I investigate how the presence of learning affects the market dynamics in three different market settings. The first chapter studies how the interplay of individual and social learning affects price dynamics. I consider a monopolist selling a new experience good over time to many buyers. Buyers learn from their own private experiences (individual learning) as well as by observing other buyers' experiences (social learning). Individual learning generates ex post heterogeneity, which affects the buyers' purchasing decisions and the firm's pricing strategy. When learning is through good news signals, the monopolist's incentive to exploit the known buyers causes experimentation to be terminated too early. After the arrival of a good news signal, the price could instantaneously go down in order to induce the remaining unknown buyer to experiment. When learning is through bad news signals, experimentation is efficient, since only the homogeneous unknown buyers purchase the experience good. The second chapter is based on the observation that workers learn at different rates about their productivity and therefore expect different wage paths across firms. We show that under strict supermodularity there is always positive assortative matching: differential learning is always dominated by the impact of productivity. Surprisingly, this holds even if learning is faster in the low type firm. The key assumption driving this result is that this is a pure Bayesian learning model.We also derive a new equilibrium condition in this class of continuous time models in addition to the common smooth-pasting and value-matching conditions. This no-deviation condition captures sequential rationality and results in a restriction on the second derivative of the value function. The third chapter develops a continuous-time war of attrition model with learning to investigate whether learning is possible to make it easier to reach an agreement. I show that with exogenous private learning, it may be easier to reach an agreement initially but it becomes more and more difficult over time. The expected delay will always be higher than the expected delay without learning. I also show that when allowing only one player to learn leads to a shorter delay than allowing both to learn.