In this thesis we are interested in how unobserved heterogeneity of agents affects the predictions from several different classical dynamic models that are widely used in economics. First, we capture heterogeneity in users’ preferences in order to obtain a better prediction for their movie ratings as our solution for the Netflix Prize competition. Our method combines user-based and item (movie) based methods in a spatial regression framework. Next, we introduce heterogeneous income profiles in a model of housing choices where households have options of renting, buying a house, and/or keeping the old house (if they already have one). While most lifecycle models of consumption and saving assume that individuals are ex-ante identical and face the same income process, we allow for the more realistic setting where each individual faces a different income process. We next investigate lifetime saving and investing behaviors of US households using the Panel Study of Income Dynamics (PSID) to detect changes in those behaviors due to retirement. Addressing heterogeneity in households’ saving and investing decision is essential in order to separate the aging effect from the household and cohort effect.