We develop a methodology to estimate the transaction costs required to arbitrage among a set of products and the probability of observing binding arbitrage. The estimation is based on a switching regimes model, identified by a truncation in the error structure. We analyze its antitrust implications, and apply it to wholesale gasoline in the Northeastern United States. The main findings are: Transaction costs are below 5% of price, increasing with distance and depend on the direction of arbitrage; arbitrage probabilities fall with distance and with transaction costs. Finally, the model discriminates well between regimes, suggesting its usefulness for other applications.