This paper asks how licensing deals change which drug projects get developed and how well markets for drug ideas work. The authors build a simple idea: some projects are incremental and easier to judge, like “me‑too” or follow‑on drugs, while others are novel and hard to evaluate, like first‑in‑class new molecular entities. When information about a project is clear, licensing can lead to good selection and fair returns. When information is noisy, markets can suffer “lemons” problems and fail to price risk correctly.

The researchers develop a theoretical model with three main predictions. First, whether licensing attracts high‑quality projects depends on how precise information is. Second, when signals are precise, competition pushes expected returns of licensed and in‑house projects toward equality. Third, because licensing involves coordination between firms, it can raise development costs compared with in‑house work if those coordination costs are large.

They test these ideas with product‑level data that include development outcomes, clinical trial costs, and commercial revenues. The authors construct predicted success probabilities using many project features and modern machine‑learning tools. For causal checks they use a two‑step approach. One step uses Double Machine Learning (DML) to control for many confounders. The other step instruments for licensing with exogenous pipeline shocks, specifically Phase III clinical trial failures at potential licensee firms, which creates a plausibly random source of “rushed” licenses. That instrument identifies a local average treatment effect (LATE) for deals accelerated by pipeline pressure.

What they find is consistent with the theory. Across the whole sample, licensed projects show higher success probabilities. But the pattern differs by innovation type. For incremental, low‑risk projects, licensing raises success probability while competition equalizes net returns between licensed and internal projects. For novel, high‑risk projects, the return equalization breaks down. Licensed frontier projects may keep higher success rates without a matching return penalty, and when licenses are rushed by pipeline shocks the trade‑off deteriorates: rushed incremental licenses still behave like a competitive market, but rushed novel licenses show a loss in returns without a clear success‑probability benefit.