Seminars 2022-2023

In this talk, I will present a series of theoretical and applied work in competitive pricing. A retailer following a competition-based dynamic-pricing strategy tracks competitors’ price changes and then must decide whether and how to respond. The answers require modeling of consumer decisions, unbiased measures of self- and cross-price elasticity as well as competitor impacts. I will discuss how we achieve them through a combination of consumer modeling, experimentation, causal inference, as well as high-dimensional statistics. I will highlight two implementations that each led to 10-20 percent revenue and profit improvement through close collaborations with leading US and international retailers, and a theoretical development that scales up choice models through high dimensional regularization.

Fulfillment by Amazon (FBA) is Amazon’s third party marketplace for e-commerce sellers. In the past decade, FBA has grown rapidly — growth that has only accelerated during the pandemic. By most measures, FBA is now larger than Amazon’s own retail business. This has created a significant challenge for capacity management, since third party sellers have considerable autonomy over how much inventory they send to Amazon. In this talk, we provide an overview of the FBA capacity management problem and describe an innovative market-based mechanism (auction) for capacity recently launched in North America. The auction uses a novel combination of securities and ex-post inspection. It admits a simple bidding strategy, implicitly scores sellers on a combination of sales and profit, and extracts minimal (often zero) revenue from sellers.

Supply chain risks, despite being a well-developed theoretical concept, are difficult to quantify, limiting the scope of empirical research. I develop a firm-level measure of supply chain risk exposure from a novel source of unstructured data—managers’ discussions of supply chain-related topics during earnings conference calls and Q&A sessions—using textual analysis techniques including seeded word embedding and bag-of-words-based content analysis. I validate the measure by showing that (1) it exhibits intuitive variations over time and across firms, and successfully captures both systematic supply chain risk events and supply risks in firms’ routine operations; (2) the measure indeed captures exposure to risks as it significantly correlates with contemporaneous and next-quarter stock return volatility, and (3) this relation remains significant after controlling for well-known aggregate risk measures such as the VIX and economic uncertainty indices. I then conduct a series of tests to demonstrate that the measure is indicative of risk that is specifically related to supply chains. (4) Consistent with theoretical predictions, firms facing higher supply chain risks have higher inventory buffers, with the majority of the inventory buildup concentrated in raw materials and intermediate inputs. High-risk firms also significantly increase their cash holding in lieu of investments and receive significantly lower trade credit from suppliers. (5) A placebo measure that only tabulates risk words does not correlate with any supply chain risk buffers. Moreover, (6) during well-known risk episodes such as the Tohoku earthquake, firms with higher ex-ante risk exposure have worse performance. These results indicate that the text-based measure provides a credible quantification of firm-level exposure to supply chain risks, and can thus be reliably utilized as outcome or explanatory variables in empirical supply chain research.

The use of artificial intelligence (AI) in healthcare is rapidly increasing, with over 500 medical AI systems having received FDA approval by 2022. While AI is unlikely to replace physicians, it is becoming plausible that physicians who use AI will replace those who don’t. Integrating AI into healthcare delivery is not a simple task and requires rethinking current workflows. In this talk, I will discuss ongoing efforts to understand and overcome barriers to the appropriate use of AI in clinical practice. Additionally, I will discuss the potential for AI to improve clinician productivity, increase access to care, and reduce disparities through studies conducted both in the United States and in developing countries. I will propose a research agenda for business scholars to consider as we continue to develop principles for the purposeful design of AI-augmented healthcare systems.

Which companies should we invest in? Who should we hire or promote? Which products should we launch? To make effective decisions, managers need to learn how different cues (e.g., the features of a product) relate to desirable business outcomes (e.g., its performance in a target market). Learning these contingencies is difficult, as many exogenous factors (e.g., the competitors’ actions, the general trend of the economy…) affect outcomes independently of the managers’ actions. Oddly, decision-makers typically report a high degree of confidence in their ability to understand business cues and drive favorable outcomes.

We highlight a possible factor driving this contradiction: some exogenous factors can paradoxically foster an illusion of learning. Specifically, we discuss the role of positive market trends. From the observation that decision-makers tend to explore sets of cues that they believe will produce favorable outcomes (a “positive test strategy”), we hypothesized that positive exogenous trends would reinforce the participants’ beliefs, regardless of their accuracy. We present a simulated model of contingency learning, and the results of five behavioral experiments, in support of this hypothesis. We conclude by discussing connections between our findings and existing research on overconfidence and illusory learning.

