This seminar will be delivered by Emmanuel Guerre (Queen Mary, University of London). It is based on the paper ‘Quantile regression methods for first-price auction: A signal approach’ joint with Nathalie Gimenes, University of Sao Paulo.
This paper considers a quantile signal framework for first-price auction. Under the independent private value paradigm, a key stability property is that a linear specification for the private value conditional quantile function generates a linear specification for the bids one, from which it can be easily identified. This applies in particular for standard quantile regression models but also to more flexible additive sieve specification which are not affected by the curse of dimensionality. A combination of local polynomial and sieve methods allows to estimate the private value quantile function with a fast optimal rate and for all quantile levels in [0,1] without boundary effects. The choice of the smoothing parameters is also discussed. Extensions to interdependent values including bidder specific variables are also possible under some functional restrictions, which tie up the bidder covariate and signal as in an auction with resale example.