联合均值与方差模型的变量选择

在许多应用方面, 特别在经济领域和工业产品的质量改进试验中, 非常有必要对方差建模. 推广经典的正态回归模型, 对联合均值与方差模型提出一种同时对均值模型和方差模型的变量选择方法. 提出的惩罚极大似然估计具有相合性和oracle性质. 随机模拟和实例研究结果表明该模型和方法是有用和有效的.     

考虑偏度特征的动态多响应稳健参数设计与优化

研究了具有动态特性的多响应稳健参数设计问题,分析了响应变量往往具有偏度特征的情况,提出了基于多元偏正态分布与响应曲面法相结合的动态多响应稳健优化模型,该模型不仅考虑了动态多响应之间的相关性,而且也考虑了尺度与偏度对动态多响应系统最优性与稳健性的影响。首先,利用非参数检验方法判断在信号因子不同水平下的各响应变量所服从的分布类型;其次,通过构建各响应变量在信号因子不同水平下的联合位置,尺度与偏度的响应曲面模型,进而建立基于多元偏正态分布的期望损失函数;然后,利用混合遗传算法对所构建的综合期望损失函数进行全局优化求解;最后,通过对具体的工业实例进行分析研究,结果表明本文所提出的方法能够有效地解决具有偏度特征的动态多响应稳健参数设计问题。     

Predicting status of Chinese listed companies based on features selected by penalized regression

China’s companies have attracted much attention due to the development of stock market in China. The listing status of listed Chinese companies becomes an important indicator which implies the potential risk of a stock. Thus predicting the status of listed Chinese companies is obviously crucial for stockholders and investors when they make further decisions. According to the four possible listing statuses for Chinese companies, researchers formulate the above issue as a classification problem which is typical in data mining area. Plenty of classification techniques have been implemented to predict the status of the listing Chinese companies based on their financial factors. Usually, there are more than 150 financial factors for each of the listed companies, and feature selection is needed before the implementation of classification methods. In the literature, researcher used t-test with variance inflation factor (VIF) analysis to select relevant factors. However, such method can not be applied in the high dimensional case. In this paper, we apply the idea of penalized regression to select the interested factors based on a logistic regression model, and then apply popular classification methods to predict the companies’ statuses. Our results show that the proposed method can find more representative factors and improves the prediction accuracy of the classification methods.     

Variable Selection via Generalized SELO-Penalized Cox Regression Models

The seamless-L_0(SELO) penalty is a smooth function that very closely resembles the L_0 penalty, which has been demonstrated theoretically and practically to be effective in nonconvex penalization for variable selection. In this paper, the authors first generalize the SELO penalty to a class of penalties retaining good features of SELO, and then develop variable selection and parameter estimation in Cox models using the proposed generalized SELO(GSELO) penalized log partial likelihood(PPL) approach. The authors show that the GSELO-PPL procedure possesses the oracle property with a diverging number of predictors under certain mild, interpretable regularity conditions. The entire path of GSELO-PPL estimates can be efficiently computed through a smoothing quasi-Newton(SQN) with continuation algorithm. The authors propose a consistent modified BIC(MBIC) tuning parameter selector for GSELO-PPL, and show that under some regularity conditions, the GSELOPPL-MBIC procedure consistently identifies the true model. Simulation studies and real data analysis are conducted to evaluate the finite sample performance of the proposed method.     

Feature Screening for Nonparametric and Semiparametric Models with Ultrahigh-Dimensional Covariates

This paper considers the feature screening and variable selection for ultrahigh dimensional covariates. The new feature screening procedure base on conditional expectation which is used to differentiate whether an explanatory variable contributes to a response variable or not, without requiring a specific parametric form of the underlying data model. The authors estimate the marginal conditional expectation by kernel regression estimator. The proposed method is showed to have sure screen property. The authors propose an iterative kernel estimator algorithm to reduce the ultrahigh dimensionality to an appropriate scale. Simulation results and real data analysis demonstrate the proposed method works well and performs better than competing methods.     

The adaptive LASSO spline estimation of single-index model

In this paper, based on spline approximation, the authors propose a unified variable selection approach for single-index model via adaptive L ₁ penalty. The calculation methods of the proposed estimators are given on the basis of the known lars algorithm. Under some regular conditions, the authors demonstrate the asymptotic properties of the proposed estimators and the oracle properties of adaptive LASSO (aLASSO) variable selection. Simulations are used to investigate the performances of the proposed estimator and illustrate that it is effective for simultaneous variable selection as well as estimation of the single-index models.     

Variable Selection of Varying Dispersion Student-t Regression Models

The Student-t regression model is a useful extension of the normal model,which can be used for statistical modeling of data sets involving errors with heavy tails and/or outliers and provides robust estimation of means and regression coefficients.In this paper,the varying dispersion Student-t regression model is discussed,in which both the mean and the dispersion depend upon explanatory variables.The problem of interest is simultaneously select significant variables both in mean and dispersion model.A unified procedure which can simultaneously select significant variable is given.With appropriate selection of the tuning parameters,the consistency and the oracle property of the regularized estimators are established.Both the simulation study and two real data examples are used to illustrate the proposed methodologies.     

