A. Serdar Simsek

Research


  • The Winner's Curse in Dynamic Forecasting of Auction Data: Empirical Evidence from eBay (with E. Haruvy, M. Hejazi Nia, and O. Ozer)

    Abstract
    Problem definition: Dynamic forecasting models in auctions have fallen short on two dimensions: (i) the lack of an equilibrium model for final stage bids and (ii) the lack of a winner's curse (i.e., a tendency to overpay conditional on winning the auction) adjustment to allow bidders to account for a common value component in the auction item. In this paper, we develop a methodology t o accurately predict final stage bids from the initial bidding dynamics and quantify the impact of the winner's curse. This methodology allows us to conduct policy simulations to optimize auction design parameters.

    Methodology: Dynamic auctions typically have a stage of high exploratory activity, followed by an inactivity period, and then a final stage of last-minute bids with sharp jumps. With a Kalman filter approach, we use exploratory stage bids to predict an auction item's valuation distribution. We feed this prediction into an equilibrium model and apply item-specific adjustments for winner's curse, bidder heterogeneity, and inactivity period. We use the resulting equilibrium model to predict the final stage bids.

    Results: Our methodology improves the forecast of final stage bids by 15.84%, on average, compared to a state-of-the-art benchmark. This improvement is even higher (22.44%) for common value auctions. We also find that: (i) significantly more (resp., fewer) bidders internalize the winner's curse in common value (resp., private value) auctions, (ii) bidders in common value (resp., private value) auctions decrease their bids by 6.69% (resp., 3.51%) due to the winner's curse, and (iii) the inactivity period has a lesser impact on the final stage bids in private value auctions.

    Managerial Implications: Our proposed methodology is intended to facilitate the need in academia and practice for real-time bid predictions that encompass different levels of the common value component in auctions. Using our methodology, auction platforms can support their choice of minimum bid increment policies and decide how to allocate resources across different auctions to mitigate the adverse effects of the winner's curse.