Yoshihiko Konno

On estimation of a matrix of normal means with unknown covariance matrix

Let X be an m - p matrix normally distributed with matrix of means B and covariance matrix Im [circle times operator] [Sigma], where [Sigma] is a p - p unknown positive definite matrix. This paper studies the estimation of B relative to the invariant loss function tr . New classes of invariant minimax estimators are proposed for the case p > m + 1, which are multivariate extensions of the estimators of Stein and Baranchik. The method involves the unbiased estimation of the risk of an invariant estimator which depends on the eigenstructure of the usual F = XS-1Xt matrix, where S: p - p follows a Wishart matrix with n degrees of freedom and mean n[Sigma].

Shrinkage estimators for large covariance matrices in multivariate real and complex normal distributions under an invariant quadratic loss

The problem of estimating large covariance matrices of multivariate real normal and complex normal distributions is considered when the dimension of the variables is larger than the number of samples. The Stein-Haff identities and calculus on eigenstructure for singular Wishart matrices are developed for real and complex cases, respectively. By using these techniques, the unbiased risk estimates for certain classes of estimators for the population covariance matrices under invariant quadratic loss functions are obtained for real and complex cases, respectively. Based on the unbiased risk estimates, shrinkage estimators which are counterparts of the estimators due to Haff [L.R. Haff, Empirical Bayes estimation of the multivariate normal covariance matrix, Ann. Statist. 8 (1980) 586-697] are shown to improve upon the best scalar multiple of the empirical covariance matrix under the invariant quadratic loss functions for both real and complex multivariate normal distributions in the situation where the dimension of the variables is larger than the number of samples.

Estimating the covariance matrix and the generalized variance under a symmetric loss

For estimating the power of a generalized variance under a multivariate normal distribution with unknown means, the inadmissibility of the best affine equivariant estimator relative to the symmetric loss is shown, and a class of improved estimators is given. The problem of estimating the covariance matrix is also discussed.

A note on estimating eigenvalues of scale matrix of the multivariate F-distribution

Let Fpxphave the multivariate F-distribution with a scale matrix Δ and degrees of freedom n1and n2. In this paper the problem of estimating eigenvalues of Δ is considered. By constructing the improved orthogonally invariant estimators % MathType!MTEF!2!1!+-% feaafeart1ev1aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn% hiov2DGi1BTfMBaeXafv3ySLgzGmvETj2BSbqefm0B1jxALjhiov2D% aebbfv3ySLgzGueE0jxyaibaiiYdd9qrFfea0dXdf9vqai-hEir8Ve% ea0de9qq-hbrpepeea0db9q8as0-LqLs-Jirpepeea0-as0Fb9pgea% 0lrP0xe9Fve9Fve9qapdbaqaaeGacaGaaiaabeqaamaabaabcaGcba% WaaCbiaeaacqqHuoaraSqabeaacaqGEbaaaOGaaiikaiaadAeacaGG% Paaaaa!402A!\[\mathop \Delta \limits^{\rm{\^}} (F)\] of Δ, which are analogous to Haff-type estimators of a normal covariance matrix, new estimators of eigenvalues of Δ are given. This is because the eigenvalues of % MathType!MTEF!2!1!+-% feaafeart1ev1aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn% hiov2DGi1BTfMBaeXafv3ySLgzGmvETj2BSbqefm0B1jxALjhiov2D% aebbfv3ySLgzGueE0jxyaibaiiYdd9qrFfea0dXdf9vqai-hEir8Ve% ea0de9qq-hbrpepeea0db9q8as0-LqLs-Jirpepeea0-as0Fb9pgea% 0lrP0xe9Fve9Fve9qapdbaqaaeGacaGaaiaabeqaamaabaabcaGcba% WaaCbiaeaacqqHuoaraSqabeaacaqGEbaaaOGaaiikaiaadAeacaGG% Paaaaa!402A!\[\mathop \Delta \limits^{\rm{\^}} (F)\] are taken as estimates of the eigenvalues of Δ.

