The Economics Revolution Will Be Televised

There’s a revolution underway in economics. It’s not due to the financial crisis, but rather something more mundane: Data, and computing power. At least that’s the claim that Betsey Stevenson and I make in our latest Bloomberg View column:

“Consider the stream of data you will create today. Your metro card will record what time you caught the train. Your Web browser will note how you go about your job, and how much you procrastinate. A mid-afternoon purchase at Starbucks will reveal your penchant for lattes and the occasional cookie. Your flow of e-mail traffic will trace out your professional and personal networks.

At the same time, computing power has made it extremely easy and cheap to analyze all the data you produce. An economist with a laptop can, in a matter of seconds, do the kind of number crunching it used to take a roomful of Ph.D.’s weeks to achieve. Just a few decades ago, economists used punch cards to program data analysis for their empirical studies.”

Two weeks ago, Harvard’s Raj Chetty gave a spectacular talk at the National Bureau of Economic Research, about what he called “The Transformative Potential of Administrative Data.” He documented that today’s cutting-edge research is based on crunching newly-available data from the vast databases which underlay our schools, welfare state and tax systems.  I’m just as optimistic that new data coming online from the private sector will prove to be just as useful.

This empirical revolution is reshaping economics in at least four important ways:

  1. The role of economic theory is changing: 

    “The shift toward an even more empirically grounded economics doesn’t mean theory is less important. When facts were expensive and scarce, the role of theory was to “fill in” for missing data. Now, its purpose is to make sense of the vast, sprawling and unstructured terabytes on our hard drives.

  2. Empirical economics is a natural bedfellow for behavioral economics:

    “The data revolution is, however, changing our theories — specifically the way we choose to model how people behave. For decades, economists assumed that people made calculated, rational decisions. Without better data to help structure our understanding of people’s preferences, it was a safe and convenient choice, even if it was often wrong.  With new data on everything from how we choose our retirement savings plans to how NBA referees call fouls, we have learned to look beyond “homo economicus.” We have a much better grasp of the systematic flaws in reasoning that often get people into trouble. We know they have a hard time committing to do difficult things in the future — to go to the gym, to lose weight, to save. So we know people can benefit from policies, such as making 401(k) contributions automatic unless they opt out, that help them commit to good behavior.”

  3. Individual-level data meant that we can say more about individual differences:

    “In the mathematical models they build to help them understand the world, economists have also long made another peculiar assumption: that the behavior of an entire group of individuals — say, U.S. consumers — can be modeled as if it were a single “representative agent.” Today, we have much better data describing the decisions of individuals, and the power of our computers allows us to populate our models with millions of such people, rather than just one.”

  4. And a theme that will be familiar to readers of this blog: Economics has become a much broader social science.

    “Perhaps the broadest insight that has come with the explosion in data is the understanding of how economic reasoning suffuses almost every aspect of our lives. The economic lens can be very helpful in parsing strategic interactions, the causes of discrimination, patterns of marriage and divorce, and how our political machinery operates.”

The bottom line:

“Technological change has brought opportunities to do economics in a way that our predecessors could only have dreamed about. Those opportunities have, in turn, yielded a field that is more connected to reality. Our hope is that these insights will improve our understanding of the economy and give us a better shot at avoiding the next crisis.”

You can read the full column here.  And let me know in the comments if you think my optimism is misplaced.


The first challenging question is how to establish causality out of all the correlations picked up by data mining. Second, the new data comes in formats (for example, my walking route) that are not easy to analyze or to match with all other data that I generate during my day. Third, the massive amount of data prohibits quick and simple analyses and it is also very costly to store. Fourth, the thesis of this article is hard to defend. It would be certainly challenging to prevent a macroeconomic crisis by using micro-analysis of millions of consumers and businesses at the individual level. Economic agents are heterogeneous and they respond to government policies in myriads of ways. Still, their aggregate and synchronized behavior is what matters most for policy making. In an ideal world, all government policies (from subsidies to taxes to healthcare) could be customized at the individual level because every one of us reacts differently to economic incentives. For this customization to become real we do not need more data, we just need relevant data. Unfortunately, many are running this frenzy of collecting data without much critical thinking about what data is really needed and for what purpose.


Milton Recht

Both the Ptolemaic geocentric and the Copernican heliocentric views of planetary orbits were data driven. Both were highly accurate. Ptolemy's system was used for over 1500 years, but the Copernican circular (elliptical) system was simpler and easier to calculate than the Ptolemaic epicycle system and eventually won out.

If modern computers were around in the 16th century, we probably would not have switched away from the Earth-centric view of the Universe to the Sun-centric view of our solar sytem. We would still be using Ptolemy's system today and making accurate orbital predictions.

Newton's view of gravity and planetary orbits and the early conceptualization of electrons and other sub-atomic particles require thinking in terms of circular orbits instead of epicycle orbits. Would these theoretical advances occur under an epicycle system?

A facility to do complex data calculations may remove the impetus to simplify theories, which may impede further theoretical advances.



Just curious if the author, or perhaps headline writer, knows the origin of the quote?


maybe not optimistic enough- maybe ai robots will become better economic predictors and supplant human economists that missed the housing bubble


Seems as though there's a problem with the underlying assumptions about the accuracy & completeness of the data. It's pretty easy to hack or spoof most of it: I could, for instance, easily create a browser-like app that would do spoof work; I don't think it'd be that much more difficult to sanitize any outgoing traffic.


I think the end result will be that economics is merely not as far behind. The economy is not a physical thing, but an emergent property of all the exchanges between members of a society. What exchanges the members make are strongly influenced by the society they're a part of, what it values, its norms, the rules and regulations placed on it. That said, there is a place for empirical economics in finding out what the rules are now but as far as prediction, I believe we'll just make bad predictions faster.


I think in all the discussion and enthusiasm about big data most people forget, that a theory is not only there because we need to fill a data gap, but to explain the world. If we only relay on data mining, we forget that correlation is not casualty. A very good example is the one of the correlation of stork population and birth rate. An example that probably everybody knows from first semester statistic. Even if we would have all the data in the world, we still would need a theory. A theory is the rule set with which we can analyse the data.
Another point that is often forgotten, is the fact that the personalized data has a massive bias. The sources where the data comes from does not represent the population in the same proportions as they occur in reality. That means that the results can not be generalized as easy as it seams to be. Just because we have a huge amount of data doesn't mean we can do something with it that is meaningful. Not to talk about the ethical issues that accure with the use of such data.



This is one of my favourite topics, opening up whole new worlds of analysis.

As far as I can tell, I coined the phrase "Digital Density" as shorthand in a presentation in 2004.

The example I use is telephony.

Originally on the phone bill we only had a fixed fee. Then we had total number of calls. Much later we started getting records of what numbers were connected (but not dialled), eventually with time of call attached. Finally we move to location data. This is an increase in density of information surrounding what is fundamentally the same activity.