Biasing Unbiased Dynamic Contests

We consider a best-of-three Tullock contest between two ex-ante identical players. An effortmaximizing designer commits to a vector of player-specific biases (advantages or disadvantages). In our benchmark model the designer chooses victory-dependent biases (i.e., the biases depend on the record of matches won by players); the effort-mazimizing biases eliminate the discouragement effect, leaving players equally likely to win each match and the overall contest. We contrast our benchmark model with one where the designer chooses victory-independent biases; the effort-maximizing biases leave players unequally likely to win each match and the overall contest. This result holds in Tullock contests and all-pay auctions, as well as under maximization of total effort and winner's effort. The appeal of our result comes from the players being ex-ante identical; thus, it challenges the conventional wisdom of optimality of unbiased contests. Our result has also an applied interest, as it shows that alternating biases, as when teams alternate home and away games, may increase total effort as opposed to an unbiased contest.