In this Paper we develop a model of intertemporal portfolio choice where an investor accounts explicitly for the possibility of model misspecification. This work is motivated by the difficulty in estimating precisely the probability law for asset returns. Our contribution is to develop a framework that allows for ambiguity about the joint distribution of returns for all stocks being considered for the portfolio, and also for different levels of ambiguity for the marginal distribution of returns for any subset of these stocks. We then use this framework to derive in closed-form the optimal portfolio weights of an investor who accounts for model misspecification. We illustrate the model by calibrating it to data on international equity returns. The calibration shows that when the overall ambiguity about the joint distribution of returns is high, then small differences in ambiguity for the marginal return distribution will result in a portfolio that is significantly under-diversified relative to the standard mean-variance portfolio.
