基于中心度的标签传播时间序列聚类方法

为了实现时间序列自动聚类,以及更为细致地描述时间序列之间的结构关系,引入社区发现方法来研究时间序列聚类.针对标签传播方法在标签传播过程中具有较强不确定性,以及算法对网络结构较为敏感等问题,提出一种基于中心度的标签传播时间序列聚类方法;通过构建时间序列网络空间结构,将每条时间序列看作一个节点,根据每个节点的中心度来得到标签更新顺序;计算节点对于每个簇的归属度,再利用节点的归属度和标签的传播实现节点的划分,从而实现时间序列聚类.所提方法通过分析时间序列之间的连接关系来发现其在欧氏空间的结构特征,进而实现空间结构的有效划分.实验结果表明,所提方法无需确定初始簇中心,能够有效划分人工数据网络和真实社会网络,在时间序列数据聚类中取得了良好的聚类效果.     

基于相空间重构的股价时间序列相关性分析

股指的相关性研究对于衍生产品定价、风险管理、套期保值和最优投资组合选择等都具有重要的意义。股指相关性研究大都是在线性理论的基础上,即认为股指时间序列是线性的,但是实际上,股指时间序列具有很强的非线性特征,因此在线性理论基础上得到的相关性结果具有一定的局限性。应用非线性的混沌理论,通过对沪深股指混沌时间序列进行相空间重构,建立了一个多维的股指时间序列系统。运用典型相关分析对构建的系统进行相关性计算,得到沪深股指之间的相关性。     

武器采购中最小费用风险的最佳套期比率法

武器采购中常常涉及到汇率风险，避免汇率风险的方法是在外汇市场上购买外汇期货和约以进行套期保值，但由于多种因素的影响，期货价格与现货价格的波动和海跌幅度却常常差别较大，如此，通常的套期保值方法不能达到预期的保值效果。本文通过求取最佳套期保值率的方法，使其并风险的变化最小，从而通过选择最优的份数来锁定基差风险，实现预期的套期保值目的。一个算例表明：本方法较通常的套欺保值可节省费用３０％。     

基于动态时间弯曲的股票时间序列联动性研究

对于股票联动性的研究,传统时间序列分析方法及目前数据挖 掘技术主要使用国内或者国外股票指数来研究市场、板块或行业之间的联动关系,并得到一 些较为宏观的结论,存在着缺少直接分析与挖掘个股数据之间的联动性的问题。鉴于此,本文提出一种基于动态时间弯曲的股票时间序列联动性研究方法。通过动态时间弯曲找出若干只形态相似的股票,并在此基础上获得相关的重要信息,再提出基于动态时间弯曲的k-means聚类方法实现股票聚类,进而得到具有相同波动趋势的股票簇。实验结果表 明,新方法能从大量股票中准确找到具有联动关系的个股,区分开不同波动趋势的股票簇,具有一定的优越性。     

基于分层伪标签的图像聚类方法

图像聚类是图像处理中一个重要且开放的问题。最近，一些方法利用联合对比学习的良好表征能力来进行端到端聚类学习，利用伪标签技术来生成高质量的伪标签以提升聚类模型的鲁棒性。伪标签方法通常需要设置一个较大的概率阈值，并对满足要求的样本生成one-hot的标签，同时利用生成的标签来更新模型。但是，这种简单的伪标签生成方法难以获得足够数量的高质量伪标签。为了解决以上问题，提出了一种基于分层伪标签的图像聚类方法，它旨在利用结构化信息与伪标签信息对分类模型进行训练和精炼。引入3个假设来指导聚类方法的设计，包括局部平滑假设、自训练假设及低密度分离假设。新方法包含两个阶段：1)基于流形的一致性学习，利用近邻一致性学习来初始化聚类模型；2)基于分层伪标签的模型精炼，基于第一阶段的结果生成伪标签，并利用其来提升聚类模型的鲁棒性。首先，将基于第一阶段的结果生成强伪标签数据集及弱伪标签数据集；然后，提出了基于标签传播及分层混合的伪标签提升技术来提升弱伪标签数据集的质量；最后，同时利用强伪标签数据集及弱伪标签数据集来提升分类模型的泛化能力。相较于最优结果，SPC算法在STL10和Cifar100-20基准数据集上，...     