We study the interplay between inventory replenishment policies and information sharing in the context of a two-tier supply chain with a single supplier and a single retailer serving an iid Gaussian market demand. We investigate how the retailer’s inventory policy impacts the supply chain’s cumulative expected long-term average inventory costs C in two extreme information sharing cases: (a) full information sharing and (b) no information sharing. To find the retailer’s inventory policy that minimizes C, we formulate an infinite-dimensional optimization problem whose decision variables are the MA(infinity) coefficients that characterize a stationary ordering policy. Under full information sharing, the optimization problem admits a simple solution and the optimal policy is given by an MA(1) process. On the other hand, to solve the optimization problem under no information sharing, we reformulate the optimization from its time domain formulation to an equivalent z-transform formulation in which the decision variables correspond to elements of the Hardy space H2. This alternative representation allows us to use a number of results from H2 theory to compute the optimal value of C and characterize a sequence of epsilon-optimal inventory policies under some mild technical conditions.  By comparing the optimal solution under full information sharing and no information sharing we derive a number of important practical takeaways. For instance, we show that there is value in information sharing if and only if the retailer’s optimal policy under full information sharing is not invertible with respect to the sequence of demand shocks. Furthermore, we derive a fundamental mathematical identity that reveals the value of information sharing by exploiting the canonical Smirnov-Beurling inner-outer factorization of the retailer’s orders when viewed as an element of H2. We also show that the value of information sharing can grow unboundedly when the cumulative supply chain costs are dominated by the supplier’s inventory costs.

This talk presents evidence from two field experiments to shed light on the effects of non-monetary incentives and how they interact with the social relational context. The settings are Wikipedia and healthcare. In one case, the non-monetary incentive had positive and long-lasting effects. In the other case, the non-monetary incentive backfired, significantly reducing people’s well-being. The talk presents an overarching theory to reconcile these findings. It lays out a framework and an agenda for future experiments on incentives and social relationships, with implications for incentive and organizational design.

A carbon offset represents one unit of reduction in greenhouse gas emissions that can be used to compensate for emissions that occur elsewhere. Companies can purchase voluntary carbon offsets under two delivery terms. The first is prompt delivery, where a seller first undertakes the investment, then yield uncertainty realizes, based on which the seller determines the price for carbon offsets as a recourse action. The second is forward delivery, where a buyer orders a certain quantity of offsets before the seller invests. Therefore, there is quantity risk due to yield uncertainty because the seller may generate fewer offsets than the buyer’s order. Motivated by the importance of choosing the right delivery term for buyers, in this paper, we investigate the economic and environmental implications of these two delivery terms. Our results offer several managerial insights: A buyer should choose prompt delivery for projects with low or high investment costs, but forward delivery for projects with an intermediate investment cost. The delivery term preferred by the buyer typically leads to a higher quantity of delivered offsets, enabling the buyer to offset more of its emissions. However, the buyer may prefer a delivery term that leads to a lower investment by the seller, and thus a lower overall emissions reduction. Accordingly, environmental groups’ preference between delivery terms may not be aligned with that of a buyer.

Professor Mullainathan’ current research uses machine learning to understand complex problems in human behavior, social policy, and especially medicine, where computational techniques have the potential to uncover biomedical insights from large-scale health data. He recently co-authored Scarcity: “Why Having too Little Means so Much and writes regularly for the New York Times”. Additionally, his research has appeared in a variety of publications including the Quarterly Journal of Economics, Science, American Economic Review, Psychological Science, the British Medical Journal, and Management Science.

Professor Hemant K. Bhargava is an academic leader in economic modeling and analysis of technology-based business and markets. His research focuses on decision analytics and how the distinctive characteristics of technology goods influences specific elements of operations, marketing, and competitive strategy, and the implications it holds for competitive markets and technology-related policy.  He has examined deeply these issues in specific industries including platform businesses, information and telecommunications industries, healthcare, media and entertainment, and electric vehicles. Bhargava has published extensively in the top journals Management Science, Operations Research, Marketing Science, Journal of Marketing Research, Information Systems Research, INFORMS Journal on Computing, and Production and Operations Management.

Antonio (Toni) Moreno is the Sicupira Family Associate Professor in the Technology and Operations Management Unit. He teaches courses related to technology and operations management in the MBA, executive, and doctoral programs. Before joining HBS, he was an associate professor in the Kellogg School of Management.

Professor Moreno’s research studies the digital transformation of operations, and he is particularly interested in the transformation of retail and service industries. Professor Moreno’s work has appeared in journals such as Management Science, Marketing Science, Manufacturing & Service Operations Management, Information Systems Research, and Sloan Management Review, and has been covered by several media outlets. He received the Wickham Skinner Early-Career Research Accomplishments Award from the Production and Operations Management Society.

