State-dependent asset allocation using neural networks

Changes in market conditions present challenges for investors as they cause performance to deviate from the ranges predicted by long-term averages of means and covariances. The aim conditional asset allocation strategies is overcome this issue adjusting portfolio allocations hedge changes investment opportunity set. This paper proposes a new approach that based on machine learning; it analyzes historical states returns identifies optimal choice period when observations become available. In approach, we directly relate state variables weights, rather than firstly modeling return distribution subsequently estimating choice. method captures nonlinearity among (predicting) weights without assuming any particular other data, fitting model with fixed number predicting data parameters. empirical results stock bond indices show proposed generates more efficient outcome compared traditional methods robust using different objective functions across sample periods.