Application of LSTM in Quantitative Trading with Big Data

Recently, the global market returns are characterized by year-long episodes of high volatility, leading market traders to struggle to maximize their total return with high-frequency trading. However, in the meantime, some classical financial characteristics have striking similarities. Therefore, an increasing amount of investors are dedicated to finding and making the most of reasonable models with computer science and machine learning, in quest of a general pattern. Accordingly, we developed a comprehensive model to help some traders with optimal transaction decisions. Given that the portfolio of cash, bitcoin and gold is gradually becoming a new choice for high returns, we chose these assets as examples to build a model to forecast the closing prices of bitcoin and gold for the next trading day based on their past stream up to today with LSTM (LongShort Term Memory) Neural Network Model. To achieve the maximum return, we used the Mean-variance Model to set the investment ratio of bitcoin and gold as variables for mathematical programming and solved the multivariate equations to find the optimal portfolio. The investment weights for each trade were obtained by comparing the lowest risk portfolio and the optimal return portfolio, i.e., the minimum variance and maximum Sharpe ratio scenarios. Finally, we evaluated our strategy through quantitative strategy indices and proved its outstanding performance.