基于径向基神经网络和正则化极限学习机的多标签学习模型*

相比径向基(RBF)神经网络,极限学习机(ELM)训练速度更快,泛化能力更强.同时,近邻传播聚类算法(AP)可以自动确定聚类个数.因此,文中提出融合AP聚类、多标签RBF(ML-RBF)和正则化ELM(RELM)的多标签学习模型(ML-AP-RBF-RELM).首先,在该模型中输入层使用ML-RBF进行映射,且通过AP聚类算法自动确定每一类标签的聚类个数,计算隐层节点个数.然后,利用每类标签的聚类个数通过K均值聚类确定隐层节点RBF函数的中心.最后,通过RELM快速求解隐层到输出层的连接权值.实验表明,ML-AP-RBF-RELM效果较好.     

模糊时间序列建模及股票市场多步预测

首先应用模糊聚类方法将数据分类,以相邻两个聚类中心的中点作为子区间的分界点来划分论域,并以此将时间序列模糊化为模糊时间序列;其次根据证券市场主要量价指标建立了具有多个前件的高阶模糊关系;最后将该模型用于上证股票综合指数和深证股票成分指数的多步预测和涨跌趋势预测。与典型模糊时间序列模型比较,涨跌趋势预测准确率有较大提高,多步预测结果表明模型具有较好的泛化能力。     

基于趋势的时间序列相似性度量和聚类研究

由于时间序列的长度很大,并且不确定时间序列在每个采样点的取值具有不确定性,导致时间序列在相似性匹配和聚类挖掘中时间复杂度很高,为了解决该问题,提出了基于趋势的时间序列相似性度量方法和聚类方法.其中基于趋势的相似性度量方法根据时间序列的整体变化趋势,将时间序列映射为短的趋势符号序列,并利用各趋势的一阶连接性指数和塔尼莫特系数完成相似性度量;基于趋势的聚类方法通过定义趋势高度,并对趋势符号序列迭代进行区间划分和趋势判断,并以此构建趋势树,最后将趋势树根节点中趋势符号相同的序列聚集为一类.实验结果表明:a)五种趋势符号的一阶连接性指数可唯一地表示一条时间序列;b)基于趋势的相似性度量方法在多项式时间内可有效完成时间序列的相似性匹配;c)基于趋势的聚类方法将序列的相似性度量和聚类过程集中在一起,聚类效果显著.     

基于节点聚类复杂度的图聚类方法

图聚类可以发现网络中的社区结构，是复杂网络分析中的一项重要任务。针对不同节点的聚类难度各异的问题，提出了一种基于节点聚类复杂度的图聚类算法(Graph Clustering Algorithm Based on Node Clustering Complexity, GCNCC),用于判断节点的聚类复杂度，为聚类复杂度低的节点赋予伪标签，利用伪标签提供的监督信息降低其他节点的聚类复杂度，进而得到网络聚类结果。GCNCC包括节点表示、节点聚类复杂度判别和图聚类3个主要模块。节点表示模块得到保持网络集聚性的表示；节点聚类复杂度判别模块用于判断网络中的低聚类复杂度节点，并利用低聚类复杂度节点的伪标签信息来优化更新网络中其他节点的聚类复杂度；图聚类模块采用标签传播方法，将低聚类复杂度节点标签传播给高聚类复杂度节点，以得到聚类结果。在3个真实的引文网络和3个生物数据集上与9种经典算法进行对比，算法GCNCC在ACC,NMI,ARI和F1等方面均表现良好。     

符号化近似SAX在时序数据挖掘中的应用研究

聚类是数据挖掘研究中最常见的一种方法,可以作为规则发现、异常发现等其它数据挖掘操作的基础,一直以来都是数据挖掘的研究热点之一。股票数据是一种典型的时间序列数据,利用股票数据进行时间序列数据挖掘的研究既有一定的实际应用价值,也是国内外的热点问题之一。文章首次将一种新型符号化方法SAX[1]应用到标准普尔500指数的股票数据的聚类研究中,使用传统的欧氏距离和动态时间弯曲两种时间序列相似性度量方法进行实验。实验结果表明将SAX应用到股票数据聚类操作,可以得到更好的趋势聚类效果和更高的效率。     