Test for Heteroscedasticity in Partially Linear Regression Models

Testing heteroscedasticity determines whether the regression model can predict the dependent variable consistently across all values of the explanatory variables. Since the proposed tests could not detect heteroscedasticity in all cases, more precisely in heavy-tailed distributions, the authors established new comprehensive test statistic based on Levene's test. The authors built the asymptotic normality of the test statistic under the null hypothesis of homoscedasticity based on the recent theory of analysis of variance for the infinite factors level. The proposed test uses the residuals from a regression model fit of the mean function with Levene's test to assess homogeneity of variance. Simulation studies show that our test yields better than other methods in almost all cases even if the variance is a nonlinear function. Finally, the proposed method is implemented through a real data-set.     

A High-Moment Trapezoidal Fuzzy Random Portfolio Model with Background Risk

In most exiting portfolio selection models, security returns are assumed to have random or fuzzy distributions. However, uncertainties exist in actual financial markets. Markets are associated not only with inherent risk, but also with background risk that results from the differences among individual investors. This paper investigated the compliance of stock yields to the fuzzy-natured high-order moments of random numbers in order to develop a high-moment trapezoidal fuzzy random portfolio risk model based on variance, skewness, and kurtosis. Data obtained from the Shanghai Stock Exchange and Shenzhen Stock Exchange was used to assess the influence on the proposed model of both background risk and the maximum level of satisfaction of the portfolio. The empirical results demonstrated that the differences between the maximum and minimum variance, skewness, and kurtosis values of the portfolio were positively correlated with the variance of the background risk.     

Variable selection for recurrent event data with informative censoring

Recurrent events data with a terminal event(e.g.,death) often arise in clinical and observational studies.Variable selection is an important issue in all regression analysis.In this paper, the authors first propose the estimation methods to select the significant variables,and then prove the asymptotic behavior of the proposed estimator.Furthermore,the authors discuss the computing algorithm to assess the proposed estimator via the linear function approximation and generalized cross validation method for determination of the tuning parameters.Finally,the finite sample estimation for the asymptotical covariance matrix is also proposed.     

带网络结构的自适应Lasso分位数回归及其应用

CVaR是衡量组合投资的重要风险测度,如何在CVaR组合模型中选择稳健的资产组合以降低管理时间和经济成本十分重要.理论上CVaR模型下的资产组合决策可转化为分位数回归,受此驱动,该文构建了带网络结构的自适应Lasso分位数回归,对高维资产进行选择.自适应Lasso对变量的回归系数进行加权约束,理论上具有变量选择的一致性.网络结构是基于复杂网络理论构造,能够体现出资产之间的复杂联动关系,因此它对改进选择结果是有利的.该文基于线性规划进行求解,对CVaR组合投资决策中特有的计算问题采取两步迭代的方式进行.多种情形下的模拟分析显示,新模型的变量选择效果和预测表现均最优,且随着变量之间相关性的增强,网络结构带来的优势愈发明显.最后,使用249只股票数据进行了实证分析,通过滚动建模的方式,得出新模型具有良好的稳健性与应用意义.     

财务困境预测中的变量筛选——基于平均影响值的SVM方法

在财务困境预测中,如何从大量备选指标中筛选出预警指标是一个重要环节。为了更有效地设计财务困境预测模型,本文将平均影响值方法应用于SVM回归来进行变量筛选,首先对训练集数据用SVM进行训练,然后分别增减每一自变量的10%来进行仿真,对两个仿真结果的差值按样本数平均,得出平均影响值;最后对各个自变量的平均影响值按绝对值大小排序,从而进行变量筛选。实证结果表明,该方法能够以较少的特征变量实现较高的分类精度,是切实有效的。     

Variable Selection for Structural Equation with Endogeneity

This paper studies variable selection problem in structural equation of a two-stage least squares (2SLS) model in presence of endogeneity which is commonly encountered in empirical economic studies. Model uncertainty and variable selection in the structural equation is an important issue as described in Andrews and Lu (2001) and Caner (2009). The authors propose an adaptive Lasso 2SLS estimator for linear structural equation with endogeneity and show that it enjoys the oracle properties, i.e., the consistency in both estimation and model selection. In Monte Carlo simulations, the authors demonstrate that the proposed estimator has smaller bias and MSE compared with the bridge-type GMM estimator (Caner, 2009). In a case study, the authors revisit the classic returns to education problem (Angrist and Krueger, 1991) using the China Population census data. The authors find that the education level not only has strong effects on income but also shows heterogeneity in different age cohorts.     

