基于符号时间序列分析的金融波动分析与预测

发布时间：2018-04-16 15:27

本文选题：符号时间序列直方图 + K-NN预测　；参考：《天津大学》2012年硕士论文

【摘要】：高频金融数据包含更多的市场信息，由于其在市场微观结构的实证研究方面的重要性而受到广泛关注。对高频金融波动的研究对股票估值、衍生产品定价、资产组合配置、风险管理、货币政策的制定等至关重要，传统分析方法针对具体的波动数据，，建立波动模型，本文则从不同的角度出发，分析与预测高频金融波动的整体模式。本文首先将符号时间序列分析方法与K-Nearest Neighbors(K-NN)算法相结合，提出了一种基于符号时间序列直方图的高频金融波动整体分布的预测方法。第一步将观测所得的时间序列变换为符号时间序列，利用符号序列直方图直观表示符号序列的分布，引入符号直方图时间序列的概念，采用K-NN算法得到下一个周期符号序列直方图的预测。在K-NN算法中，针对符号序列直方图的特点，提出以欧几里得范数，χ2统计量和相对熵作为选择邻居时的符号直方图序列相似度的度量方法，并利用系统自身的几何特性确定符号直方图序列的嵌入维数。其次，利用可以有效提取日内信息的“已实现”波动来度量高频金融时间序列的波动，首次使用具有鲁棒性的排列熵方法分析“已实现”波动序列的顺序模式、序列之间的广义同步，利用全概率理论，在已知历史“已实现”波动顺序模式的情况下，预测下一个交易日的“已实现”波动处于不同水平的概率。针对本文所提的方法，均以上证综指或深证成指5分时的高频数据检验了方法的可行性与有效性。结果表明直方图时间序列的预测所得结果整体误差均在可以接受的范围内，预测所得的分布与真实分布均值相同，但是方差较小；而基于排列熵方法分析时，发现这两个指数的“已实现”波动序列之间基本不存在广义同步，确定了它们的主要顺序模式，并基于主要顺序模式对“已实现”波动水平进行预测，结果显示主要顺序模式的条件顺序模式仍然占主要地位。
[Abstract]:High frequency financial data contains more information from the market, because of its importance in the empirical research of market microstructure and attracted widespread attention. The research on high frequency financial volatility of stock valuations, derivatives pricing, asset allocation, risk management, is the formulation of monetary policy, the traditional analysis method for wave specific data, establish fluctuation model, this article from a different perspective, the overall pattern analysis and prediction of high frequency financial volatility.
Firstly, the symbolic time series analysis method and K-Nearest Neighbors (K-NN) algorithm, presents a high frequency financial volatility of symbolic time series based on the histogram of the overall distribution forecasting method. The first step is to transform the time series of observed symbols for time series, using the visual representation of distribution symbol sequence histograms of symbol sequences, concept the introduction of symbolic time series prediction histogram, K-NN algorithm is used to get the next cycle of symbol sequence histograms. In K-NN algorithm, according to the characteristics of symbol sequence histograms, the Euclidean norm x 2, statistics and relative entropy as a measure method of symbol sequence similarity histogram when choosing neighbors, and using geometric characteristics the system itself to determine the embedding dimension histogram sequence of symbols. Secondly, use can effectively extract information on Realized Volatility To measure the high volatility of financial time series, the first use of permutation entropy method is robust analysis of realized volatility series order model, generalized synchronization between sequences, using the theory of probability, in the known history of realized volatility sequence model, forecast the next trading day "has been achieved" volatility in different levels of probability.
According to the method proposed in this paper, the high frequency data 5 points are the Shanghai Composite Index Shenzhengchengzhi or test the feasibility and validity of the method. The results show that the prediction results of the histogram of time series the overall error in the acceptable range, the income distribution prediction and the real distribution of mean the same, but the smaller variance analysis; permutation entropy based method, it was found that the two index of the realized volatility series does not exist between the basic generalized synchronization, and determine the main sequence pattern, and based on the main sequence model to predict the realized volatility level. The results showed that the main conditions of sequential mode sequential mode is still dominant.

【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：F224;F830.91

【参考文献】