融合多源数据的股指预测研究

金融市场中股票价格的变动受到多方面因素的影响，如何更好地利用更多的大数据为投资决策进行服务始终是各方研究的重点。本研究以沪深300指数为研究对象，采用图像化处理的方式融合财经新闻、市场交易数据和技术指标等多源异构数据，建立了卷积神经网络模型，对未来不同时间长度的股指图像数据进行涨跌预测。通过对模型结构的稳健性检验，使用60天融合新闻情绪、技术指标与股价三类图片的三层图片预测模型，预测股指未来5天后的涨跌样本外准确率可达65.2%。融合多源数据后的图片数据能够丰富单一的股价数据，从而提升了模型的预测准确率。通过与传统的线形模型、LSTM循环神经网络模型及其他经典卷积神经网络模型比较，本研究构建的预测模型在样本外预测效果最佳，表明本研究构建的基于多源异构数据的图片预测模型在股指预测中具有可行性和一定优势。

Research on Stock Index Prediction Based on Multi-Source Data

The changes in stock prices in the financial market are affected by many factors. How to better use the current rich and diverse big data to serve investment decisions has always been the focus of research. This study took the HS300 Index as the research object and used image processing to integrate multi-source heterogeneous data including financial news, market transaction data and technical indicators. By establishing a convolutional neural network model, the stock index image data of different time lengths in the future was predicted to rise and fall. Through the robustness test of the model structure, using a 60-day three-layer picture prediction model integrating news sentiment, technical indicators and stock price pictures, the out-of-sample accuracy rate of predicting the stock index''s rise and fall in the next five days reached 65.2%. The image data after the fusion of multi-source data could enrich a single stock price data, thereby improving the prediction accuracy of the model. Compared with the traditional linear model, LSTM recurrent neural network model and other classic convolutional neural networks, the prediction model constructed in this study has the best prediction effect outside the sample, which shows that the image prediction model based on multi-source heterogeneous data constructed in this study is in the stock index. There are feasibility and advantages in forecasting.