Nonparametric Tests for Common Values

We develop tests for the presence of common value components in bidders' valuations at symmetric first-price sealed bid auctions. These tests are fully nonparametric and require observations only of the bids submitted at each auction. The main principle of the test relies on the observation that the winner's curse is only present in common value auctions. In previous work, Guerre, Perrigne, and Vuong (1999) have devised a method for recovering particular functionals of bidders' latent signals in nonparametric fashion from data on bids. We build on this by noting that, under the common value hypothesis, the distribution of these functionals in strictly increasing (in a first-order stochastic dominance sense) in the number of bidders, whereas under the private value hypothesis these distributions should be identical across all numbers of bidders. This forms the basis of the tests that we propose. We develop the asymptotic distribution theory for our tests, and evaluate their performance in a variety of Monte Carlo experiments. We illustrate their use in a case study of the United States Forest Service (USFS) timber auctions.