In this work, an extension of the neural network model introduced by Li and Forsyth (2019) is proposed. This extension allows for a data-driven solution of two types of dynamic portfolio allocation problems: a general asset allocation problem where the neural network model directly chooses the asset allocation, and a problem where it determines deviations from the allocation of a pre-determined benchmark strategy. The neural network model is applied to the problem of nding deviations from the 1/N-Strategy in the case of two assets on a single path of historical return data for an investor whose utility is represented by a power utility function. Across several levels of risk-aversion, we nd that the neural network model does not manage to consistently outperform the 1/N-Strategy or other simple constant proportion strategies in terms of average utility. Considering a di erent score for model evaluation the model consistently performs better than the constant proportion strategies for lower levels of risk-aversion.      «

In this work, an extension of the neural network model introduced by Li and Forsyth (2019) is proposed. This extension allows for a data-driven solution of two types of dynamic portfolio allocation problems: a general asset allocation problem where the neural network model directly chooses the asset allocation, and a problem where it determines deviations from the allocation of a pre-determined benchmark strategy. The neural network model is applied to the problem of nding deviations fro...     »