, which helped to fill a gap in 2017 when the widely mistrusted national statistics on inflation were rebooted. Pablo Neumeyer, chief economist at the central bank, says its economists have successfully developed a nowcasting model that draws on PriceStats data: “In the last two weeks of each month, this mixed-frequency estimator is the best predictor of inflation. This is especially valuable, as the official statistics have a two-week delay.”
PriceStats data comes with an additional benefit for Argentina because the prices in the index are not subject to controls. “Many prices – such as public transport, energy, health insurance, etc – are regulated by the government, and, hence, they do not paint a clear picture of inflation trends for free prices,” says Neumeyer.
PriceStats data comes with an additional benefit for Argentina because the prices in the index are not subject to controls
Recently, PriceStats has been focusing on rolling out its purchasing power parity indexes. “Our PPP indices are like a Big Mac Index but expanded to about 200 product categories within food, electronics and fuel,” says Iglesias. “Those indices help better understand whether a currency is expected to appreciate or depreciate.” She says the indicators have been “well-received” by clients so far: “The PPP portfolio represented a significant achievement because it required the development of new methodologies, technologies and processes that we hadn’t used before.”
As such, central banks can now assess their economies using high-frequency indicators of both domestic prices and relative prices across borders, helping them to determine how inflation and the exchange rate are likely to move day by day. PriceStats’ work has pioneered a new way for policymakers to think about their economies and study how policy changes propagate through the system. Although the data should – like all data – be used with caution, it seems to open up major new possibilities for the future of central banking.
1. Cavallo, Alberto and Roberto Rigobon, ‘The Billion Prices Project: using online prices for measurement and research’ in Journal of Economic Perspectives, Volume 30, Number 2, Spring 2016, pages 151–178.
2. Cavallo, Alberto, ‘Are online and offline prices similar? Evidence from large multi-channel retailers’ in American Economic Review, Volume 107, Number 1, January 2017, pages 283–303.
The Central Banking Awards were written by Christopher Jeffery, Daniel Hinge, Dan Hardie, Rachael King, Victor Mendez-Barreira, Iris Yeung, Joel Clark and Tristan Carlyle