Professor Moreno earned degrees in electrical engineering and industrial engineering from Technical University of Catalonia. He has an MA in statistics and a PhD in operations and information management from the Wharton School of the University of Pennsylvania.

Professor Gupta’s research focuses on data-driven decision-making and optimization, particularly in settings where data are scarce. Such settings are common in applications that rely on personalization (like precision healthcare) and real-time decision-making (like risk management). Consequently, his research spans a wide variety of areas including revenue management, education, healthcare, and artificial intelligence. Professor Gupta has received a number of recognitions for his work, including the Wagner Prize for Excellence in the Practice of Advanced Analytics and Operations Research, the Pierskalla Best Paper Prize, the Jagdish Sheth Impact of Research on Practice Award.

Professor Sahin’s research interests include pricing, revenue and marketplace analytics, consumer choice models and strategic capacity management. Some of her recent research projects include analysis of pricing strategies with consumer search cost, ancillary service pricing, identifying consumer biases and heuristics for sequential decision making in retail. Professor Sahin has published papers in prestigious academic journals including Management Science, Operations Research, Manufacturing and Service Operations Management, among others. She is currently the Department Editor of revenue management and pricing area of Decision Sciences Journal.

Professor Pope is a behavioral economist that researches a variety of topics at the intersection of economics and psychology. He has published work in prestigious journals within economics, psychology, and management. His research primarily uses observational data and studies how psychological biases play out in important economic markets. Examples include left-digit bias and projection bias in car markets and present bias in housing markets. Prior to joining Chicago Booth faculty in 2010, Professor Pope was on the faculty at the Wharton School at the University of Pennsylvania. He earned a PhD in economics from UC Berkeley in 2007 and a BA in economics from Brigham Young University in 2002.

Michael leads economics at Microsoft. His Office of the Chief Economist has a team of economists and data scientists; in addition, economists across the company who report into product groups are hired and evaluated with the help of Michael and his team. The Office of the Chief Economist engages primarily in three areas: (1) Demand Forecasting – to optimize capacity planning/supply chain management and to identify top trends, (2) Product Pricing and Discounting (e.g. Azure Services, Surface, Windows) and (3) Market Design – to optimize the effectiveness of marketplaces. The team engages in a variety of other topics such as antitrust issues and zero-carbon energy. The team also helps the Microsoft sales organization engage with governments and our largest customers. Michael drives key topics and themes that need cross-company Economics representation.

We study the matching of jobs to workers in a queue, e.g. a ridesharing platform dispatching drivers to pick up riders at an airport. Under FIFO dispatching, the heterogeneity in trip earnings incentivizes drivers to cherry-pick, increasing riders’ waiting time for a match and resulting in a loss of efficiency and reliability. We first present the direct FIFO mechanism, which offers lower-earning trips to drivers further down the queue. The option to skip the rest of the line incentivizes drivers to accept all dispatches, but the mechanism would be considered unfair since drivers closer to the head of the queue may have lower priority for trips to certain destinations. To avoid the use of unfair dispatch rules, we introduce a family of randomized FIFO mechanisms, which send declined trips gradually down the queue in a randomized manner. We prove that a randomized FIFO mechanism achieves the first best throughput and the second best revenue in equilibrium. Extensive counterfactual simulations using data from the City of Chicago demonstrate substantial improvements of revenue and throughput, highlighting the effectiveness of using waiting times to align incentives and reduce the variability in driver earnings.

In emerging economies, there is a growing number of farm equipment sharing platforms that connect smallholder farmers with tractor owners who are willing to fulfill farmers’ requests for mechanization services. Due to the small farm sizes and the low digital literacy in rural areas of emerging economies, these platforms often rely on the so-called “booking agents” to collect demand from individual farmers and submit the aggregated demand on the platform (rather than having individual farmers submit their service requests). This paper studies how the presence of such booking agents affects the platform’s optimal pricing and wage decisions and equilibrium outcomes. Our analysis reveals several insights. First, we show that the platform should consider paying out a higher percentage of the price to booking agents when the number of tractors on the platform increases, when tractor owners’ operating cost decreases, and surprisingly, when the platform puts more emphasis on its profit (over farmer surplus). Second, in contrast to conventional sharing settings, we show that in the presence of booking agents, an increase in the number of service providers (i.e., tractors) may lead to a lower optimal platform commission. Finally, while efforts to enhance the supply side of the platform (e.g., increasing the number of tractors or reducing tractor owners’ operating cost) generally lead to a decrease in price and an increase in total farmer surplus, we find that reducing booking agents’ demand aggregation cost can lead to an increase in price and a decrease in total farmer surplus.

