Forecasting of Stock Price Trend Based on CART and Similar Stock

Most of the existing literature on stock price analysis is based on the historical price of a single stock, where models are trained to identify the pattern of price movements and, predict future stock the prices. However, the stock price is not an isolated variable. It is correlated with and influenced by many factors. These factors are highly time dependent, it which means that the stock price movement is also time dependent. Therefore, existing prediction models often fail in applications. In this paper, our model is not based on single stock. Instead, we study a class of stocks with similar historical price movements. We get the training and forecast data of the prediction model based on time series window sliding. We then train a CART (Classification and Regression Trees) and evaluate the model on testing data. Compared with the classical models for price movement forecast including the time series analysis model ARIMA (Autoregressive Integrated Moving Average) and the recursive neural network model LSTM (Long-Short Term Memory), our l empirical results support the effectiveness of the proposed method in the stock trend forecast.