| Summary: | <p>Feeding America, an organisation responsible for feeding 130,000 Americans every day, distributes donated food among a network of participating food banks. Feed- ing America’s allocation mechanism, the ‘Choice System’, uses repeated rounds of simultaneous first-price auctions to allow food banks to signal which types of food they need from Feeding America. This provides food banks a large degree of choice over the types of food they receive. This thesis examines the welfare and distribu- tional consequences of enabling this choice.</p>
<p>I develop an empirical model of bidding in repeated rounds of simultaneous first-price auctions. I prove non-parametric identification of primitives in this dy- namic multi-object auction model, and introduce a computationally feasible pro- cedure to estimate this type of game. The difficulty arises because, when players cannot place package bids, bids do not uniquely identify bidders’ valuations.</p>
<p>I then apply this model to Choice System bidding data, estimating the distribu- tion of food banks’ heterogeneous and time-varying needs. The central challenge is that I cannot observe food banks’ inventories — the key determinant of bidding be- haviour. Nonetheless, I prove that observations of food banks’ winnings, which are just observed changes in their unobserved stocks, are sufficient for nonparametric identification. I propose a Bayesian estimator to estimate food banks’ needs in the presence of the latent inventories. I then use these estimates to compare the Choice System to the previous allocation mechanism employed by Feeding America which gave food banks very limited choice. I estimate that the Choice System increased welfare by the equivalent of a 17.1% increase in the quantity of food being allocated, and that on average 85% of food banks are strictly better off from this change. I find that the majority of this welfare gain arises because the Choice System allocates food in batches, rather than sequentially.</p>
|
|---|