支持向量机择时的Alpha套利研究

发布时间：2018-01-30 20:52

本文关键词： 支持向量机 Alpha套利交叉验证遗传算法粒子群算法　出处：《首都经济贸易大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着经济持续发展,国民财富积累,如何让自身资产保值升值越来越为民众重视。一方面由于国内外经济的波动,金融市场规则的不完善,我国金融市场的剧烈波动时有发生,因此,Alpha策略作为一种本质上能够规避系统风险的投资方式近年来备受投资者青睐。另一方面传统Alpha策略存在明显缺陷,长期双边持仓风险敞口较大,而且一旦遭遇金融市场急剧波动,作为对冲工具的股指期货与现货沪深300指数协整关系骤然减弱,这时传统双边Alpha策略不仅不能对冲系统风险,反而会加大双边亏损的程度。考虑到在市场行情已知的情况下Alpha套利策略往往有较好的表现事实,本文研究如何通过市场行情走势的预判断,减少双边持仓的持有期,缩小风险敞口,扩大了盈利能力,以弥补上述传统Alpha策略缺陷的问题。首先是市场行情走势预判断研究。本文选取2006年1月1日至2015年12月31日沪深300指数的开盘价、最高价、最低价、成交额、成交量、涨跌幅和振幅七个指标,运用机器学习中的支持向量机技术对沪深300指数未来五个交易日的走势进行预测。并且运用了交叉验证、遗传算法和粒子群算法方法,改进了其中的参数估计问题。在参数优化基础上,根据平均平方误差和平方相关系数指标比较了线性核函数、多项式核函数、径向基核函数和Sigmoid核函数等四种核函数的预测表现情况,发现交叉验证和遗传算法优化参数时选择径向基核函数预测市场行情走势效果较好,粒子群算法优化参数时选择线性核函数预测市场行情走势效果较好。其次是,Alpha套利组合构造方法研究。本文选取深市129只股票2006年1月1日至2010年12月31日的数据,将收盘价、总市值、净利润、营业收入、成交量、换手率年度指标换算成收益率、市盈率、市销率、波动率、换手率、流动性等六个因子,用面板二值选择模型,寻找股票中超越市场指数作用显著的因子。发现波动率、换手率和市盈率这三个因子作用显著。因此本文根据这三个因子对129只股票加权打分排序,选取其中排名前30股票根据不同市场行情构造不同Alpha套利组合。策略是,如果预判断股票走势上涨,则Alpha套利只持有多头股票组合;如果下跌,则同时持有多头股票组合和空头股指期货。最后是套利效果实证研究。本文分别使用了2011年1月1日至2014年6月31日和2014年7月1日至2015年12月31日的数据(前者波动不大,后者波动较大),从多个角度比较了本文方法和传统方法的市场表现,发现前者明显优于后者,且市场波动不大时用市场强度指标预判的套利效果好,反之则是支持向量机的效果好。另外,本文只考虑选取一种核函数方法,如果能够采取集成学习方法对不同核函数预测结果加权,则可能取得更好的套利效果。
[Abstract]:With the sustained development of the economy and the accumulation of national wealth, how to keep and appreciate the value of its assets has become more and more important to the public. On the one hand, due to the fluctuation of domestic and foreign economy, the financial market rules are not perfect. The violent fluctuation of our country's financial market occurs from time to time, therefore. As a kind of investment mode which can avoid system risk in essence, Alpha strategy is favored by investors in recent years. On the other hand, the traditional Alpha strategy has obvious defects. Long-term bilateral position exposure is large, and once the financial market experienced sharp fluctuations, as a hedge tool of stock index futures and spot CSI 300 index cointegration relationship suddenly weakened. At this time the traditional bilateral Alpha strategy not only can not hedge against system risk. On the contrary, it will increase the degree of bilateral losses. Considering that Alpha arbitrage strategy often has a good performance in the case of known market situation, this paper studies how to pre-judge the market trend. Reducing the holding period of bilateral positions, reducing risk exposure, and expanding profitability. In order to make up for the defects of the traditional Alpha strategy mentioned above, the first is the pre-judgment study of the market trend. This paper selects the opening of the CSI 300 Index from January 1st 2006 to December 31st 2015. Price. The highest price, the lowest price, the turnover, the trading volume, the rise and fall and the amplitude seven indicators. Using the support vector machine technology in machine learning to predict the trend of CSI 300 index in the next five trading days, and using cross-validation, genetic algorithm and particle swarm optimization method. On the basis of parameter optimization, linear kernel function and polynomial kernel function are compared according to mean square error and square correlation coefficient index. The prediction performance of four kinds of kernel functions, such as radial basis kernel function and Sigmoid kernel function, is found to be better when selecting radial basis kernel function to predict market trend when cross-validating and genetic algorithm optimizing parameters. The linear kernel function is used to predict the market trend in PSO. The second is the linear kernel function. Alpha arbitrage combination construction method. This paper selects the data of 129 stocks from January 1st 2006 to December 31st 2010, the closing price, total market value, net profit. The annual index of operating income, turnover and turnover rate is converted into six factors, such as yield, price-earnings ratio, market sales rate, volatility rate, turnover rate, liquidity and so on, and the panel binary-value selection model is used. It is found that volatility, turnover and price-earnings ratio play a significant role in the stock market. Therefore, this paper gives a weighted ranking of 129 stocks according to these three factors. Select the top 30 stocks according to different market prices to construct different Alpha arbitrage portfolio. The strategy is that if the stock advance, Alpha arbitrage only holds long stock portfolio; If it falls. The last is the empirical study of arbitrage effect. This paper uses January 1st 2011 to June 31st 2014 and July 1st 2014 to 2 respectively. Data for December 31st 2005 (. The former fluctuates little. The latter fluctuates a lot, comparing the market performance of this method and the traditional method from many angles, it is found that the former is obviously better than the latter, and when the market fluctuation is not big, the arbitrage effect of pre-judging with the market intensity index is good. On the other hand, this paper only considers the selection of a kernel function method. If the integrated learning method can be used to weight the prediction results of different kernel functions, better arbitrage effect may be obtained.
【学位授予单位】：首都经济贸易大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F832.51;F224

【相似文献】