Augmenting Economics with AI

Assistant Professor of Economics Ekaterina Seregina has developed an augmented method of economic modeling using AI

Ekaterina Seregina
Assistant Professor of Economics Ekaterina Seregina
By Linzy Rosen '22Photography by Tristan Spinski
June 21, 2022

Distinct but not detached.

That’s how Assistant Professor of Economics Ekaterina “Kate” Seregina thinks of the 2008 recession, the coronavirus, and the ongoing war in Ukraine. Those three events, while separate, are inextricably linked. They all impacted the global economy, and Seregina wants to compare how those impacts differ.

But existing modeling techniques are limited. Utilizing artificial intelligence, Seregina has built her own version of traditional analytical and modeling tools, unlocking the ways that economists can more accurately analyze complex relationships.

Seregina’s work at Colby is an example of how artificial intelligence and machine learning are being incorporated across the economics curriculum, said Amanda Stent, inaugural director of Colby’s Davis Institute for Artificial Intelligence. She also cited the work of Herbert E. Wadsworth 1892 Professor of Economics Michael Donihue, Associate Professor of Economics Dan LaFave, Visiting Assistant Professor of Economics Ben Scharadin, and others.

“Their work to incorporate techniques from and discussions about machine learning and AI across the economics curriculum will help students acquire a more critical understanding and deeper ability to model data for decision-making,” Stent said, noting that economics students have approached the Davis Institute to talk about AI more than any other major. “Through my experience in industry, I have seen firsthand the relevance of machine learning to finance, economics, and public policy.”

Originally from Russia, Seregina joined Colby in 2020 with a background in statistics and machine learning from her advanced econometrics Ph.D. program at the University of California, Riverside. In pursuing econometrics—a blend of economics and statistics—Seregina found herself using graphical models for complex datasets.

A graphical model is a type of optimization tool used across disciplines, from biology to mathematics. These models help to demonstrate that behavior is best predicted not by examining drives, attitudes, or demographic characteristics, but instead by looking at complex webs of connections. In economics, a graphical model can help estimate how different entities, such as financial institutions in the global economy, may affect each other during and following global or regional events. But such models are limited for economists, Seregina said, making them less widely used, despite their capacity to model large datasets.

Current graphical models can’t account for how connections between individual entities, such as various banks, change over time, Seregina explained, as the economy is not static. Traditional models also can’t distinguish between economic fluctuations that only apply to certain companies or industries, as opposed to all of them. 

“Through my experience in industry, I have seen firsthand the relevance of machine learning to finance, economics, and public policy.”

Amanda Stent, Director of the Davis Institute for Artificial Intelligence

To address these flaws, Seregina is developing a new approach. Her proposed Factor Graphical model integrates graphical modeling with the factor structure to help account for common mistakes in forecasting and the relationships between entities. Taking into account these aspects incorporates AI and machine learning techniques.

Seregina is currently employing her augmented model to study the implications of the recession of 2008, the coronavirus, and the war in Ukraine on the macroeconomy and, separately, on corporate portfolios. The augmented model can better illustrate the impacts of the pandemic, for example, on relationships among firms, which is helpful for macroeconomic forecasts and shareholder decisions. For the latter, the information derived from these relationships can be used to detect stocks or portfolios most sensitive to changes and help investors select mutual funds that may perform the best. 

Ultimately, spreading more awareness of the power of graphical tools—and her augmentation— is key. Seregina was recently invited to write a chapter on graphical models for the forthcoming book Econometrics with Machine Learning published in Springer’s Advanced Studies in Theoretical and Applied Econometrics series. This work is aimed at helping theoretical and applied econometricians as well as graduate students promote the use of machine learning in their scholarship.

In the fall, Seregina will teach a new senior seminar on financial technology, the first of its kind at Colby. Using a grant from the Davis Institute, she developed a curriculum that includes topics such as cryptocurrency, blockchain, machine learning, and AI for finance. 

“Most schools don’t offer these courses in person,” Seregina added. “[They’re] usually offered online, as part of an M.B.A. program. So I think it’s really unique that we get a chance to offer something like that.”

For Seregina, developing improved graphical models is only the beginning. Drawn to Colby for its close-knit learning community, she aspires to mentor and bring students to the forefront of her work.

In her first year at Colby, Seregina hired Yaoyao “Chris” Zhu ’22 as a research assistant and has plans to hire more. “I learned a lot about applying neural network models to real-life data, which will benefit me with my career in economics consulting,” said Zhu, a computer science and economics double major. “In the future, I hope to use my machine learning and AI skills in finance to create innovative approaches to financial modeling.” 

Sign up to read the latest each week.