Exchange-Traded Fund Price Prediction Based on the Deep Learning Model

With the exchange-traded fund (EFT) expanding rapidly, more and more people begin to invest in it, due to its lower risk and ideal returns. Forecasting the price of ETFs can greatly lower the risk and increase their revenue. Similar to the prices of other financial assets, the price of ETF is affected by many factors. This paper uses two models to predict the highest price of an ETF of the next day. The daily trade data and eight indicators are candidate features. However, more features bring not only information but noise. So before building the model, the maximum information coefficient feature extraction algorithm is applied to choose the best features as the input of the model. The first model is based on LSTM, which is a class of neural networks for processing sequence data. The other one is CNN-BiLSTM-AM model. CNN can extract the information from the input, BiLSTM can find the laws hidden in time series, and AM focuses on the more important information. The models are used to predict the highest price of the next day for 700 trading days of Shanghai 50 exchange-traded securities investment fund. Compared with the ARIMA and DNN, LSTM has the highest accuracy. Using the model and the Kelly Criterion, this paper proposes a quantitative investment strategy. Simulate trading on the test data set of about three years and the net gain reaches 63.7%. Both the model and the trading strategies can provide investors with a good guide to maximize benefits.