Speech

Data in the Digital Age

November 17, 2017

Good afternoon. Thank you so much for joining us this year for the fifth annual IMF Statistical Forum. My special thanks to all our speakers.

It has fallen to me to try to summarize and comment on your discussions in just a few minutes. I’ll do my best.

I know Moore’s Law was mentioned this morning, and that got me thinking. In the 90s, we marveled at the rapid succession of technical advances with the microprocessors that powered our computers.

Now, we have essentially reached the point extrapolated from Moore’s Law where the physical limits of the laws of nature have begun to constrain aspects of chip development predicted by Moore’s Law.

Yet our digital experience long ago moved beyond those limits. We seem at times to be experiencing infinite possibilities: cloud computing, big data, the internet of things, Amazon amassing $135 billion in revenue last year… and the wonder of 140, and now 280 character tweets. Most importantly, the notion embedded in Moore’s Law that computers can outpace human linear thinking with their ability to process exponentially remains essential to the digital revolution.

As we’ve been discussing, the emergence of the digital economy has presented many new challenges for both economic and financial statistics. We first saw this in the 1990s, when improved price measurement methodology and compilation practices enabled us to measure the acceleration of productivity spurred by the initial phase of the IT revolution.

New Statistical Challenges

But lightning-fast change and growth mean that the next statistical challenge is constantly upon us. Third-party mobile payments are growing toward $9 trillion dollars a year in China. In Africa, mobile banking services like M-Pesa are providing banking to rural and urban populations that until recently were unserved. You addressed the implications of these changes yesterday.

Now we are witnessing a debate about the impact of the “gig” economy, and the sharing economy, especially in terms of measuring non-market production in households. It wasn’t so many decades ago that household production in this country centered on weaving and farming.

Statisticians face the challenge of updating compilation practices and developing new forms of data.

The research presented at this forum offered noteworthy perspectives on this challenge. So, allow me to take a moment to summarize some of our conclusions.

First, online platforms, and the businesses they enable, need to be promptly included in the source data for economic statistics. This will require effective data-sharing arrangements between statistical agencies and other branches of government that produce administrative data, such as tax authorities and regulators.

Second, statistical agencies need to provide more data on the digital economy for national accounts and trade statistics.

Third, the international classification systems for industries and products must be updated to guide the incorporation of source data on the digital economy and the development of new statistics.

Fourth, there needs to be a renewed push to update and maintain quality adjustment procedures of price statistics for ICT-related goods and services. This encompasses a range of services such as mobile telephony, broadband internet, streaming videos, international calling and messaging, and even time spent online. Right now, it is hard to determine whether the 5 million smartphone apps available are reflected in volume indexes for software.

Fifth, GDP and productivity statistics have been criticized for omitting consumption of free services from digital platforms, and for neglecting household production that uses digital technologies. Better communication is necessary between compilers and users of statistics. This also suggests a need for data on the welfare effects of digitalization.

A Role for Developing Countries

Sixth, developing countries must be part of the research on these welfare effects. Observation in places like East Africa tells us that the impact may be greater where most people had no phones or access to financial services before the advent of mobile phones and mobile money. We need to be able to measure this. The implications for inequality are important.

Seventh, digitalization does not just mean new measurement challenges and data demands. It also means new statistical opportunities. For example, with data-reporting requirements for e-money transactions, including cross-border payments, we can shed new light on the unobserved economy.

And finally, we must always keep in mind that measuring the digital economy and capturing these new opportunities will require resources and collaboration. We will need budget and human capital resources well beyond what is now available in the “old economy.”

Having said that, I’d like to offer an observation on new statistical opportunities. Digitalization is not just a matter of new measurements. It’s not just a matter of making statistics more accurate and timely.

More fundamentally, it is about using Big Data to explore numbers in ways that have never been possible before. We now have the ability to crunch data sets that previously were beyond our technical or temporal capabilities. That means being able to learn more about economic relationships and the effectiveness of policies.

Let me give you an example. A few weeks ago, Professor Raj Chetty of Stanford gave a presentation here at the Fund about the work he and colleagues are doing on economic opportunity and social mobility. He described how they are working with vast data bases drawn from the Census Bureau and Social Security Administration to track the income trends of millions of people over 20-30 years.

They have been able to quantify how moving from one neighborhood to another affects the ability to escape poverty, and how the quality of a third-grade teacher can influence lifetime earnings.

New Sources of Information

This type of work may soon dwarf what statistical agencies do with the data collected from their surveys—as important as those are. The new reality is that there are troves of information held by governments, banks, internet companies and others—some of it available, some of it so far beyond our reach. The use of this information can transform how we come to understand economics and markets.

This has implications for the IMF. Our economists and statisticians do superb work with the data they have while conducting Fund surveillance and offering policy advice. Our Statistics Department makes an essential—and largely unsung—contribution to governments around the world by helping them develop their data compilation and analytical capabilities.

But there is so much information now available that we are not tapping with the computing power at our fingertips. The Fund needs to give serious thought to the ways in which Big Data will transform our work in the future. And I’d imagine that right now there are university and high school students around the world who will arrive here in the coming years to help us do just that.

One final point on measuring GDP. As you have been discussing at this conference, I agree that we are not reflecting important trends in the digital economy in our statistical work. For example, the way that we differentiate prices changes and quality changes in the computer industry. When a more powerful device costs the same as the previous generation—and I’m not talking about the iPhone X—is that quality upgrade adequately reflected in GDP numbers?

This issue also plays out at the level of metadata amassed by online retailers—in a way that we will have to come to grips with going forward. Every time you buy from Amazon you engage in two transactions—one monetary and one essentially a barter of services in the form of information. Every link you click on, every purchase you make, provides a wealth of information that is magnified across billions of transactions in the global marketplace.

How are we valuing these transactions? As the cost of a book or a pair of shoes? Or something much more important to the economy.

This conference has discussed the issue of measuring free internet services to consumers. The question of information-provided-for-free may represent a more significant element of value-added that is being translated into revenue by the extrapolating power of massed servers for future work.

It is one of many issues that I am sure we will be grappling with at future statistical forums. I look forward to those discussions.

So, on that note, let me thank you all once again for your contributions to this year’s conference.