Covariance tapering for prediction of large spatial data sets in transformed random fields

The best linear unbiased predictor (BLUP) is called a kriging predictor and has been widely used to interpolate a spatially correlated random process in scientific areas such as geostatistics. However, if an underlying random field is not Gaussian, the optimality of the BLUP in the mean squared error (MSE) sense is unclear because it is not always identical with the conditional expectation. Moreover, in many cases, data sets in spatial problems are often so large that a kriging predictor is impractically time-consuming. To reduce the computational complexity, covariance tapering has been developed for large spatial data sets. In this paper, we consider covariance tapering in a class of transformed Gaussian models for random fields and show that the BLUP using covariance tapering, the BLUP and the optimal predictor are asymptotically equivalent in the MSE sense if the underlying Gaussian random field has the Matérn covariance function.     

Admissibility of estimators of the probability of unobserved outcomes

The problem of estimating the probability of unobserved outcomes or, as it is sometimes called, the conditional probability of a new species, is studied. Good's estimator, which is essentially the same as Robbins' estimator, namely the number of singleton species observed divided by the sample size, is studied from a decision theory point of view. The results obtained are as follows: (1) When the total number of different species is assumed bounded by some known number, Good's and Robbins' estimators are inadmissible for squared error loss. (2) If the number of different species can be infinite, Good's and Robbins' estimators are admissible for squared error loss. (3) Whereas Robbins' estimator is a UMVUE for theunconditional probability of a new species obtained in one extra sample point, Robbins' estimator is not a uniformly minimum mean squared error unbiased estimator of the conditional probability of a new species. This answers a question raised by Robbins. (4) It is shown that for Robbins' model and squared error loss, there are admissible Bayes estimators which do not depend only on a minimal sufficient statistic. A discussion of interpretations and significance of the results is offered. Research supported by NSF Grant DMS-88-22622.     

A Generalization of Rao's Covariance Structure with Applications to Several Linear Models

This paper presents a generalization of Rao's covariance structure. In a general linear regression model, we classify the error covariance structure into several categories and investigate the efficiency of the ordinary least squares estimator (OLSE) relative to the Gauss–Markov estimator (GME). The classification criterion considered here is the rank of the covariance matrix of the difference between the OLSE and the GME. Hence our classification includes Rao's covariance structure. The results are applied to models with special structures: a general multivariate analysis of variance model, a seemingly unrelated regression model, and a serial correlation model.     

Covariance Tapering for Interpolation of Large Spatial Datasets

Interpolation of a spatially correlated random process is used in many scientific areas. The best unbiased linear predictor, often called a kriging predictor in geostatistical science, requires the solution of a (possibly large) linear system based on the covariance matrix of the observations. In this article, we show that tapering the correct covariance matrix with an appropriate compactly supported positive definite function reduces the computational burden significantly and still leads to an asymptotically optimal mean squared error. The effect of tapering is to create a sparse approximate linear system that can then be solved using sparse matrix algorithms. Monte Carlo simulations support the theoretical results. An application to a large climatological precipitation dataset is presented as a concrete and practical illustration.     

Second-order nonlinear least squares estimation

The ordinary least squares estimation is based on minimization of the squared distance of the response variable to its conditional mean given the predictor variable. We extend this method by including in the criterion function the distance of the squared response variable to its second conditional moment. It is shown that this “second-order” least squares estimator is asymptotically more efficient than the ordinary least squares estimator if the third moment of the random error is nonzero, and both estimators have the same asymptotic covariance matrix if the error distribution is symmetric. Simulation studies show that the variance reduction of the new estimator can be as high as 50% for sample sizes lower than 100. As a by-product, the joint asymptotic covariance matrix of the ordinary least squares estimators for the regression parameter and for the random error variance is also derived, which is only available in the literature for very special cases, e.g. that random error has a normal distribution. The results apply to both linear and nonlinear regression models, where the random error distributions are not necessarily known.     

High-Dimensional Covariance Matrix Estimation Based on Network

A new method for estimating high-dimensional covariance matrix based on network structure with heteroscedasticity of response variables is proposed in this paper. This method greatly reduces the computational complexity by transforming the high-dimensional covariance matrix estimation problem into a low-dimensional linear regression problem. Even if the size of sample is finite, the estimation method is still effective. The error of estimation will decrease with the increase of matrix dimension. In addition, this paper presents a method of identifying influential nodes in network via covariance matrix. This method is very suitable for academic cooperation networks by taking into account both the contribution of the node itself and the impact of the node on other nodes.     