Asymtotics of Dantzig selector for a general single-index model

As two popularly used variable selection methods, the Dantzig selector and the LASSO have been proved asymptotically equivalent in some scenarios. However, it is not the case in general for linear models, as disclosed in Gai, Zhu and Lin’s paper in 2013. In this paper, it is further shown that generally the asymptotic equivalence is not true either for a general single-index model with random design of predictors. To achieve this goal, the authors systematically investigate necessary and sufficient conditions for the consistent model selection of the Dantzig selector. An adaptive Dantzig selector is also recommended for the cases where those conditions are not satisfied. Also, different from existing methods for linear models, no distributional assumption on error term is needed with a trade-off that more stringent condition on the predictor vector is assumed. A small scale simulation is conducted to examine the performances of the Dantzig selector and the adaptive Dantzig selector.     

异方差非参数回归模型均值与方差变点的小波估计与应用

金融市场中,受突发事件的影响反映资产平均收益的均值函数和反映资产收益波动的方差函数都有可能出现变点. 本文讨论了均值和方差都存在变点的异方差非参数回归模型的变点估计问题. 给出均值函数与方差函数的局部线性估计,利用函数小波系数的特性求得均值与方差变点位置的估计值并给出其收敛速度.在模拟实验中分析变点估计值的样本特性及均值变点估计与方差变点估计的相互影响.最后通过对两组股票数据的均值变点和方差变点进行估计,说明方法的有效性.     

基于主基底分析的两阶段变量筛选方法

在实际系统分析及建模中,人们往往需要保留一些特别重要的分析变量。本文改进了基于主基底的变量筛选方法,分两个阶段来筛选系统分析所需变量。用重要变量构建初始主基底超平面,作为筛选其他普通变量的起点。该方法既结合了人们的定性分析经验,又保留了基于主基底分析的变量筛选方法能够自动筛选系统分析所需最简变量集合的特点,达到了数据降维目的。实际案例分析验证了该方法的有效性。     

基于变换核密度估计的半参数GARCH模型

针对金融资产收益率分布呈现的尖峰、厚尾及有偏的特点,沿袭变换核密度估计的思想,提出一种广义Logistic变换,对变换后的样本应用Beta核密度估计以消除边界偏差. 模拟试验表明,该方法显著提高了对尖峰厚尾分布密度的估计精度. 继而将该方法与参数化的GARCH设定相结合,建立一种半参数GARCH模型. 该模型具有两个优点：第一,基于变换核密度估计可更加准确地估计收益率的条件分布;第二,通过迭代提高了参数估计的稳健性. 模拟试验表明,较之伪极大似然估计法和基于离散最大惩罚似然估计的半参数方法,该方法大大提高了参数估计的相对效率. 对沪深300指数的实证研究验证了本文模型的有效性.     

基于矩阵分解和自适应图的无监督特征选择

在高维数据分析中,一个不可避免且棘手的问题是维度诅咒,因而如何将高维数据通过特征选择降维为低维数据显得尤为重要。对此, 提出了基于鲁棒矩阵分解和自适应图的无监督特征选择模型(unsupervised feature selection model based on robust matrix factorization and adaptive graph, MFAGFS), 实现在一个统一的学习框架下执行鲁棒矩阵分解、特征选择以及局部结构学习。模型首先通过鲁棒矩阵分解可获得聚类标签, 将聚类标签和局部结构信息用来引导特征选择过程, 再从特征选择的结果中自适应地学习数据局部结构。通过局部结构学习和特征选择这两个基本任务的相互作用, MFAGFS可以精确捕获数据的结构信息以及选择出具有判别性的特征。然后,详细阐述了算法优化求解方法, 并证明了算法的收敛性。最后,在6个公开数据集上进行试验对比分析, 参数敏感性分析, 验证了所提模型的有效性。实验结果表明, 所提的方法与其他方法相比, 性能均有不同程度的提高。     

考虑偏度风险和峰度风险的非线性期货套期保值模型

期货市场中,不仅存在方差风险,还存在偏度风险和峰度风险。但是现有的期货套期保值模型研究基本都是建立在方差风险基础上的,并没有考虑偏度风险和峰度风险对于套期保值的影响。针对现有研究的这一共同问题,本文以负指数效用函数为决策函数,提出了考虑偏度风险和峰度风险的非线性期货套期保值模型,并以原油的套期保值为例,讨论了偏度风险和峰度风险对于期货套期保值模型的影响。     

基于多尺度相似度特征的答案选择算法

答案选择的主要任务是对问答系统中问题的候选答案排序,当前主流的方法是基于表示学习方法,通过神经网络对问题和答案进行向量表示,然后根据向量相似度对候选答案排序,该类方法忽略了问题和答案的局部关联性。针对这一问题,提出了一种基于多尺度相似度特征的深度学习模型。该模型采取传统的深度学习模型分别提取问题和答案的特征,然后计算各个尺度下的特征相似度得到问答的相似度矩阵,最后采取三种不同的相似度特征学习模型对相似度矩阵学习得到联合相似度。在公开数据集WebQA上进行实验验证,实验结果表明将相似度特征学习方法引入传统深度学习模型获得了较为明显的提升。