We consider centralized dynamic matching markets with finitely many agent types and heterogeneous match values. A network topology determines the feasible matches in the market and the value generated from each match. An inherent trade-off arises between short- and long-term objectives. A social planner may delay match decisions to thicken the market and increase match opportunities to generate high value. This inevitably compromises short-term value, and the planner may match greedily to maximize short-term objectives.

We characterize networks where this tradeoff is moot and introduce matching policies that achieve the optimal balance between short- and long-term objectives.  This balance depends, we show, on the general position gap, a measure that quantifies the demand-imbalance in the network. The optimal policies highlight the fundamental difference between multi-way and two-way networks.

The talk is based on joint work with Suleyman Kerimov and Itai Ashlagi.

Complementor innovation is an essential form of value co-creation in open platform ecosystems. However, the increasing platform fragmentation, i.e., users in the ecosystem adopting different versions of the platform technologies, has significantly hindered complementor innovation. For instance, there are a dozen different versions of the Android operating system currently running among Android devices, which increases the cost of complementor innovation as app developers must exhaustively test on different versions of the operating system before releasing new updates. Reducing platform fragmentation is a complex coordination problem involving multiple parties in the ecosystem, thus it is unclear whether and how a platform’s efforts to fight fragmentation would affect complementor innovation. Focusing on recent efforts by Google to address Android fragmentation, we find that while the platform investment does not immediately reduce Android fragmentation, app developers respond to the platform’s action by significantly increasing their innovation efforts in updating apps shortly after the platform investment. We find support of two possible explanations for the developers’ forward-looking behavior: a lower anticipated cost structure and a higher perceived platform value in future. Our research highlights the role of platform commitment to improve platform infrastructure in complementor innovation and provides implications for platform investment and intervention.

Continuity of care, defined as an ongoing therapeutic relationship between a patient and a physician, is a defining characteristic of primary care. However, arranging a consultation with one’s regular doctor is increasingly difficult as practices face physician shortages. We study the effect of declining care continuity on the productivity of physicians by analyzing data of over 10 million consultations in 381 English primary care practices over a period of 11 years. Specifically, we examine whether a consultation with the patient’s regular doctor is more productive than with another doctor in the practice. Using statistical models that account for confounding and selection bias and restricting the sample to consultations with patients who had at least three consultations over the past two years, we find that the time to a patient’s next visit is on average 18.1% (95% CI: [16.9%, 19.2%]) longer when the patient sees the doctor they have seen most frequently over the past two years, while there is no operationally meaningful difference in consultation duration. The data shows that the productivity benefit of care continuity is larger for older patients, patients with multiple chronic conditions, and patients with mental health conditions. We estimate that the total consultation demand in our sample could have fallen by up to 5.2% had all practices offered continuity of care at the level of the top decile of practices while prioritizing patients expected to yield the largest productivity benefits. We discuss operational and strategic implications of these findings for primary care practices and for third-party payers.

We develop a framework for quantifying delay propagation in airline networks. Using a large comprehensive data set on actual delays and a model-selection algorithm (elastic net) we estimate a weighted directed graph of delay propagation for each major airline in the US. We use these estimates to decompose the airline performance into “luck” and “ability.” We find that luck may explain about 38% of the performance difference between Delta and American in our data. We further use these estimates to describe how network topology and other airline network characteristics (such as aircraft fleet heterogeneity) affect the expected delays. Finally, we propose a model of aircraft scheduler who decides which flights to delay and by how much. We then use the estimated model to evaluate counterfactual scenarios of investments in airport infrastructure in terms of their impact on delays.

How should a platform or a society address the problem of fake news? The spread of misinformation is ancient, complex, yet ubiquitous in media concerning elections, vaccinations, and global climate policy.  After examining key attributes of “fake news” and of current solutions, this article presents design tradeoffs for curbing fake news. The challenges are not restricted to truth or to scale alone.  Surprisingly, there exist boundary cases when a just society is better served by a mechanism that allows lies to pass, even as there are alternate boundary cases when a just society should put friction on truth.  Harm reflects an interplay of lies, decision error, scale, and externalities.  Using mechanism design, this article then proposes three tiers of solutions: (1) those that are legal and business model compatible, so firms should adopt them (2) those that are legal but not business model compatible, so firms need compulsion to adopt them, and (3) those that require changes to bad law.

The first set of solutions, grounded in choice architecture, seek to alter information sets available to those affected by misinformation. By enabling transparency into not simply the content and sources but also the distribution and destination, the system provides effective means for counter narratives that are infeasible under current transparency proposals.

