UCLA

My dissertation contributes to the literature on prior-free (robust) mechanism design. Prior-freeness can be interpreted differently, but a common feature is that certain mechanisms can be ranked above the others without the exact knowledge of distributions and/or utilities. According to the Wilson critique, the knowledge of fine details of the setting such as distributions and utilities is an unrealistic assumption and, moreover, optimal mechanisms in the classic (Bayesian) sense are often too complex to be implemented in reality.

In the first chapter I study a scoring auction and the welfare implications of switching between the two leading designs of the scoring rule: linear (``weighted bid'') and log-linear (``adjusted bid''), when the designer's preferences for quality and money are unknown. Motivated by the empirical application, I formulate a new model of scoring auctions, with two key elements: exogenous quality and a reserve price, and characterize the equilibrium for a rich set of scoring rules. The data is drawn from the Russian public procurement sector in which the linear scoring rule was applied from 2011 to 2013. I estimate the underlying distribution of firms' types nonparametrically and simulate the equilibria for both scoring rules with different weights. The empirical results show that for any log-linear scoring rule, there exists a linear one, yielding a higher expected quality and rebate. Hence, at least with risk-neutral preferences, the linear design is superior to the log-linear.

In the second chapter (Co-authored with Jernej Copic and Byeong-hyeon Jeong, UCLA) I study robust allocation of a divisible public good among n agents with quasi-linear utilities, when the budget is exactly balanced. Under several additional assumptions, we prove that such mechanism is equivalent to a distribution over simple posted prices. A robustly optimal mechanism minimizes expected welfare loss among robust divisible ones. For any prior belief, I show that a simple posted prices is robustly optimal. This justifies a restriction to binary allocations commonly found in the mechanism design literature. Robustness comes at a high cost. For certain beliefs, we show that the expected welfare loss of an optimal posted price is as big as 1/2 of the expected welfare in the corresponding optimal Bayesian mechanism, independently of the size of the economy. This bound is tight for the special case of two agents.

In the third chapter (Co-authored with Tomasz Sadzik, UCLA) I provide mechanisms for exchange economies with private information and interdependent values, which are ex-post individually rational, incentive compatible, generate budget surplus and are ex-post nearly efficient, when there are many agents. Our framework is entirely prior-free, and I make no symmetry restrictions. The mechanisms can be implemented using a novel discriminatory conditional double auction, without knowledge of information structure or utility functions. I also show that no other mechanism satisfying the constraints can generate inefficiency of smaller order.