Computational problem solving advances are opening new paths for economists to further understand an increasingly complex economy

Every decision we make, every subset of information we digest and process — from shopping for shoes to seeking out meaningful news to hailing a cab — is driven in some part by an unseen algorithmic hand. While underlying influences on our decisions are not new, the nature of the data and the speed at which it can be accessed are changing how consumers and businesses behave and how markets actually function. Similarly, for academic economists, these computational advances and access to new data afford exciting research opportunities, allowing researchers to address new substantive problems in a computationally rich economic environment and to revisit existing problems in more ambitious ways.

According to Lars Peter Hansen, BFI Director, “Achieving these aims will require rethinking and altering some of the analytical frameworks used by economists. To fully realize the expanded opportunities facilitated by computational advances, the BFI is bringing together experts in computation and data science with experts in economic analysis to engage in productive exchanges of ideas along these two different fronts.”

For Mohammad Akbarpour, embracing computation in his study of marketplaces is a matter of necessity—the very nature of markets is becoming more complicated. “Resource allocation problems are becoming increasingly complex, and we cannot solve optimization problems for them in a reasonable amount of time,” says Akbarpour, institute research fellow and co-organizer of an August conference on the intersection of computer science and economic theory.

Modeling approaches of previous decades continue to be useful as thought experiments—frameworks through which we can understand the variables in play within a given market. But especially as these markets expand in scope and complexity, economists have to make a lot of assumptions to optimize models of how they work, then reflect on how their assumptions affect what conclusions they can draw from them. Moreover, if economists want to design more efficient marketplaces, they have to untangle those assumptions in order to really see what variables can truly make a difference. Simple models of 30 years ago are useful, and, of course, insightful, says Akbarpour, but they are not enough for designing new marketplaces.

Models with complex constraints allow more robust optimization of the problem: finding the feature of the model or auction that makes a real difference, and being able to explain it in real terms.

Computational tools might help us reduce the assumptions we make in considering not only classic macroeconomic questions, but also as “economists as engineers” strike into new frontiers for economics. Akbarpour’s work focuses on markets that involve choosing, as well as being chosen—specifically, kidney transplant markets, which involve a complex web of timing and donor-recipient compatibility. He says that, conventionally, economists would ignore the network graph that connects these donors and recipients in a complex web, solving a simplified model exactly. Computer science theory tools such as graph theory and algorithms present alternative options: rather than exactly optimizing a simplified model, a scholar can “approximate” the optimum of a more realistic model. Which approach is superior, of course, depends on the specific economic problem in hand. Graph theory and similar models with complex constraints allow more robust optimization of the problem: finding the feature of the model or auction that makes a real difference, and being able to explain it in real terms.

Ben Brooks, co-organizer of the August conference on computation and economic theory, agrees that there are obvious complementarities between computation and economic theory. “There may be institutions which have good properties in terms of incentives but are impossible to implement in a reasonable amount of time and with a practical amount of computational power, and the tools of computer science can be used to better understand and incorporate those constraints into economic modeling,” writes Brooks. “On the flip side, computer scientists may propose technological solutions that are efficient but only if one ignores the self-interested behavior of economic agents, and economics has tools for understanding human decision making and strategic behavior.” Each field can benefit from the tools that the other has developed, argues Brooks.

“What are useful notions of “complexity” for understanding human decision making and institutional design?” asks Brooks.

Brooks says that this and other open philosophical questions provide both computer scientists and economists a common jumping off point for conversation and debate. “What are useful notions of “complexity” for understanding human decision making and institutional design?” asks Brooks. “And how should we model the bounded rationality of real-world decision makers, who face similar constraints as computer systems in their ability to perform complex computations and process large amounts of information?”

The August conference is part of an ongoing commitment from the institute to deepen the collaborative spirit between economists and computer scientists here at the University of Chicago, says Hansen. In addition to the August conference on economic theory and computation and a September conference on machine learning, the institute will provide resources to aid scholars seeking to expand the use of computation within their own work, via experienced computational economists like Richard Evans, who will lend his expertise both to young students learning to work together using the improved collaborative tools and established scholars looking to integrate new approaches into classic economic problems.

In new ways of thinking through, collaborating on and tackling problems, the next generation of economists may find far-reaching new approaches to old problems at the same time that new avenues of inquiry are emerging surrounding the study of markets in a computationally rich environment. Computational advances are not by themselves a panacea, but these cross-disciplinary conversations are important to have now as economists discover what works and what doesn’t. With a slate of conferences ahead, the institute is helping jumpstart the conversation in 2016 and beyond.

*—Mark Riechers*