Empirical Research for Investment Model Based on VMD-LSTM

Abstract

Investment diversification has become an inevitable trend with the development of the world economy. In this work, we first compare the K-Nearest Neighbor model, the Artificial Neural Network model, the grey prediction model, and the LSTM (Long Short-Term Memory Networks) prediction model for a period of data analysis. The experimental results show that LSTM is superior, and thus LSTM model is selected to forecast the long-term prices in this work. Then, we introduce some indicators, such as convergence divergence ratio and risk coefficient to qualitatively analyze the market price. The five-day moving average method is used to formulate the best trading strategy based on the above-introduced indicators. We apply the commonly used regression indicators ( R 2 and R M S E ) to verify the reliability of the prediction model. Then we introduce new strategies to compare the performance of different ones with them. We found that the five-day moving average method achieved 20% higher returns than the other strategies we used for comparison. Considering the fact that transaction costs may change, we perform the polynomial fitting based on existing strategies by changing the commission cost. The results show that a 1% increase in the gold commission will reduce the total return by 24% to 25%, while a 1% increase in the bitcoin commission will only reduce the total return by 7% to 8%.