The second set of solutions, based in externality economics, considers how to protect free speech while updating Section 230. Revisions have faced two main critiques: one, that holding platforms actionable for false speech would cause them to take down user speech, and two, that ambiguity in individual messages makes judgement of false speech infeasible at scale. Whistleblower testimony before congress emphasized platform amplification of content in pursuit of engagement.  A targeted solution, therefore, can separate original speech from amplified speech, generously protecting the former while reverse amplifying the latter.  The posting and even discovery of false speech is protected even as amplification is unprotected.  The second element uses scale as an advantage.  Rather than vet every message, the system takes only statistical samples.  The Central Limit Theorem guarantees that establishing the presence of misinformation in amplified speech is feasible to any level of desired accuracy simply by taking larger samples.

The third set of solutions imports insights of antitrust jurisprudence into free speech jurisprudence.  The paradox of antitrust before 1978 was that legal decisions, intended to protect consumers and free markets, artificially raised prices by protecting inefficient firms from consequences of competition.  Free speech rulings vigorously protect speakers on the basis of enabling a free market of ideas. Overzealous protection of those promoting false facts, however, prevents the market from clearing itself.  No government intervention is required.  Rather, it simply needs to step aside in such cases as WASHLITE v Fox News, where numerous false stories that covid is no worse than flu and that vaccines do not work have been causally implicated in thousands of unnecessary deaths. The free speech paradox is that legal decisions intended to protect citizens and free idea markets artificially raise harms and protect those pushing false claims from consequences of acting on those claims.

Paper:  https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3997980

Problem definition: To respond to pandemics such as COVID-19, policy makers have relied on interventions that target specific population groups or activities. Since targeting is potentially contentious, rigorously quantifying its benefits is critical for designing effective and equitable pandemic control policies.

Methodology/results: We propose a flexible modeling framework and a set of associated algorithms that compute optimally targeted, time-dependent interventions that coordinate across two dimensions of heterogeneity: age of different groups and the specific activities that individuals engage in during the normal course of a day. We showcase a complete implementation focused on the Île-de-France region of France, based on commonly available public data. We find that targeted policies generate substantial complementarities that lead to Pareto improvements, reducing the number of deaths and the economic losses, as well as the time in confinement for each age group. Optimized dual-targeted policies are interpretable: by fitting decision trees to our raw policy’s decisions across many problem instances, we find that a feature corresponding to the ratio of marginal economic value prorated by social contacts is highly salient in explaining the confinements that any group – activity pair experiences. We also propose coarser and thus more practical targeting mechanisms that avoid explicitly enforcing differential confinements across age groups, while still capturing some of the benefits of dual targeting.

Implications: Given that some amount of targeting of activities and age groups is already in place in real-world pandemic responses, our framework highlights the significant benefits in explicitly and transparently modelling targeting and identifying the interventions that rigorously optimize overall societal welfare.

When discussing the ills afflicting social media, there is a great deal of concern about the role played by partisan animosity. Among other negative consequences, dislike of counter-partisans – and political motivations more generally – have been suggested to promote belief in, and sharing of, misinformation. While partisanship may be part of the misinformation problem, here I will argue that political motivations are also essential for one of the only possibilities for identifying (and therefore combatting) misinformation at scale – the wisdom of crowds. While professional fact-checkers play a critical role in countering misinformation, they are relatively few in number, and cannot possibly keep up with the vast amount of content posted on social media every day. Recent work has suggested that it is possible to supplement professional fact-checking by harnessing the wisdom of crowds: the ratings of fairly small politically-balanced groups of laypeople can generate high levels of agreement with professional fact-checkers. A central challenge for conducting crowd-based evaluations at scale, however, is the problem of encouraging participation – why should people bother to flag misleading content? In this talk, I will argue that not wanting people to be exposed to posts by counter-partisans helps to solve this participation problem by motivating people to flag. Although extreme partisans would flag all counter-partisan content as misleading regardless of its actual truth value, such extreme partisans are rare. A much larger group of people care somewhat about both truth and partisanship, such that they would only be sufficiently motivated to flag when content is both misleading and counter-partisan. For these people, the partisan motivation is needed to drive participation – without any partisan motive they would flag nothing. I will present experimental and observational social media data supporting this account.

We discuss how smart logistics can facilitate environmental aspirations to combat climate change.

Some smart logistics utilize dual sourcing models to increase slower carbon-friendly transport modes while remaining responsive.

We review analytic solutions to this classic model and compare them with recent deep learning solutions.