Could the jump diffusion technique enhance the effectiveness of futures hedging models?: A reality test

A multivariate Markov-switching ARCH (MVSWARCH) model in which variance/correlations for futures and spot returns is controlled by a state-varying mechanism is introduced and used to design a state-varying stock index futures hedge ratio. Additionally, a conventional random-variance framework, the MVGARCH (multivariate GARCH) model with a time-varying technique is employed and subjected to a benchmark model. The feasibility of these proposed models is examined using two types of spot positions selected from the U.K. stock markets: (1) the FTSE-100 market index, representing a well-diversified market portfolio, and (2) ten sub-stock indices defined by the Data Stream database, representing the sub-set of the market portfolio. The empirical results are consistent with the following notions. First, when futures and spot returns are simultaneously (individually) based on low or high volatility states, the corresponding correlation measure between futures and spot returns is higher (lower). Second, consistent with prior studies, the in-sample hedging effectiveness tests demonstrated the superior performance of the stat-varying hedge ratio generated by the MVSWARCH model in all cases. However, our empirical results further indicate that the out-of-sample performance of the MVSWARCH-based hedge ratio is statistically marginal when investors hold a well-diversified market portfolio as their spot position and tranquil periods are experienced.     

标度曲线拟合与金融时间序列聚类

针对金融时间序列具有的多重分形特征,提出基于标度曲线测度沪深300指标股之间的相似性并实现聚类。该方法首先使用多标度退势波动分析(MSDFA)拟合不同自相关阶数下收益率序列的标度曲线,然后抽取其分布或形态特征构造模式向量。聚类通过含权K-means算法实现,最优类别数根据分类适确性指标(DBI)确定。结果显示,基于标度曲线的聚类能够揭示出股市的行业聚集性和板块间的关联性,在此基础上构造的投资组合可以显著降低风险,并且效果优于基于原始序列线性趋势特征的聚类。     

Forecasting the volatility of stock price index

Accurate volatility forecasting is the core task in the risk management in which various portfolios’ pricing, hedging, and option strategies are exercised. Prior studies on stock market have primarily focused on estimation of stock price index by using financial time series models and data mining techniques. This paper proposes hybrid models with neural network and time series models for forecasting the volatility of stock price index in two view points: deviation and direction. It demonstrates the utility of the hybrid model for volatility forecasting. This model demonstrates the utility of the neural network forecasting combined with time series analysis for the financial goods.     

Optimal dynamic hedging via copula-threshold-GARCH models

The contribution of this paper is twofold. First, we exploit copula methodology, with two threshold GARCH models as marginals, to construct a bivariate copula-threshold-GARCH model, simultaneously capturing asymmetric nonlinear behaviour in univariate stock returns of spot and futures markets and bivariate dependency, in a flexible manner. Two elliptical copulas (Gaussian and Student's-t) and three Archimedean copulas (Clayton, Gumbel and the Mixture of Clayton and Gumbel) are utilized. Second, we employ the presenting models to investigate the hedging performance for five East Asian spot and futures stock markets: Hong Kong, Japan, Korea, Singapore and Taiwan. Compared with conventional hedging strategies, including Engle's dynamic conditional correlation GARCH model, the results show that hedge ratios constructed by a Gaussian or Mixture copula are the best-performed in variance reduction for all markets except Japan and Singapore, and provide close to the best returns on a hedging portfolio over the sample period.     

