Auctions vs. Posted Prices: When Do Sellers Switch Formats?
Job Market Paper [Draft]
Sellers often do not know the demand for their goods. An auction reveals it but risks no sale
and adds delay; posting a price is immediate and certain but ends demand learning. This paper
studies this trade-off on an online used-car platform, where sellers repeatedly auction the same
car and may switch to a posted price. In a dynamic discrete–continuous choice model with
Bayesian learning, risk-averse sellers update their beliefs based on auction outcomes and, in
each period, choose a format and a secret reserve or a posted price. Repeated auctions of the
same car identify how informative auctions are, and posted prices after a switch identify the
seller’s risk aversion and delay costs. The model is estimated by maximum likelihood using an
expectation–maximization algorithm with a machine-learning first stage. A single auction resolves
most of a seller’s uncertainty, learning delivers roughly 70% of the value of full information, and
bad news triggers prompt switching. The two formats are complements, and mandating a switch
after two auctions shortens time on market at no cost to sellers. A reserve-recommendation tool
and a standing guaranteed-purchase offer each raise seller welfare; the offer is taken up precisely
after sellers learn that demand is weak.
Simultaneous Auctions with Common and Private Value Bidders: Evidence from Online Car Auctions
Revise and Resubmit at The RAND Journal of Economics | 2026 [Draft]
Presented at NASMES 2026, SEA 2026 (scheduled)
This paper develops a structural model of second-price auctions with two bidder types using
data from an online used-car platform. Dealers have common values and participate in multiple
simultaneous auctions. Individual buyers have independent private values and bid once. The
model captures interactions between common-value and private-value bidders and incorporates
complementarities arising from simultaneous dealer wins. I establish identification of signal
distributions and complementarities, then estimate the model parametrically in three steps. Results
show that dealers gain positive complementarities from winning auctions with similar characteristics.
Counterfactual analyses show that bundling auctions increases platform revenue but generates
non-negligible allocative inefficiency.
