函数型数据分析方法在股票价格预测上的应用

发布时间：2018-03-05 18:31

本文选题：价格预测　切入点：函数型数据　出处：《厦门大学》2014年硕士论文　论文类型：学位论文

【摘要】：股票价格的预测模型可以帮助交易者做更好的交易策略,但是因为股票价格受到很多方面因素的影响,要建立一个合适的模型去模拟股票价格的波动是不容易的。时间序列模型在预测方面的应用受到了广泛的认可,因此,以往有很多研究股票价格预测的文章都用到了时间序列模型。但是很多时间序列模型,如著名的ARIMA模型,通常需要假设样本数据本身或变换后有平稳性和线性性,这一假设并不一定能满足,因此人们要寻找新的模型,可以适用于更宽松的假设,减少由于不满足模型假设而引起预测结果出现较大偏差的情形。文献上有Wang and Leu(1996)利用神经网络方法来建模并进行非线性拟合和预测,但这类模型多用于低频数据的分析。随着金融市场的迅速发展,证券市场中的交易越来越频繁,交易量也越来越大,相应的证券价格的变化也越来越频繁,所以传统的用低频数据来做证券市场的研究已经很难满足市场发展的需求,人们开始转向对时间间隔更小而数据量更大的高频数据的研究。高频的股票交易数据蕴含更丰富的信息,因而对建模选用的模型的灵活性和适用性有更高的要求。结合自回归模型和非参数回归思想,本文提出了一个新的混合模型以预测未来股票的开盘价。该模型的自回归部分,反映了过去开盘价的信息,该模型的非参数部分,是对前一个交易日的日内交易价格与一个未知函数作积分所得。由于利用了前一个交易日的日内交易价格的综合信息,故有望能提高我们对未来股价的预测能力。我们对混合模型中的未知函数不作任何参数形式的设定。通过对日间交易价格进行函数型主成分分析,我们可以巧妙地拟合混合模型非参数部分。最后我们用沪深300指数的数据进行实证分析。分析结果显示,本文提出的混合模型相比于传统的自回归模型有更好的预测表现。
[Abstract]:The forecasting model of stock price can help traders to make better trading strategy, but because the stock price is affected by many factors, It is not easy to establish a suitable model to simulate the fluctuation of stock price. The application of time series model in forecasting is widely accepted, so, In the past, many researches on stock price forecasting used time series models, but many time series models, such as the famous ARIMA model, usually need to assume that the sample data itself or the transformed data are stable and linear. This assumption is not necessarily satisfied, so people are looking for new models that can be applied to more relaxed assumptions. In the literature, Wang and Leuer (1996) uses neural network method to model and carry out nonlinear fitting and prediction. With the rapid development of the financial market, the transactions in the securities market are more and more frequent, the trading volume is also increasing, and the corresponding securities prices are changing more and more frequently. Therefore, the traditional use of low-frequency data to do securities market research has been very difficult to meet the needs of market development. People begin to study the high-frequency data with smaller interval and larger amount of data. The high-frequency stock trading data contain more information, so the flexibility and applicability of the models used in modeling are higher. Combined with the idea of autoregressive model and nonparametric regression, this paper presents a new mixed model to predict the opening price of future stocks. The autoregressive part of the model reflects the information of the past opening price, and the non-parametric part of the model. Is obtained by integrating the intraday trading price of the previous trading day with an unknown function. As a result of the use of comprehensive information on the intraday trading price of the previous trading day, It is expected to improve our ability to predict future stock prices. We do not set the unknown functions in the mixed model in any parameter form. We can fit the non-parametric part of the hybrid model skillfully. Finally, we use the data of the Shanghai and Shenzhen 300 index to carry on the empirical analysis. The results show that the hybrid model proposed in this paper has better prediction performance than the traditional autoregressive model.
【学位授予单位】：厦门大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：F224;F832.51

【参考文献】