Asset Trading Strategies Based on LSTM - Bitcoin and Gold as an Example

Inhalt:
At the moment, how to invest is a question that everyone has to encounter. Thus, this paper constructs an investment model and helps investors to adopt a reasonable investment strategy. Firstly, to determine an investment strategy you must first make a forecast of asset movement. We have taken a variety of forecasting models using Bitcoin and gold daily data as examples. LSTM GRU and ARIMA are established and trained with the past price data, and these models are useful in predicting the trend of time series data. The LSTM obtains the optimal prediction model with MAE equal to 0.04. Secondly, we consider the return and risk indicators in discussing the best trading strategy. To quantify the risk indicator, GARCH, a risk scoring system, was used to perform risk assessment, and the return indicator is quantified using the predicted rate of return. Then Sharpe Ratio is constructed from the income and risk indexes based on the idea of maximization of return per risk, which is rational since every investor must balance the relationship between return and risk in portfolio construction. Further, PSO is used to search for optimum, which reached 27 263 in 9/10/2021 value, equivalent to an annualized yield of 93.68% with an overall VAR value of 502.51. To test the effectiveness of our strategy, we adopted Data Envelopment Analysis, a multi-stage portfolio evaluation model to prove the effectiveness of the trading strategy. Finally, we did a sensitivity analysis of the results. This process saves investors from different cost functions and highlights the best strategy.