基于支持向量机的高频量化投资研究

发布时间：2018-05-16 15:31

本文选题：支持向量机 + 量化投资　；参考：《山东科技大学》2017年硕士论文

【摘要】：量化投资在国外已经发展了三十多年,由于投资方式基于模型化交易,业绩比较稳定,在国外对冲基金投资策略中占据了很大的份额。量化投资的本质就是模型化交易。量化的过程是对交易行为进行建模,对交易变量进行参数量化,一旦市场触发了条件,就会进行交易。Corinna Cortes和Vapnik在1995年的时候首先提出了支持向量机理论,它是基于统计理论的机器学习算法,由于支持向量机指导基础是扎实的统计学理论,不会收敛于局部最优解。基于支持向量机适合解决非线性的、小样本的数据问题,在高维模式识别问题也有许多独特的优势,本文基于股指期货IF高频TICK数据进行回归预测,从以下几个方面进行了研究:一方面考虑到目前为止在核函数的选取甚至于有针对性地构造核函数还没有较好的理论指导,为了更好得研究解决支持向量机核函数参数问题,本文选取R]BF核函数进行研究,把重心放在了核函数参数g还有惩罚因子C的优化选择上,用改进的网格法进行参数寻优,寻参效果较好,同时减少了参数寻优计算时间。另一方面是交易成本处理问题,高频交易除了量化模型很重要以外,交易成本也是影响结果的重要因素,尤其是买卖价差,手续费水平,因此本文通过建立数学模型,对买卖价差进行了较好的分析研究,建立了对数回归拟合模型,拟合效果极好,比较充分反映了实际的买卖价差成本,并在最终策略实现的时候导入了买卖价差成本进行减值,计算出了成本模型。本文最终基于买卖价差分析模型以及支持向量机预测模型建立了拟合分类预测模型,研究了高频量化投资在国内市场的IF合约下的策略构建可行性,并基于研究过程进行了实践作业,进行了策略实现,给出了不同市场下对高频策略影响的手续费评估阈值,通过实际效果验证了理论分析,具有实际指导意义。
[Abstract]:Quantitative investment has been developed for more than thirty years in foreign countries. Because of the model trading based on the investment mode, the performance is relatively stable, and it occupies a large share in the foreign investment strategy of hedge funds. The essence of the quantitative investment is model trading. The quantitative process is to model the transaction behavior and quantify the parameters of the transaction variables. The market triggered the conditions, and the transaction.Corinna Cortes and Vapnik first proposed the support vector machine theory in 1995. It is a machine learning algorithm based on statistical theory. Because the support vector machine guidance foundation is a solid statistical theory, it will not converge to the local optimal solution. The support vector machine is suitable for the non line solution. The problem of small sample data has many unique advantages in the high dimensional pattern recognition problem. This paper studies the regression prediction based on the IF high frequency TICK data of stock index futures. On the one hand, there is no good theory on the selection of the kernel function and even the targeted construction of the kernel function. In order to better study and solve the kernel function parameter problem of support vector machine, this paper chooses R]BF kernel function to study the kernel function of kernel function parameter g and penalty factor C. In addition to the cost handling problem, high frequency transaction is very important in addition to the quantitative model, and the transaction cost is also an important factor affecting the result, especially the sale price difference and the service fee level. Therefore, this paper establishes a logarithmic regression fitting model by establishing a mathematical model and establishes a logarithmic regression fitting model. It reflects the actual cost of the price difference, and introduces the cost of the sale price difference when the final strategy is realized, and calculates the cost model. Finally, based on the sale price difference analysis model and the support vector machine prediction model, the fitting classification prediction model is established, and the IF combination of the high frequency quantitative investment in the domestic market is studied. The proposed strategy is feasible, and the practical operation is carried out based on the research process. The strategy is implemented, and the evaluation threshold of the fee assessment for the high frequency strategy under different markets is given. The theoretical analysis is verified by the actual effect, which has practical guiding significance.

【学位授予单位】：山东科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F832.51;F224

【参考文献】