Currency hedging strategies using dynamic multivariate GARCH

This paper examines the effectiveness of using futures contracts as hedging instruments of: (1) alternative models of volatility for estimating conditional variances and covariances; (2) alternative currencies; and (3) alternative maturities of futures contracts. For this purpose, daily data of futures and spot exchange rates of three major international currencies, Euro, British pound and Japanese yen, against the American dollar, are used to analyze hedge ratios and hedging effectiveness resulting from using two different maturity currency contracts, near-month and next-to-near-month contract. We estimate four multivariate volatility models (namely CCC, VARMA-AGARCH, DCC and BEKK), and calculate optimal portfolio weights and optimal hedge ratios to identify appropriate currency hedging strategies. The hedging effectiveness index suggests that the best results in terms of reducing the variance of the portfolio are for the USD/GBP exchange rate. The empirical results show that futures hedging strategies are slightly more effective when the near-month future contract is used for the USD/GBP and USD/JPY currencies. Moreover, the CCC and AGARCH models provide similar hedging effectiveness, which suggests that dynamic asymmetry may not be crucial empirically, although some differences appear when the DCC and BEKK models are used.     

Forecasting trends of high-frequency KOSPI200 index data using learning classifiers

Recently many statistical learning techniques have been applied to the prediction of financial variables. The aim of this paper is to conduct a comprehensive study of the applications of statistical learning techniques to predict the trend of the return of high-frequency Korea composite stock price index (KOSPI) 200 index data using the information from the one-minute time series of spot index, futures index, and foreign exchange rate. Through experiments, it is observed that the spot index change is better predictable with high-frequency time series data and the futures index information significantly improves the prediction accuracy of the return trends of the spot index for high-frequency index data, while the information of exchange rate does not. Also, dimension reduction process before training helps to increase the accuracy and dramatically for some classifiers. In addition, the trained classifiers with which a virtual trading strategy is applied to, noticeable better profits can be achieved than just a buy-and-hold-like strategy.     

状态变化下的连续时间动态投资组合选择

运用两状态隐马尔可夫模型刻画金融资产收益率序列的非线性变化,建立状态变化下的连续时间动态投资组合模型,利用动态规划得到最优投资决策的一般解,使用蒙特卡罗方法模拟投资者的投资决策行为.仿真结果表明:状态变化产生了对冲需求,对冲组合的大小依赖于投资者对市场状态的预期;当风险资产的波动率越小时,投资者状态信念的轻微变化都会引起对冲组合较大幅度的变化;当风险厌恶程度越大时,对冲组合对初始状态信念的变化越不敏感.     

SDAE-LSTM模型在金融时间序列预测中的应用

针对金融时间序列预测的复杂性和长期依赖性，提出了一种基于深度学习的LSTM神经网络预测模型。利用堆叠去噪自编码从金融时间序列的基本行情数据和技术指标中提取特征，将其作为LSTM神经网络的输入对金融时间序列进行预测；通过LSTM神经网络的长期依赖特性来提高金融时间序列的预测精度。利用股价指数数据，与传统的神经网络的预测结果进行比较，结果表明基于深度学习的LSTM神经网络具有比较高的预测精度。     

Application of enhanced cluster validity index function to automatic stock portfolio selection system

This paper presents an automatic stock portfolio selection system. In the proposed approach, 53 financial indices are collected for each stock item and are consolidated into six financial ratios [Grey relational grades (GRGs)] using a Grey relational analysis model. The GRGs are processed using a modified form of the PBMF index method (designated as the Huang index function) to determine the optimal number of clusters per GRG. The resulting cluster indices are then processed using rough set theory to identify the stocks within the lower approximate sets. Finally, the GRGs of each stock item in the lower approximate sets are consolidated into a single GRG, indicating the ability of the stock item to maximize the rate of return. It is demonstrated that the proposed stock selection mechanism yields a higher rate of return than several existing portfolio selection systems.     

Hedging global environment risks: An option based portfolio insurance

This paper introduces a financial hedging model for global environmental risks. Our approach is based on portfolio insurance under hedging constraints. Each investor is assumed to maximize the expected utility of his/her portfolio which includes financial and environmental assets. The optimal investment is determined for quite general utility functions and hedging constraints. Our results show how and why derivative assets should be introduced in the portfolio to hedge environmental risks.The main conclusion of the paper is that new types of options which combine both equity and environmental assets should be used, contrary to the current practice which considers two separate option markets.