The Center for Data Innovation spoke with Giselle Guzman, founder and chief executive officer of Now-Cast Data Corporation, an economic analytics company based in New York. Guzman discussed how economists only recently could have the ability to study the economy in real time, as well as how real-time economic insights could help bring about a more stable, peaceful world.
Joshua New: You have worked with Nobel Prize-winning economists Lawrence Klein and Joseph Stiglitz, whom you have said have always expressed the desire to monitor the economy in real time, rather than rely on quarterly or yearly government data, but could not because there was no technology powerful enough to do so. What has changed to allow Now-Cast to do just that?
Giselle Guzman: This project has been in the making for many years. Building a real-time system of the economy was something that my mentor, the late Lawrence Klein, spoke of constantly.
When I started building the Now-Cast system in 2010, the technology wasn’t really available to do it. We laid the foundation at that time in anticipation of the technology developing in that direction, because we could kind of see it looming on the horizon, even though it wasn’t there yet. When we launched our alpha product in 2012, the best the state of the art would allow was weekly forecasting. It wasn’t until 2014 that all the pieces finally came together—the data, the technology, and the talent—to be able to do it in true real-time. It was just a function of the way the Big Data landscape evolved from 2010 to 2015. There have been big changes in data and technology from 2010 until now, and the landscape continues to evolve rapidly.
New: Your team is predicting four thousand different economic indicators. How did you identify all of these? Or, if these are indicators that everyone is privy to, how did you manage to get them in real time when nobody else can?
Guzman: The four thousand indicators of U.S. macroeconomic and U.S. regional economic activity we work with at Now-Cast are the main economic indicators covering prices, population and labor markets, national accounts, money and banking, and production and business activity. These are the main indicators measuring U.S. economic activity that investors and policymakers look at. But we’re working on incorporating even more indicators into our forecasting, so stay tuned!
Getting Now-Cast in real-time was a matter of persistence and creativity. We had to be singularly focused and very open-minded about how to accomplish the mission of achieving truly real-time, minute-by-minute estimates of economic activity. We were willing to experiment with different techniques and borrow methods from other scientific disciplines, mainly computer science. We ended up building a system that fuses economics together with computer science, artificial intelligence, machine learning, data-mining, and statistics. So, we broke a lot of rules and did a lot of things that traditional economists would never do, but it works.
New: Relatedly, you’ve said that “data mining” is a dirty word to a lot of economists. Why is that?
Guzman: Economics has traditionally been an a priori science, which means theory is antecedent to experience. This is fundamentally different from the natural sciences where observation or laboratory experiments are crucial in the development of theory. The natural sciences are empirical sciences, relying on data, which means they are a posteriori as opposed to a priori.
Data-mining is about knowledge discovery, where we explore data in search of patterns and relationships between variables. Data mining, which the Now-Cast system embraces, lets the data “do the talking,” and this is at odds with the a priori nature of traditional economics. The Now-Cast system strikes a careful balance by being on the one hand a properly interrelated system of the economy that respects basic economic theory, but on the other hand allows for knowledge discovery through data mining. So the Now-Cast system is both a priori and a posteriori, which is at odds with traditional economics.
New: Are there any accuracy tradeoffs in your approach? Just how accurate is this real-time forecasting?
Guzman: Now-Cast is very accurate. The Now-Cast system uses machine intelligence to scan billions of data points of live web activity for economic risk signals and calculates, on the fly, their impact on the economy. This means that the system can react to new information faster than humans possibly could. Because the Now-Cast system is 100 percent algorithmic, it is free of certain biases that may influence a human economist’s judgment. Moreover, the Now-Cast system has sophisticated learning algorithms built into its econometric framework. This means that forecast errors are reduced over time—they asymptote to zero, in fact. In the battle of man versus machine, machine wins in this case.
New: When we talked previously, you described how someone once half-jokingly told you that what Now-Cast is doing could help bring about world peace. Is there any truth to that? What are the far-reaching implications of real-time economic insights?
Guzman: I did think the suggestion that Now-Cast could help bring about world peace was a bit far-fetched when it was first suggested to me, but after thinking about it, I could see the point. Now-Cast provides economic transparency by showing what the state of the economy is on a minute-by-minute basis. It is a continuous data stream of estimates that is extremely accurate. Transparent views of the economy can minimize surprises and data shocks, help people make better decisions, help lead to better economic policy, and therefore lead to more economic stability, which is a crucial prerequisite for a stable and peaceful society. In this sense, the real-time economic insights Now-Cast provides can improve social welfare, and hopefully help bring about world peace. It might sound a bit corny, but it isn’t that far-fetched.