极限学习机新华富时A50股指期货交易中的应用

发布时间：2018-03-09 08:30

本文选题：新华富时A50股指期货　切入点：极限学习机　出处：《深圳大学》2017年硕士论文　论文类型：学位论文

【摘要】：新华富时A50股指期货是在中国境外上市以中国A股市场股票为标的金融衍生品。同国内的指数期货相比具有较低的交易成本、高杠杆、高流通性等特点,是金融活动参与者进行风险管理及投资的重要金融工具。由于A50股指期货受多种复杂因素的影响,以及时间序列自身非线性、动态、高噪声等特性,运用传统金融时间序列研究方法无法逼近其内在运动规律,随着计算机科学、人工智能、大数据的发展,基于统计理论的机器学习算法广泛应用到时间序列的分析中并取得显著成效。本文主要目的是研究极限学习机在新华富时A50股指期货交易中的应用,由于极限学习机具有良好的非线性逼近能力、运算速度快、结构简单等特点,因此本文选用极限学习机为主要建模工具,本文样本数据为2013年7月26日至2016年9月23日期间的775个交易日的A50期货指数收盘价数据,其中前600个数据作为训练集,后175个数据为预测集。为了进行比较,本文选取四个投机策略(A1至A4)和两个对冲策略(B1和B2)进行计算和分析,其中A1和A2为一般的根据趋势技术指标进行的交易策略,A3和A4为根据极限学习机结果进行的交易策略。B1为根据一般对冲原则的交易策略,B2为根据模型结果进行波段操作的交易策略(具体说明见正文)。由于支持向量机也有广泛的应用,本文还选取支持向量机作为参照模型进行计算和比较。研究发现:(1)极限学习机在模型的结构、学习能力、预测精度等方面都要优于高斯核支持向量机。极限学习机模型的平方根均方误差、正则均方误差、平均绝对误差均小于高斯核支持向量机模型的对应误差。(2)极限学习机由于自身网络结构简单、运算速度快、泛化能力强等特点,较好的逼近新华富时A50股指期货每日收盘价的走势。预测值与实际值的误差最大相差0.5点。精准的预测有利于提高交易策略的胜率、盈利空间,同时降低价格波动带来的风险。(3)实证结果表明:无论是投机策略还是对冲策略,根据极限学习机模型结果建立的交易策略均优于其它交易策略,前者不仅能够有起到风险预警作用,还能大幅提高收益。在预测集中,四种投机策略收益率分别为21.84%、24.16%、140.79%、62.80%。其中A3策略的收益率分别是A1、A2策略的6.4、5.8倍。A4策略的收益率分别A1、A2策略的2.9、2.6倍,在对冲策略中,依据极限学习机模型结果的对冲策略B2的收益率是B1(依据一般对冲原则)7.1倍。由此可见,根据极限学习机模型结果而采取的交易策略有显著优势。
[Abstract]:Xinhua FTSE A50 stock index futures are financial derivatives listed outside China with Chinese A-share market shares. Compared with domestic index futures, they have the characteristics of low transaction cost, high leverage, high liquidity, etc. A50 stock index futures are influenced by many complicated factors, and the time series are nonlinear, dynamic, high noise and so on. With the development of computer science, artificial intelligence and big data, the traditional financial time series research method can not approach its internal motion law. The machine learning algorithm based on statistical theory is widely used in the analysis of time series and has achieved remarkable results. The main purpose of this paper is to study the application of extreme learning machine in the trading of stock index futures of Xinhua FTSE A50. Because extreme learning machine has good nonlinear approximation ability, fast operation speed, simple structure and so on, this paper chooses the ultimate learning machine as the main modeling tool. The sample data in this paper are A50 futures index closing price data for 775 trading days from July 26th 2013 to September 23rd 2016, of which the first 600 data are training sets and the last 175 data are forecast sets. In this paper, four speculative strategies (A1 to A4) and two hedging strategies (B _ 1 and B _ 2) are selected for calculation and analysis. Where A1 and A2 are general trading strategies based on trend technical indicators. A3 and A4 are trading strategies based on extreme learning machine results. B1 is trading strategy based on general hedging principle. Trading strategies for segment operations (see text for details.) because support vector machines are also widely used, This paper also selects support vector machine as the reference model to calculate and compare the structure and learning ability of the ultimate learning machine. The prediction accuracy is better than that of Gao Si kernel support vector machine. The square root mean square error, regular mean square error of extreme learning machine model, The average absolute error is smaller than the corresponding error of Gao Si kernel support vector machine model. Better approach to the trend of the daily closing price of the Xinhua FTSE A50 stock index futures. The biggest difference between the predicted value and the actual value is 0.5 points. Accurate prediction is helpful to improve the winning rate of trading strategy and profit space. The empirical results show that both speculative and hedging strategies are superior to other trading strategies based on the results of extreme learning machine model. The former can not only serve as a warning of risks, but also significantly increase returns. The return rate of the four speculative strategies is 21.84 / 24.16 / 140.79 / 62.80 respectively. The yield of A3 strategy is 6.4or 5.8 times that of A1A _ 2 strategy. The return rate of A4 strategy is 2.9or 2.6 times that of A _ 1 / A _ 2 strategy, respectively. The yield of the hedging strategy B2 based on the LLM model is 7.1 times that of B1 (according to the general hedging principle). It can be seen that the trading strategy based on the LLM model has significant advantages.
【学位授予单位】：深圳大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F724.5

【相似文献】