沪铜期贷价格预测模型的构建与预测研究

发布时间：2018-06-29 05:16

本文选题：沪铜期货 + 价格预测　；参考：《兰州交通大学》2017年硕士论文

【摘要】：随着中国市场经济地位和人们理财意识的提高,期货交易成为金融交易以及金融衍生品交易中的一个重要交易品种,期货市场的健康稳定发展也成为管理者和投资者研究的重点。不管是利用期货市场进行投资还是投机,做好风险控制是尤为重要的。而做好风险控制的前提就需要对期货价格进行预测,在预测分析之后制定相应的交易原则,按照交易原则进行风控处理。目前,随着现代科技的不断进步与发展,期货的预测方法也越来越多,以统计学中的方法为例,有时间序列预测模型、灰色预测模型、神经网络预测模型等等。本文首先选取了其中的ARIMA模型、GARCH模型以及BP神经网络模型三种单一的预测模型对沪铜期货价格进行预测。选取沪铜主连合约从2015年1月5日到2015年9月25日共180个交易日的收盘价数据作为研究对象,其中2015年1月5日到2015年8月28日的沪铜主连合约收盘价数据用于拟合估计模型,剩余的数据用于预测结果的对比分析。实证的结果表明:BP神经网络模型的累计相对误差值和MAPE值都比ARIMA模型和GARCH模型要小,说明BP神经网络模型是这三种模型中预测精度最高的。这主要源于BP神经网络模型有着强大的自学能力,它可以通过训练学习找到参数之间的规律和特点,掌握数据间的依存关系。其次,根据前面三个单一预测模型的预测效果,本文又在此基础上优化得到了两个组合预测模型,即最优权重线性组合预测模型和基于BP神经网络的组合预测模型,并用这两个组合预测模型同样做了实证分析,得到的结论是:两种组合预测模型相比单一的预测模型,都在一定程度上提高了预测精度;在这两种组合预测模型中,基于BP神经网络的组合预测模型具有更高的预测精度,预测效果更好。最后,考虑到不同模型对数据的适用性不同,本文又选取了2016年10月10日到2016年11月18日共30个交易日的沪铜主连收盘价进行短期时间跨度预测研究,对比180个交易日的长期时间跨度的预测结果,来探究各模型在沪铜期货价格预测中的适用条件和范围。结果表明,基于BP神经网络的组合预测模型无论是在短期时间跨度还是长期时间跨度预测中都有着很好的预测效果。
[Abstract]:With the development of China's market economy status and people's awareness of financial management, futures trading has become an important trading variety in financial transactions and financial derivatives transactions. The healthy and stable development of futures market has also become the focus of managers and investors. Whether using futures market for investment or speculation, risk control is particularly important. The premise of risk control is to forecast the futures price. After forecasting and analyzing, the corresponding trading principles should be worked out, and the wind control should be carried out according to the transaction principles. At present, with the progress and development of modern science and technology, there are more and more forecasting methods of futures, such as time series forecasting model, grey forecasting model, neural network forecasting model and so on. This paper first selects Arima model GARCH model and BP neural network model to forecast Shanghai copper futures price. The closing price data of 180 trading days from January 5, 2015 to September 25, 2015 are selected as the research objects. The closing price data of Shanghai Copper main Company contract from January 5, 2015 to August 28, 2015 are used to fit the estimated model. The remaining data are used for comparative analysis of the predicted results. The empirical results show that the cumulative relative error and MAPE of the BP neural network model are smaller than those of Arima model and GARCH model, indicating that the BP neural network model is the most accurate of the three models. This is mainly due to the strong self-learning ability of BP neural network model, which can find the rules and characteristics of parameters through training and learning, and grasp the dependence of data. Secondly, according to the prediction effect of the first three single prediction models, this paper optimizes two combined forecasting models, that is, the optimal weight linear combination prediction model and the BP neural network based combination forecasting model. The two combined forecasting models are also used for empirical analysis. The conclusion is that the two combined forecasting models have improved the prediction accuracy to some extent compared with the single prediction model; in these two combined forecasting models, the two combined forecasting models have improved the prediction accuracy to a certain extent. The combined prediction model based on BP neural network has higher prediction accuracy and better prediction effect. Finally, considering the different applicability of different models to the data, this paper also selects 30 trading days from October 10, 2016 to November 18, 2016 to carry out short-term time span prediction research on the closing price of Shanghai Copper main Line. By comparing the forecasting results of 180 trading days with a long time span, this paper probes into the applicable conditions and scope of each model in Shanghai copper futures price forecasting. The results show that the combined prediction model based on BP neural network has good prediction effect in both short and long term time span prediction.
【学位授予单位】：兰州交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F224;F724.5;F764.2

【参考文献】