The Covariance Adjusted Location Linear Discriminant Function for Classifying Data with Unequal Dispersion Matrices in Different Locations

Classification between two populations dealing with both continuous and binary variables is handled by splitting the problem into different locations. Given the location specified by the values of the binary variables, discrimination is performed using the continuous variables. The location probability model with homoscedastic across location conditional dispersion matrices is adopted for this problem. In this paper, we consider presence of continuous covariates with heterogeneous location conditional dispersion matrices. The continuous covariates have equal location specific mean in both populations. Conditional homoscedasticity fails when strong interaction between the continuous and binary variables is present. A plug-in covariance adjusted rule is constructed and its asymptotic distribution is derived. An asymptotic expansion for the overall error rate is given. The result is extended to include binary covariates.     

带Gauss型误差的GMANOVA-MANOVA模型中未知参数的精确置信域

对于带Gauss型误差的GMANOVA-MANOVA模型,在均匀协方差结构下,求出了其中未知参数的极大似然估计及其均值和方差,并依据极大似然估计构造了未知参数的精确置信域.     

An innovation approach to the smoothing problem from uncertain observations with correlated signal and noise

This paper treats the least mean‐squared error linear fixed‐point and fixed‐lag smoothing problems from uncertain observations, when the variables describing the uncertainty are independent, and the signal and observation white noise are correlated. Using an innovation approach, recursive algorithms are derived for both estimators without requiring the whole knowledge of the state‐space model generating the signal, but only covariance information of the signal and the observation noise, as well as the probability that the signal exists in the observed values. Copyright © 2005 John Wiley & Sons, Ltd.     

Prediction of Components in Random Sums

We consider predictions of the random number and the magnitude of each iid component in a random sum based on its distributional structure, where only a total value of the sum is available and where iid random components are non-negative. The problem is motivated by prediction problems in a Poisson shot noise process. In the context, although conditional moments are best possible predictors under the mean square error, only a few special cases have been investigated because of numerical difficulties. We replace the prediction problem of the process with that of a random sum, which is more general, and establish effective numerical procedures. The methods are based on conditional technique together with the Panjer recursion and the Fourier transform. In view of numerical experiments, procedures work reasonably. An application in the compound mixed Poisson process is also suggested.     

Bayesian Computation of Software Reliability

Abstract

Bayesian methods for the Jelinski and Moranda and the Littlewood and Verrall models in software reliability are studied. A Gibbs sampling approach is employed to compute the Bayes estimates. In addition, prediction of future failure times and future reliabilities is examined. Model selection based on the mean squared prediction error and the prequential likelihood of the conditional predictive ordinates is developed.     

Curve forecasting by functional autoregression

This paper deals with the prediction of curve-valued autoregression processes. It develops a novel technique, predictive factor decomposition, for the estimation of the autoregression operator. The technique is based on finding a reduced-rank approximation to the autoregression operator that minimizes the expected squared norm of the prediction error.Implementing this idea, we relate the operator approximation problem to the singular value decomposition of a combination of cross-covariance and covariance operators. We develop an estimation method based on regularization of the empirical counterpart of this singular value decomposition, prove its consistency and evaluate convergence rates.The method is illustrated by an example of the term structure of the Eurodollar futures rates. In the sample corresponding to the period of normal growth, the predictive factor technique outperforms the principal components method and performs on a par with custom-designed prediction methods.     

Bayesian shrinkage prediction for the regression problem

We consider Bayesian shrinkage predictions for the Normal regression problem under the frequentist Kullback-Leibler risk function.Firstly, we consider the multivariate Normal model with an unknown mean and a known covariance. While the unknown mean is fixed, the covariance of future samples can be different from that of training samples. We show that the Bayesian predictive distribution based on the uniform prior is dominated by that based on a class of priors if the prior distributions for the covariance and future covariance matrices are rotation invariant.Then, we consider a class of priors for the mean parameters depending on the future covariance matrix. With such a prior, we can construct a Bayesian predictive distribution dominating that based on the uniform prior.Lastly, applying this result to the prediction of response variables in the Normal linear regression model, we show that there exists a Bayesian predictive distribution dominating that based on the uniform prior. Minimaxity of these Bayesian predictions follows from these results.     

多维金融高频协方差阵预测模型的比较分析

现代投资组合理论大部分是从组合风险控制的角度展开,协方差矩阵扮演着非常重要的角色.将高频协方差阵应用在投资组合或风险管理时,就需要考虑采用何种预测模型来对高频协方差阵进行预测,较好的预测模型能够更加准确的对资产的波动性进行预测.高频协方差阵预测模型的建立较为复杂,目前还没有一种广泛被认可的模型.采用MCS检验法来选择最优的预测模型,研究发现高频协方差阵预测模型LOG-HAR模型在所有的损失函数下预测能力最好,并且高频协方差阵预测模型的预测能力要优于低频协方差阵预测模型.     

