Optimal Portfolio Model based on LSTM Neural Network and Markovitz Theory

Asset price forecasting is essential for portfolio decision-making. This paper establishes an asset price prediction model based on LSTM neural network to achieve asset price prediction. First, the historical asset price dataset is used as the training set of the model, and this paper set two hidden layers with 50 and 80 neuron units, respectively. Second, the Adam optimizer is used for the second hidden layer to optimize the neural network and minimize the loss function. Finally, the output layer data of asset price prediction is obtained considering the environment and other factors to achieve accurate price prediction. Meanwhile, this paper constructs a Markowitz-Dynamic programming model based on Markowitz and dynamic programming theories. It uses the output data cost of the prediction model to establish optimal portfolio planning, optimize portfolio decisions, and maximize investment returns. The model shown in this paper has significant reference value for investors' portfolio decisions and is essential to help investors obtain higher investment returns to a greater extent.