A Two-stage Model: Investment Market Trading Model Based on Prediction and Adaptive Strategies

Abstract

Market traders buy and sell volatile assets frequently, with a goal to maximize their total return. Two such assets are gold and bitcoin. This paper constructs a two-stage model for price prediction and trading strategy formulation. Firstly, we build a dynamic ARIMA-LSTM hybrid model. And before applying it, we use CEEMDAN method to decompose the non-stationary time series first and then reconstruct the final result by predicting each IMF and summing up weighted. And then the model can update the training set dynamically when new price data is released. After getting prices predicted, we calculated several quantitative trading indicators so that we can make decisions more comprehensively instead of only focusing on the predicted price. And we use semi-supervised SVM to develop an adaptive strategy to maximize the total return. Finally, we demonstrate the superiority of our strategy from two perspectives. In actual investment transactions, the two-stage model can be used as a guide for the formulation of trading strategies, thereby avoiding risks and increasing returns.