Algorithm Optimization Model of Trading Strategy based on CEEMDAN-SE-LSTM and Artificial Intelligence

Abstract

The key challenges of the financial industry are the volatility and complexity of the stock market, so how to make optimal trading strategy to maximize the total profit in all market conditions has become an important issue to the professional researchers and investors. This paper describes a hybrid stock trading strategy model based on long short-term memory (LSTM) networks. The Complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) algorithm and sample entropy (SE), combined with LSTM, are used to construct the integrated prediction model, which has dramatically improved the forecast precision. On the premise of accurate prediction, the extreme value theory (EVT) is introduced to improve the predictive ability of dynamic value at risk (VaR), which can manage the risk of portfolio. To forecast stock trends, the approach of analytic hierarchy process (AHP) is applied to assign weights to related factors. The final trading decisions are made by establishing trading signals and scoring models. Based on models above, the integrated trading strategy model is constructed as an automated trading decision tool. Taking Gold and Crude oil as examples, the profit results are proved to be decent through trading simulations.