In this article, we bring forward a newly-made asset combination optimization method that combines Genetic Algorithms (GA) with Long Short-Term Memory (LSTM) networks for market prediction. LSTM networks are utilized by us to build a model that is based on past market changes, for the forecasting of the future earnings of assets. The forecasted numerical results of these targets help the GA to find out a portfolio which has more fitting allocation. This multi-objective method is able to assign the weights of risk and profit more effectively than one single-objective model. In this place, an adaptive genetic algorithm is put forward. It makes changes to the crossover and mutation operators on the basis of the fitness of the portfolio. Iterative methods can be utilized by people for the optimization of portfolio distribution. In the end, the model’s performance is evaluated by means of risk-adjusted profits, steadiness, and out-of-sample operational results. We then carry out a comparison between it and the traditional models which are like mean-variance optimization. We carry out backtesting work, therefore, in order to further make confirmation of the robustness and the generality of this optimized combination of investment. Through making use of LSTM forecast and GA optimization method, an optimal portfolio may be obtained. Therefore, the function and adaptive ability of this novel method are better than what the traditional method has.
