In this paper, I address the stochastic behavior of asset prices set by an imperfectly informed specialist who uses learning technology to refine her knowledge of the order flow. The specialist’s endogenous choice of an information structure is analyzed which fully characterizes her responses to large orders in the market. Specifically, large orders can either be of structural origin, i.e., a disturbance in the asset’s payoff, or an exogenous one associated with noise trading. A specialist with a large learning capacity optimally chooses a pricing function where structural shocks display high persistence, whereas exogenous shocks disappear rapidly. This market structure provides a natural setup to address market resilience, in the sense of the recovery speed of prices.