摘要
为了提高汇率预测精度,本文创新性地将深度学习方法 GRU神经网络应用于欧元汇率预测,进一步通过加入百度指数数据改进预测模型。研究结果表明:GRU神经网络相比传统机器学习方法和经典深度学习方法能更精准地预测汇率;将百度指数为代表的互联网搜索行为数据应用于汇率预测模型有助于提升预测准确度;GRU神经网络对于预测步长并不敏感。此研究表明GRU神经网络可以对外汇预测管理提供重要参考,在外汇市场中具有较大应用价值。
In order to apply deep learning methods to the financial economy, the GRU neural network is used to predict EUR/USD, and then introduce the Baidu index into the model for prediction. The research results show that the GRU neural network can predict the exchange rate more accurately than the traditional machine learning method and the classical deep learning method. The application of the Internet search behavior data represented by the Baidu index to the exchange rate prediction model can improve the prediction accuracy. GRU neural networks are not sensitive to the change of historical days.This research shows that GRU neural network can provide an important reference for foreign exchange forecasting management and has great application value in foreign exchange market.
出处
《浙江金融》
2019年第3期12-19,28,共9页
Zhejiang Finance
基金
上海市哲学社会科学规划项目"欧洲央行非常规货币政策的非对称溢出效应研究"(项目编号:2017EJL001)
2018年上海高校青年教师培养资助项目(项目编号:ZZslg18015)
2018年上海理工人文社科攀登计划(编号:SK18PB04)