A new economics study examines how private information and reputation-building can reshape competition in markets where firms repeatedly decide how much to produce. Andrew Sweeting, Xuezhen Tao, and Qian Wang develop a model of oligopoly quantity competition that unfolds over a finite number of periods. In each period, multiple sellers supply a homogeneous product facing linear demand, but each seller privately knows its own marginal cost—a cost that also evolves over time and is correlated from one period to the next.
That serial correlation matters because it links today’s production choice to beliefs about tomorrow’s costs. In the authors’ framework, rivals infer information from observed quantities: producing more can be interpreted as a sign that a seller has low marginal costs. The central idea is that firms may “produce to signal”—choosing higher output not only for immediate profit, but also to convince competitors they are efficient, thereby influencing future competitive behavior.
The paper focuses on “fully separating” equilibria, in which different cost types choose different quantities, allowing actions to reveal private information. Within that class of outcomes, the authors derive conditions under which such equilibria exist and when they are unique. Establishing existence and uniqueness is crucial for making clear predictions, since dynamic games with private information can otherwise support multiple, sharply different patterns of behavior.
A key conclusion is that signaling incentives can meaningfully change market outcomes compared with common benchmark settings—both static models with complete information (where all costs are known) and static models with incomplete information (where costs are private but there is no intertemporal learning). In the dynamic setting, the desire to be perceived as low-cost can push firms’ quantities away from what would be chosen if output were determined only by contemporaneous profit considerations. The authors report that these effects can be large enough to alter total production, the allocation of production across firms (productive efficiency), and overall welfare. Notably, the distortions from signaling do not disappear simply because there are many competitors: the paper argues that the qualitative and quantitative impacts can remain substantial even in industries with ten or more sellers.
To connect the theory to a major real-world market, the researchers apply the model to global oil production. The oil market is a natural testing ground because producers repeatedly choose quantities, costs can fluctuate persistently, and market participants often try to infer others’ capabilities from their production decisions. The study also compares the model’s predictions with an alternative environment in which sellers can commit to quantities, highlighting how the ability (or inability) to make binding commitments changes strategic behavior and the informational content of observed output.
The project received support from the U.S. National Science Foundation (Grant SES-1658670, supporting Sweeting) and China’s National Natural Science Foundation (Grant 72673107, supporting Tao). The authors note that Xinlu Yao contributed as a graduate student at the University of Maryland, and they acknowledge Mario Leccese’s role in related work (Leccese, Sweeting and Tao, 2025). They also credit Dan Vincent for comments and Gaurab Aryal for discussions connected to Aryal and Zincenko (2024). The team reports using the AI system Claude in summer 2026 to improve performance of previously written MATLAB code, while emphasizing that any remaining mistakes are their own. As with other National Bureau of Economic Research outputs, the findings reflect the authors’ views, not necessarily those of the NBER.
Why it matters
If firms produce extra output to look low-cost, markets can end up with different prices, efficiency, and welfare than standard textbook models predict—even when many sellers compete. For policymakers and analysts following industries like oil, the work suggests that observed production may reflect strategic signaling as much as fundamentals. What to watch next is how well these dynamic, information-driven predictions match production patterns in real data and how outcomes shift when producers can credibly commit to future quantities.