Long Short-Term Memory Neural Network Forecast Model Based on Improved Particle Swarm Optimization Algorithm

Volatile assets often attract market traders in trading markets. In order to better help market traders, allocate the asset portfolio reasonably based on the asset price data of the day to maximize the asset income in the future. We build a Revenue-Return Strategy Analysis Model based on LSTM that gives traders the best combination of trades based on the day's price data and the price data of the past few days. The model includes strategy analysis, risk assessment, growth rate forecast, optimization algorithm four parts. We first discussed the weekly and monthly growth rates of bitcoin and gold. Then we use the VAR model and Granger causality test to find that there is no Granger causality between gold and bitcoin. Finally, according to the price of the day, future price growth rate, and weight coefficient, we get the formula of profit. When the profit is the largest, the trading portfolio is the best strategy of the day. We analyze the risk of past price fluctuations on future returns according to Markowitz's risk theory. We get the weight coefficient needed in the formula of the return value. Then we use LSTM training and testing data to predict the price trend of the future days. We substitute the weight coefficient and future price growth rate into the profit value formula and use particle swarm optimization algorithm to get the maximum profit value.