Publications

Machine learning for asset management faces a unique set of challenges that differ markedly from other domains where machine learning has excelled. Understanding these differences is critical for developing impactful approaches and realistic expectations for machine learning in asset management. We discuss a variety of beneficial use cases and potential pitfalls, and emphasize the importance of economic theory and human expertise for achieving success through financial machine learning.

From 2011 to 2019, the Connecticut Department of Social Services (DSS), under the direction of Roderick Bremby, managed to transform its $570 million food stamp program from one of the worst in the nation to one of the best. Having achieved such a remarkable turnaround, observers wondered what else the DSS might do to further deliver on its mission of "providing person-centered programs and services to enhance the well-being of individuals, families and communities."

Connecticut's food stamp program became overwhelmed when applications for assistance surged in the wake of the 2008 recession. At the same time, reduced state revenues forced cuts in the number of employees to process the paperwork. Bremby took over a demoralized program with antiquated technology, siloed staff and inefficient work flow. Over the next eight years, the department took a number of steps to improve the speed and accuracy of issuing benefits. The incremental moves improved matters fitfully, but Bremby and his staff persisted. In 2018, Connecticut SNAP was recognized by Washington as best in class. Federal officials awarded the state several million dollars in bonuses for the remarkable improvement.

Yet workers in Connecticut's social services agency were ambivalent about the changes. To a person, workers were pleased that they no longer faced mountains of unprocessed claims on their desks and hours of unreturned telephone messages on their answering machines. Nonetheless, many wondered if a vital human touch had been lost. They believed they had fewer opportunities to use their social work skills to safeguard clients’ welfare and build bridges to better lives. While employees were proud the Connecticut SNAP had embraced a more efficient business model, they worried that the DSS had become more like an airline reservation system than a social service agency.

In 2019, Bremby left state government for the private sector, providing a moment to wonder how SNAP's gains in efficiency translated to gains in effectiveness in alleviating the burdens of poverty. Having made improvements on the metrics that mattered to the federal government, could the DSS leverage its success with the SNAP program to improve the lives of recipients in other ways? Were there other avenues of service delivery the department could pursue?

What to do about Covid? With nearly 60 million cases and 1.4 million deaths worldwide as of the end of November 2020, there are still no consistently effective treatments or approved vaccines yet (though large-scale vaccine trials have already produced promising results). Social distancing, mask wearing, and infection control practices can reduce the rate of spread somewhat, but as long as infectious individuals circulate amongst susceptible persons, continued spread is inevitable, given that most populations have not built immunity against SARS-CoV-2 to any meaningful extent.

We examine how operational or technological transformation impacts consumer value, as well as the effectiveness of a firm’s pricing strategies. We develop a model of multidimensional screening featuring forward-looking consumers who make short-run consumption and long-run purchase decisions. Using a detailed panel of consumer data from a rental-by-mail firm, we estimate consumer utility for current consumption, obtaining heterogeneous preferences for bunching and smoothing consumption. Using counterfactual analysis, we evaluate the impact of improving service time. We find that the firm with improved service time might create more value for all consumers, but its profits and even revenues could diminish because value extraction becomes more difficult. We find a novel mechanism that causes this effect, which is driven by increased consumer heterogeneity in the valuation for each product and reduced differentiation across products. This result persists even when the firm can reoptimize its price levels based on the service time. We find that a change in the pricing strategy might be required for the firm to obtain higher revenue with improved service time.

The Main Street Lending Program (MSLP) was established by the Federal Reserve to provide loans to small and mid-sized firms and large below-investment-grade firms that were financially sound before the onset of the COVID-19 pandemic. The program, which was established under the Fed’s Section 13(3) emergency authorities, is supported by capital from the U.S. Treasury and became operational in July 2020; however, utilization has been slight. We describe the economic challenges in designing a loan support program and evaluate the MSLP program in terms of how it manages significant asymmetric information, adverse selection, poor targeting, and moral hazard problems while protecting taxpayer funds. We contrast the MSLP with other possible approaches, such as subsidies or loan guarantees. We conclude by recommending changes to the program to increase its usage and effectiveness.

Motivated by viral marketing on social networks, we study the diffusion process of a new product on a network where each agent is connected to a random subset of others. The number of contacts (i.e., degree) varies across agents, and the firm knows the degree of each agent. Further, the firm can seed a fraction of the population and incurs a fixed cost per contact. Under any bounded degree distribution and for any target adoption proportion, we compute both the cost and the time it takes to reach the target in the limit of network size. Our characterization of the diffusion process in such a general setting is the first of its kind for a problem that is generally deemed intractable and solved using approximation methods such as mean-field. Our solution indicates that the degree distribution impacts the diffusion process even beyond its first and second moments. Using our limit results, we conduct comparative statics on degree distribution and uncover a trade-off between cost-efficiency and fast growth. Fixing the average degree, a minimum-variance degree distribution incurs the minimum cost to reach any adoption proportion. On the other hand, higher variance results in faster growth for low or moderately high target adoption proportions, but it incurs higher cost. This trade-off arises partly because of an endogenous effect of diffusion on the distribution of adopters’ neighbors: as the diffusion progresses, adopters become more likely to be connected to other adopters. This highlights the benefit of our exact analysis compared to mean-field approximation methods that rely on perfect mixing of adopters and non-adopters (i.e., there is no change in the distribution of adopters’ neighbors with the progress of diffusion). Further, we study the impact of the degree distribution on optimal seeding strategies for a given seeding budget. Somewhat surprisingly, we show that to minimize cost, it is optimal to seed low-degree agents. Even if the objective is to minimize time, for certain regimes, the optimal seeding strategy is a mixture of low- and high-degree agents.