I am an applied microeconomist and a Ph.D. candidate in Economics at Duke University. I study what happens to government-funded R&D, how efficiently it turns public money into innovation, and what policies can improve its allocation. I take these questions to two countries at different levels of R&D activity (the United States and Türkiye), using administrative data and novel datasets I build from public records.
Applied Microeconomics · Industrial Organization · Economics of Innovation
Job Market Paper
Lobbying for Innovation? Evidence from Revolving-Door Connections in Defense R&D
Firms can exert influence over the congressional appropriations process that funds the R&D they perform. I study how lobbying directed at this process affects the allocation and innovation outcomes of publicly funded R&D. To answer this question, I build a novel dataset of late-stage R&D projects funded by the U.S. Department of Defense from 2004 to 2023 that records their funding, milestones, and their outcomes (successful completion or termination), linked to the lobbying activities of the firms that perform those R&D projects. To address the endogeneity of lobbying, I exploit revolving-door connections: a firm gains political access to a defense appropriator’s office when that office hires a former congressional coworker of a lobbyist at one of the firm’s lobbying shops. Within four years of gaining this access, the share of years in which firms lobby rises by 23 percentage points, Congress appropriates 17% more funding for their projects, and the share of their projects terminated by the Department of Defense falls by 70%. But their projects achieve no more milestones than before, and the share that reaches successful completion does not rise. Political access thus increases funding and prolongs development without advancing it, a pattern consistent with a rent-seeking view of lobbying.
Works in Progress
Unfinished Innovation: Project Risk and the Design of R&D Grants
Abstract
I study how project risk shapes what firms do with public R&D grants, using application records for the grants that the Scientific and Technological Research Council of Türkiye (TÜBİTAK) awards to firms, linked to administrative data on Turkish firms. With a novel regression discontinuity design that exploits variation in the reimbursement rates of these grants, I show that the grants increase the private R&D expenditures of recipient firms by more than the amount of the grant. The grants also raise these firms’ domestic sales, with no noticeable increase in their exports. These effects are concentrated among firms that do low-risk research. Grants that fund high-risk research are mostly underused. As firms learn more about the feasibility of their projects or about demand for the products they are developing, many stop drawing on the money and drop the project midway. Preliminary estimates imply that firms spend substantially less on R&D than they would if their projects carried no risk. Administrative rules fix the terms of each grant at approval, so a grant cannot adjust to what the firm learns, which limits its effects on high-risk research. My estimates therefore point to the need for a different grant design for risky projects. In the next step of this project, I ask what that design should be when firms learn about their projects after approval.
Allocative Efficiency of Public R&D: Evidence from NIH Funding
Abstract
This paper studies the allocative efficiency of public R&D funding using detailed data on NIH grants. We develop a framework in which heterogeneous research recipients transform public funds into knowledge and innovation outputs, measured by scientific publications and patents. The model highlights how distortions in funding allocation – captured as wedges between marginal products and marginal costs – affect aggregate research productivity. We estimate knowledge and innovation production functions exploiting panel variation in funding and introduce a novel identification strategy based on shifts in the composition of congressional committees and using dynamic panel data methods. Combining these estimates with administrative data linking grants to publications and patents, we recover recipient-level productivity and quantify misallocation across research areas. Our results show substantial heterogeneity in returns to funding across fields and recipients. Misallocation in knowledge production is economically large, with aggregate productivity losses ranging from roughly 10% to 85% relative to an efficient benchmark, while misallocation in innovation (patents) is more limited due to lower returns to scale. These findings imply that reallocating existing funds could generate sizable gains in scientific output without increasing total spending.
Data
I share the data used in my research. To request access, please email me at cem.ozdemir@duke.edu.
Congressional Staffer Data
The record of congressional staffers who have served a member of Congress since 1965, with their employment trajectories. Each staffer is linked to the offices and committees they served during their time in Congress.
DoD R&D Project Panel
A project-level panel of late-stage, unclassified Department of Defense R&D projects, built from the annual budget justification documents: requested and enacted funding, milestones and schedules, contractor information, and how each project evolves over time.