基于EEMD的金融时间序列多尺度分析

发布时间：2018-05-07 12:14

本文选题：金融时间序列 + 多尺度　；参考：《中国科学技术大学》2016年硕士论文

【摘要】：近年来,随着我国金融市场的发展以及投资品种的日益丰富,量化投资逐渐步入人们的视野,同时也将金融计量分析推至前所未有的高度。时间序列作为金融市场中最常见的观测数据,是时市场行为的真实刻画,通过对其进行量化分析,能够挖掘市场中潜在的信息,进而为投资决策提供理论和技术支撑,是策略制定、风险管理、资产定价和产品设计等工作的前提。基于金融时间序列的多尺度、非线性和非平稳和的多重特性,本文将集成经验模态分解(EEMD)应用到金融时间序列分析中。首先,利用EEMD建立多尺度集成预测模型。先用EEMD将原始序列分解并重构成高频、趋势和低频三个子序列；再结合Elman神经网络、支持向量机(SVM)和GM(1,1)对各部分进行拟合；集成模型最后的预测值为各部分预测值之和。实证结果表明：该多尺度集成模型的预测精度要显著高于传统的单一模型和集成模型。其次,利用EEMD探究股市与汇市的波动溢出效应。利用EEMD将股价和汇率序列分解并重构成高频(短期波动项)、。低频(中期波动项)和长期趋势三个波动成分,从时域和频域的双重视角来探究股市与汇市的波动溢出效应。在不同的波动层次上,分别结合Granger因果检验和时变Copula探究波动溢出的方向和强度。研究表明：短期波动项对原始序列波动的贡献最大；不同时间尺度上的波动溢出方向和强度是不同的。最后,利用EEMD去噪建立跨期套利策略。将价差信号看作趋势和噪声的叠加,用EEMD方法对价差序列进行消噪,参考均值回复策略的原理,结合价差在趋势周围的波动状况寻找套利机会,相比于传统的小波去噪方法,避免了小波基函数参数选择的这一难题。
[Abstract]:In recent years, with the development of our financial market and the increasing variety of investment, the quantitative investment has gradually stepped into the people's vision, at the same time, the financial econometric analysis has been pushed to an unprecedented height. As the most common observation data in the financial market, time series is the real depiction of the market behavior. Through the quantitative analysis of the time series, the potential information in the market can be excavated, thus providing theoretical and technical support for the investment decision. It is a prerequisite for strategy making, risk management, asset pricing and product design. Based on the multi-scale, nonlinear and non-stationary characteristics of financial time series, this paper applies the integrated empirical mode decomposition (EMD) to the analysis of financial time series. Firstly, the multi-scale integrated prediction model is established by using EEMD. First, the original sequence is decomposed by EEMD to form three sub-sequences of high frequency, trend and low frequency; then, combined with Elman neural network, support vector machine (SVM) and GM-1) are fitted to each part, and the final prediction value of the integrated model is the sum of the predicted values of each part. The empirical results show that the prediction accuracy of the multi-scale integrated model is significantly higher than that of the traditional single model and integrated model. Secondly, using EEMD to explore the volatility spillover effect of stock market and foreign exchange market. Using EEMD to decompose stock price and exchange rate sequence to form high frequency (short term fluctuation term). From the perspective of time domain and frequency domain, the volatility spillover effects of stock market and foreign exchange market are studied from the perspective of low frequency (medium term fluctuation term) and long term trend. At different volatility levels, the direction and intensity of volatility spillover are explored with Granger causality test and time-varying Copula respectively. The results show that the short-term fluctuation term has the greatest contribution to the fluctuation of the original series, and the direction and intensity of the volatility spillover are different in different time scales. Finally, the cross-period arbitrage strategy is established by using EEMD denoising. The spread signal is regarded as the superposition of trend and noise, and the spread sequence is de-noised by EEMD method. Referring to the principle of mean recovery strategy and combining the fluctuation of price difference around the trend, the arbitrage opportunity is found, compared with the traditional wavelet de-noising method. The problem of parameter selection of wavelet basis function is avoided.
【学位授予单位】：中国科学技术大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：F224;F832.51

【相似文献】