A manufacturer sets wholesale prices for a retailer. Both want to learn the demand distribution, but neither observes it directly. What they see is sales — which equals demand only when inventory doesn't run out. When it does, the true demand is censored: they know at least this many people wanted to buy, but not how many more.
Kanoria and Kim characterize equilibrium pricing under this mutual ignorance. The public belief about demand updates as a Markov system, where each period's observation (censored or not) shifts both players' estimates. The manufacturer and retailer each optimize against their expectations of the other's strategy, producing a dynamic game where learning and pricing are inseparable.
For Weibull demand distributions, the scaling method from extreme value theory adapts to the strategic context: equilibrium exists and can be computed through a standardized recursion that exploits the distribution's self-similarity. For exponential demand, equilibrium is unique and computable by backward induction. The computational tractability comes from the demand family's structure, not from simplifying the game.
The through-claim is about what censoring does to strategic learning. In single-agent settings, censored demand is a technical nuisance — you correct for the truncation and learn the true distribution. In strategic settings, the censoring itself becomes a weapon. A manufacturer who prices high causes stockouts, which censor demand information, which makes the retailer uncertain, which shifts bargaining power. The ignorance is not just an obstacle to learning — it is an outcome of the game, produced by the players whose strategies depend on it.