On an Alternative Parameterization of the Solutions of the Partial Realization Problem

The solutions of the partial realization problem have to satisfy a finite number of interpolation conditions at . The minimal degree of an interpolating deterministic system is called the algebraic degree or McMillan degree of the partial covariance sequence and is easy to compute. The solutions of the partial stochastic realization problem have to satisfy the same interpolation conditions and have to fulfill a positive realness constraint. The minimal degree of a stochastic realization is called the positive degree. In the literature, solutions of the partial realization problem are parameterized by the Kimura–Georgiou parameterization. Solutions of the partial stochastic realization problem are then obtained by checking the positive realness constraint for the interpolating solutions of the corresponding partial realization problem. In this paper, an alternative parameterization is developed for the solutions of the partial realization problems. Both the solutions of the partial and partial stochastic realization problem are analyzed in this parameterization, while the main concerns are the minimality and the uniqueness of the solutions. Based on the structure of the parameterization, a lower bound for the positive degree is derived.     

Simultaneous estimation of eigenvalues

The problem of simultaneous estimation of eigenvalues of covariance matrix is considered for one and two sample problems under a sum of squared error loss. New classes of estimators are obtained which dominate the best multiple of the sample eigenvalues in terms of risk. These estimators shrink or expand the sample eigenvalues towards their geometric mean. Similar results are obtained for the estimation of eigenvalues of the precision matrix and the residual matrix when the original covariance matrix is partitioned into two groups. As a consequence, a new estimator of trace of the covariance matrix is obtained.The results are extended to two sample problem where two Wishart distributions are independently observed, say, S _i W _p ( _i , k _i ), i=1, 2, and eigenvalues of _{1 2} ^-1 are estimated simultaneously. Finally, some numerical calculations are done to obtain the amount of risk improvement.     

Thermal behaviour prediction utilizing artificial neural networks for an open office

This study investigates a neural network-based non-linear autoregressive model with external inputs (NNARX), a non-linear autoregressive moving average model with external inputs (NNARMAX), and a non-linear output error model (NNOE) to predict the thermal behaviour of an open-plan office in a modern commercial building. External and internal climate data recorded over one summer, autumn and winter season were used to build and validate the models. The paper illustrates the potential of using these models to predict room temperature and relative humidity for different time scales ahead (30 min or 2 h ahead). The prediction performance is evaluated using the criteria of goodness of fit, coefficient of determination, mean absolute error and mean squared error between predicted model output and real measurements. To obtain an optimal network structure (avoiding overfitting) after training, a pruning algorithm called optimal brain surgeon (OBS) was used to remove unnecessary input signals, weights and hidden neurons. The results demonstrate that all models provide reasonably good predictions but the NNARX and NNARMAX models outperform the NNOE model. These models can all potentially be used for improving the performance of thermal environment control systems.     

增长曲线模型中一致最小风险无偏估计的存在性

考虑协方差阵任意，或具有均匀协方差结构，或具有序列协方差结构的正态增长曲线模型本文将文[19]在设计矩阵满秩，且仅估计回归系数矩阵的情形获得的结果推广到设计矩阵不必列满秩，且同时估计回归系数矩阵的线性可估函数和协方差阵（或有关参数）的情形；在凸损失函数类和矩阵损失函数下，给出存在一致最小风险无偏估计的充分必要条件．     

IMPROVED ESTIMATES OF THE COVARIANCE MATRIX IN GENERAL LINEAR MIXED MODELS

In this article, the problem of estimating the covariance matrix in general linear mixed models is considered. Two new classes of estimators obtained by shrinking the eigenvalues towards the origin and the arithmetic mean, respectively, are proposed. It is shown that these new estimators dominate the unbiased estimator under the squared error loss function. Finally, some simulation results to compare the performance of the proposed estimators with that of the unbiased estimator are reported. The simulation results indicate that these new shrinkage estimators provide a substantial improvement in risk under most situations.     

Uncertainty under a multivariate nested-error regression model with logarithmic transformation

This work aims to predict exponentials of mixed effects under a multivariate linear regression model with one random factor. Such quantities are of particular interest in prediction problems where the dependent variable is the logarithm of the variable that is the object of inference. Bias-corrected empirical predictors of the target quantities are defined. A second-order approximation for the mean crossed product error of two of these predictors is obtained, where the mean squared error is a particular case. An estimator of the mean crossed product error with second-order bias is proposed. Finally, results are illustrated through an application related to small area estimation.