Pairs trading is a very common trading strategy, and being able to obtain parameters that tell us when to trade and when to get out is of great importance. In this paper I propose a methodology that can improve the performance of traditional pairs trading strategy. I use stochastic beam search algorithm to find the best parameters in in-sample data within a given industry, and then test those parameters on out of sample data. My results outperform the results obtained by tools traditionally employed by the industry practitioners.