Bayes, minimax and nonnegative estimators of variance components under Kullback–Leibler loss

Abstract In a balanced one-way model with random effects, the simultaneous estimation of the variance components are considered under the intrinsic Kullback–Leibler loss function. The uniformly minimum variance unbiased (UMVU) or ANOVA estimators are known to have a drawback of taking negative values. The paper shows the minimaxity of the ANOVA estimators of the variance components and obtains classes of minimax estimators. Out of these classes, two types of minimax and nonnegative estimators are singled out, and they are characterized as empirical Bayes and generalized Bayes estimators. Also, a restricted maximum likelihood (REML) estimator is interpreted as an empirical Bayes rule. The risk performances of the derived estimators are investigated based on simulation experiments.

Entropy loss and risk of improved estimators for the generalized variance and precision

Let the distributions of X(p×r) and S(p×p) be N(ζ, Σ⊗Ir) and Wp(n, Σ) respectively and let them be independent. The risk of the improved estimator for |Σ| or {ei329-1} based on X and S under entropy loss (=d/|Σ| −log(d/|Σ|)−1 or d|Σ|−log(d|Σ|)−1) is evaluated in terms of incomplete beta function of matrix argument and its derivative. Numerical comparison for the reduction of risk over the best affine equivariant estimator is given.

Risk of Improved Estimators for Generalized Variance and Precision

Let the distributions of observed random matrices X(pxr) and S(pxp) be N(ξ, Σ⊗Іr) and Wp(n, Σ) respectively. Assume that they are independent. The risk of improved estimators for |Σ| or |Σ−1| based on X and S under squared loss is evaluated in terms of incomplete beta function of matrix argument. Numerical comparison for the reduction of risk over the best equivariant estimators is given.

Statistical inference based on the nonparametric maximum likelihood estimator under double-truncation

Doubly truncated data consist of samples whose observed values fall between the right- and left- truncation limits. With such samples, the distribution function of interest is estimated using the nonparametric maximum likelihood estimator (NPMLE) that is obtained through a self-consistency algorithm. Owing to the complicated asymptotic distribution of the NPMLE, the bootstrap method has been suggested for statistical inference. This paper proposes a closed-form estimator for the asymptotic covariance function of the NPMLE, which is computationally attractive alternative to bootstrapping. Furthermore, we develop various statistical inference procedures, such as confidence interval, goodness-of-fit tests, and confidence bands to demonstrate the usefulness of the proposed covariance estimator. Simulations are performed to compare the proposed method with both the bootstrap and jackknife methods. The methods are illustrated using the childhood cancer dataset.

A goodness-of-fit test for parametric models based on dependently truncated data

Suppose that one can observe bivariate random variables (L,X) only when L@?X holds. Such data are called left-truncated data and found in many fields, such as experimental education and epidemiology. Recently, a method of fitting a parametric model on (L,X) has been considered, which can easily incorporate the dependent structure between the two variables. A primary concern for the parametric analysis is the goodness-of-fit for the imposed parametric forms. Due to the complexity of dependent truncation models, the traditional goodness-of-fit procedures, such as Kolmogorov-Smirnov type tests based on the Bootstrap approximation to null distribution, may not be computationally feasible. In this paper, we develop a computationally attractive and reliable algorithm for the goodness-of-fit test based on the asymptotic linear expression. By applying the multiplier central limit theorem to the asymptotic linear expression, we obtain an asymptotically valid goodness-of-fit test. Monte Carlo simulations show that the proposed test has correct type I error rates and desirable empirical power. It is also shown that the method significantly reduces the computational time compared with the commonly used parametric Bootstrap method. Analysis on law school data is provided for illustration. R codes for implementing the proposed procedure are available in the supplementary material.

Construdtion of shrinkage estimators for the regression coefficient matrix in the gmanova model

This paper extensively investigates the theory of estimating the regression coefficient matrix in the normal GM.4KOVA model. We explicitly construct estimators which improve upon the maximum likelihood estimator under an invariant scalar loss function. These include the double shrinkage estimatois and those shrinking the maximum likelihood estimators directly. The underlying method is the decomposition of the problem into the conditional subproblems due to Kariya, Konno, and Strawderman(l996) and application of integration-by-parts technique to derive an unbiased estimate of the risk for certain class of invariant estimators.