Why Vote for Losers?

Voting Theory generally concludes that -in first-past-the-post elections- 1) All voters should go to effective candidates (Duvergers Law); 2) Parties platform should converge (Median Voter Theorem). Observations, though, suggest that such predictions are not met in practice. We show that divergence and dispersion of votes is a natural election outcome when there is uncertainty and repetition of elections. "Voting for Losers" increases the informational content of elections, and forces main parties to relocate towards extremists. As a result, they maximize their probability of being elected, not by converging to the median but by diverging to a certain extent. Ideological behavior results then from optimizing considerations alone.