Daily deals sites, such as Groupon and LivingSocial, attract millions of customers in the hunt for products and services at substantially reduced prices (i.e., deals). An important aspect for the profitability of these sites is the correct prediction of how many coupons will be sold for each deal in their catalog—a task commonly referred to as deal size prediction. Existing solutions for the deal size prediction problem focus on one deal at a time, neglecting the existence of similar deals in the catalog. In this paper, we propose to improve deal size prediction by taking into account the context in which a given deal is offered. In particular, we propose a topic modeling approach to identify markets with similar deals and an expectation-maximization approach to model intra-market competition while minimizing the prediction error. A systematic set of experiments shows that our approach offers gains in precision ranging from 8.18% to 17.67% when compared against existing solutions.