Working Papers
Communicating with Models and Data (Job Market Paper, Economics)
September 2026
Many real-world decisions rely on external data collection and modeling: consultants provide clients scenarios, portfolio managers present models of asset returns, and AI chatbots offer recommendations based on internal data. I study how strategic data collection and model simplification distort these interactions. A sender collects data and estimates a model that helps a receiver make a choice. However, he estimates a simple model that artificially restricts the curvature of the relationship between choices and outcomes: either a "naive" model that assumes low curvature and neglects the costs of actions, or a "cautious" model that assumes high curvature and exaggerates them. I show that senders who estimate naive models strategically sample data across a wide range of potentially irrelevant choices. Those who estimate cautious models sample locally around an extreme choice. When choices are multivariate, the sender samples data to maximize collinearity between different variables. These distortions generate gains from model simplicity: a sender who reports an accurate model gains if the receiver misperceives his model as naive or cautious. In equilibrium, accurate senders surround themselves with simple senders who estimate models that fix curvature at the most extreme values the receiver finds plausible. These senders polarize into two groups: those who estimate maximally naive models and those who estimate maximally cautious ones. In the limit, the former group estimates a linear regression model and the latter a "worst-case scenario" model.
Persuasion with Partisanship: The Informational Content of Policymaking with Application to US Governors (Job Market Paper, Political Science)
June 2026
I study a persuasion model where a political executive's policy agenda generates information about her ability. Since partisan policies are harder to pass, an agenda's partisanship affects what success or failure communicates about ability. The model predicts a U-shape between office-motivated executives' popularity and their agendas' partisanship. I then use novel text data to analyze the partisanship of U.S. governors' policy proposals from 1990-2020. Despite a rise in partisanship over the sample period, I argue that electoral incentives can explain substantive variation in partisanship, testing the model and showing a U-shape between partisanship and approval as well as other comparative statics.
Optimal Push and Pull Funding for Innovation (with Chris Snyder)
August 2026
We study whether a designer should fund underprovided private sector innovations using "push funding" — such as research grants — or "pull funding" — such as prizes. We show that incentive-compatible contracts involve rewarding higher-probability-of-success firms with more in pull funding and less in push funding, and show that they can be characterized by convex functions. While push funding screens firms based on their costs of developing an innovation, pull funding selects for firms who are more likely to develop the innovation, but allows them to collect greater rents than push funding, generating a tradeoff. For a large class of distributions, the optimal contract takes a simple, "affine" form: all firms are offered the same amount of each type of funding, regardless of their type. In these cases, we show that the most cost-efficient contract always involves a positive amount of pull funding, and show precisely when a designer should also utilize a positive amount of push and pull.
Misconduct in Organizations (with Kim Sarnoff)
June 2026
We study how policies that disincentivize misconduct in organizations can generate changes in abusers' behaviors that negatively impact victims. We describe a model where "managers" choose to commit harmful actions of varying intensities against "employees," who can report these actions as "misconduct." We show that when the marginal disutility from managers' actions is particularly small, increasing the ease of reporting misconduct, the severity of punishment for managers, or the efficacy of investigation technology may in fact harm employees. These policies may motivate managers to commit harmful actions that employees do not want to report or induce managers to opt out of interacting with employees altogether. We provide a dynamic extension where reports generate precedents for the organization and employees, showing that the model converges to a steady state where employees are worse off than initially and harmful actions are never punished.
The Twin Puzzles of U.S. Productivity Growth Dynamics: A New Interpretation (with Robert J. Gordon)
November 2025, NBER Working Paper 30267, CEPR DP19569
This paper provides a unified framework that resolves recent puzzles in U.S. productivity growth that we show are interrelated. First, why was productivity growth in the 2010-19 decade the slowest of any decade in U.S. history? Second, why did the cyclicality of productivity growth change from procyclical in 1950-85 to acyclical in 1986-2006 and then back to procyclical in 2010-19? Third, why was productivity growth strongly countercyclical in the recessions of 2008-09 and 2020? The fundamental dynamic driving cyclical productivity fluctuations originates in the gradual adjustment of hours of work to demand-driven output fluctuations due to the costs of hiring and firing labor. Since productivity growth is a residual by definition, equal to output growth less hours growth, productivity growth immediately jumps in response to an upward output movement because hours are slow to respond and then falls back in subsequent quarters as hours complete their adjustment. We are able to explain the temporary 1986-2006 disappearance of procyclicality and its reappearance after 2009 as the result of changes in the standard deviation and serial correlation of output changes.
We explain countercyclical productivity surges in 2008-09 and 2020 by showing that business firms in those two episodes overreacted with “excess layoffs,” cutting hours in response to the sharp output decline with a much higher elasticity than normal. By coupling these excess layoffs with a post-recession rehiring effect that boosted hours growth and depressed productivity growth for 2010-19, our regression analysis explains why productivity growth on average was so slow during that decade. If this recession/rehiring effect had not occurred, productivity growth in the 2010-19 decade would have been 2.1 percent per year instead of 1.2 percent, suggesting that concern about U.S. “secular stagnation” has been overstated.
Press coverage in The Economist, NBER Digest.
