基于LSTM-ARIMA模型股票预测研究

Research on Stock Forecasting Based on LSTM-ARIMA Model

摘要: 随着理财观念的不断深化，股票作为金融资产在资本市场的投资价值逐渐显现。因此，对股票价格的预测越来越成为当下专家学者的研究重点。其中，股价涨跌幅趋势的研究能够帮助投资者制定个性化选股策略，从而提高可行性、降低风险，以此达到投资收益率最大化。本文选取不同领域的6支股票进行分析，经过相关性分析和熵权法权重分析后确定收盘价作为股价评价指标，选用2018年2月2日至2022年3月30日收盘价时序列数据数据建立LSTM神经网络进行长期股价走势分析，选用2021年6月1日至2022年3月30日收盘价建立ARIMA模型进行短期股价走势分析，结合拟合值、真实值和模型预测误差，结果显示预值和真实值相差不大。由测试结果可以得出结论，LSTM神经网络模型对于长期时间序列数据预测结果拟合精度高，ARIMA模型对于短期走势走势拟合程度高。因此，结合LSTM神经网络模型和ARIMA模型模型可以对长短期股价进行预测分析，能够得到一个较为精确的预测走势。

Abstract: With the deepening of managing money matters, the investment value of stock as a financial asset in the capital market gradually appears. As a result, stock market forecasting began to converge on the research of experts and scientists. Among them, Research on the upward and downward trend in stock prices can help investors develop a personalized stocks election strategy, so as to improve the feasibility and reduce the risk, so as to maximize the return on investment. In this paper, six stocks in different fields are selected for analysis. After correlation analysis and entropy weight analysis, the closing price is determined as the stock price evaluation index. LSTM neural network is used to establish the sequence data of the closing price from February 2, 2018 to March 30, 2022 for long-term stock price trend analysis. The ARIMA model is used for short-term stock price trend analysis based on the closing price from June 1, 2021 to March 30, 2022. Combined with the fitting value, the real value and the model prediction error, the results show that there is little difference between the forecast value and the real value. It can be concluded from the test results that LSTM neural network model has a high fitting accuracy for long-term time series data prediction results, and ARIMA model has a high fitting degree for short-term trends. Therefore, LSTM neural network model and ARIMA model can be combined to predict and analyze long-term and short-term stock prices, and a more accurate fluctuation trend can be obtained.