We model endogenous correlation in asset returns via the role of heterogeneous expectations in investor types, and the dynamic impact of imitative learning by investors. Learning is driven by relative performance. In addition, we allow a cautious slow learning pace to reflect institutional conditions. Imitative learning shapes the market ecology that influences price formation. Using the model of non-imitative agents as a benchmark, our results show that the dynamics of imitative learning endogenously induce a significant degree of asset dependency and patterns of non-constant correlation. The asymmetric learning effect on correlation, however, implies a self-reinforcing process, where a bearish condition amplifies the effect that further exacerbates asset dependency. We conclude that imitative learning, even when rational, can to a certain extent account for the phenomena of market crashes. Our results have implications for transparency in regulation issues.