Publications
June 2026, Forthcoming: American Journal of Political Science
When constituents are uncertain about whether leaders' policy preferences align with their interests, policy choices generate a tradeoff between payoffs and information. Leaders can implement their preferred policy, but if the outcome is bad, constituents learn this and threaten office removal. I use a repeated model of political agency to analyze how this tradeoff shapes long-lived, term-unlimited leaders' policy choices. When there is uncertainty about the optimal policy for constituents, leaders implement "learning traps": spatially intermediate policies that halt learning about what the best policy is. They experiment only when their preferences are very likely to be aligned with constituents, a contrast to the "gambling for resurrection" logic ubiquitous in the literature. I provide a case study that examines how the 1861 reforms ending serfdom in Imperial Russia represented the logic of a "learning trap" and illustrate additional applications of the model to political contagion and economic policymaking.
Transatlantic Technologies: The Role of ICT in the Evolution of U.S. and European Productivity Growth (with Robert J. Gordon)
Spring 2020, International Productivity Monitor No. 38
We examine the role of the ICT revolution in driving productivity growth behavior for the United States and an aggregate of ten Western European nations (the EU-10) from 1977 to 2015. We find that the standard growth accounting approach is deficient when it separates sources of growth between ICT capital deepening and TFP growth, because much of the effect of the ICT revolution was channeled through spillovers to TFP growth rather than being limited to the capital deepening path- way. Using industry-level data from EU KLEMS, we find that most of the 1995-2005 U.S. productivity growth revival was driven by ICT-intensive industries producing market services and computer hardware. In contrast the EU-10 experienced a 1995- 2005 growth slowdown due to a paucity of ICT investment, a failure to capture the efficiency benefits of ICT, and performance shortfalls in specific industries in- cluding ICT production, finance-insurance, retail-wholesale, and agriculture. After 2005 both the United States and the EU-10 suffered a growth slowdown, indicating that the benefits of the ICT revolution were temporary rather than providing a new permanent era of faster productivity growth. Also circulated as NBER WP 27425.
Press coverage in NBER Digest, VoxEU.
The Industry Anatomy of the Transatlantic Productivity Growth Slowdown: Europe Chasing the American Frontier (with Robert J. Gordon)
Fall 2019, International Productivity Monitor No. 37
By merging KLEMS data covering 16 industry groups within the total economy and 11 manufacturing sub-industries, we compare and contrast productivity growth from 1950 to 2015 in the United States with an aggregate of the ten largest European nations (EU-10) from 1972 to 2015. We interpret the EU-10 performance as catching up to the United States in stages. Strikingly, the total economy "early-to-late" productivity growth slowdown from 1972-1995 to 2005-2015 in the EU-10 (-1.68 percentage points) was almost identical to the U.S. slowdown from 1950-1972 to 2005-2015 (-1.67 percentage points). There is a very high EU-U.S. correlation in the magnitude of the early-to-late slowdown in each industry, suggesting that the productivity growth slowdown from the early postwar years to the most recent decade was due to a retardation in technical change that affected the same industries by roughly the same magnitudes on both sides of the Atlantic. Also circulated as NBER WP 25703.
Press coverage in The Economist, VoxEU.
Policy and Public Writing
Generic Drug Repurposing Incentive Program Proposal (with Chris Snyder, Akhil Bansal, and Sarrin Chethik)
Submitted to FDA RFI: Drug Repurposing for Unmet Medical Needs
July 2026
We provide a detailed policy proposal for a drug repurposing incentive program. To incentivize companies to pursue promising new uses for generic drugs, we propose that a government funder, such as the Centers for Medicare & Medicaid Services (CMS) and/or the National Institutes of Health (NIH), use "pull" funding – a prize-like approach – to reward the successful discovery and development of a generic drug for a new use. Our proposal includes a mechanism where a government funder rewards companies who repurpose generic drugs with adoption-linked payments, determines who is eligible for those payments prior to clinical trials using a secret application process, and potentially runs an auction as part of that application process.
Transatlantic technologies: Why did the ICT revolution fail to boost European productivity growth? (with Robert J. Gordon)
VoxEU
August 2020
The benefits of the ‘ICT revolution’ are readily seen in labour productivity statistics for the US, but a similar acceleration of productivity growth was not seen in Western Europe. This column argues that most of the 1995-2005 US productivity growth revival was driven by ICT-intensive industries producing market services and computer hardware. In contrast, the EU10 experienced a 1995-2005 growth slowdown due to a paucity of ICT investment, a failure to capture the efficiency benefits of ICT, and performance shortfalls in specific industries. After 2005 both the US and the EU10 suffered a growth slowdown, indicating that the benefits of the ICT revolution were temporary rather than providing a new permanent era of faster productivity growth.
The industry anatomy of the transatlantic productivity growth slowdown: Similarities outweigh the differences (with Robert J. Gordon)
VoxEU
August 2019
Since 2005, productivity growth in the US and Europe has dipped below 1%. Using new industry-level from the US and ten EU countries, this column shows that that the industrial composition of the slowdown was similar in Europe and the US. Falling multifactor productivity growth explains both the magnitude and composition of falling productivity growth on both sides of the Atlantic. Decelerating technical change, rather than slowing investment, was the primary driving force in the transatlantic slowdown.
Selected Work in Progress
Inference from Choices and Spatial Differentiation
(Slides available on request)
Producers or politicians can differentiate themselves horizontally (in product/policy spaces) or vertically (in quality space). Consumers and voters make inferences about products or politicians by observing others' choice data — such as who won an election or who purchased a product — but without observing others' information. I show how career concerns distort producers' (or politicians') willingness to horizontally differentiate themselves from the competition, particularly in settings of dynamic choice, leading to convergence on inefficient policies.
Model-Based Inference (with Kim Sarnoff)
How do everyday Americans make inferences about causal processes underlying economic outcomes? This experimental project uses a "coefficient weight" method to elicit how participants create mental models that organize data.