Search

In recent years, model averaging, by which estimates are obtained based on not one single model but a weighted ensemble of models, has received growing attention as an alternative to model selection. To-date, methods for model averaging have been developed almost exclusively for point-valued data, despite the fact that interval-valued data are commonplace in many applications and the substantial body of literature on estimation and inference methods for interval-valued data. This paper focuses on the special case of interval time series data, and assumes that the mid-point and log-range of the interval values are modelled by a two-equation vector autoregressive with exogenous covariates (VARX) model. We develop a methodology for combining models of varying lag orders based on a weight choice criterion that minimises an unbiased estimator of the squared error risk of the model average estimator. We prove that this method yields predictors of mid-points and ranges with an optimal asymptotic property. In addition, we develop a method for correcting the range forecasts, taking into account the forecast error variance. An extensive simulation experiment examines the performance of the proposed model averaging method in finite samples. We apply the method to an interval-valued data series on crude oil future prices.

In this paper, we propose a frequentist model averaging method for quantile regression with high-dimensional covariates. Although research on these subjects has proliferated as separate approaches, no study has considered them in conjunction. Our method entails reducing the covariate dimensions through ranking the covariates based on marginal quantile utilities. The second step of our method implements model averaging on the models containing the covariates that survive the screening of the first step. We use a delete-one cross-validation method to select the model weights, and prove that the resultant estimator possesses an optimal asymptotic property uniformly over any compact (0,1) subset of the quantile indices. Our proof, which relies on empirical process theory, is arguably more challenging than proofs of similar results in other contexts owing to the high-dimensional nature of the problem and our relaxation of the conventional assumption of the weights summing to one. Our investigation of finite-sample performance demonstrates that the proposed method exhibits very favorable properties compared to the least absolute shrinkage and selection operator (LASSO) and smoothly clipped absolute deviation (SCAD) penalized regression methods. The method is applied to a microarray gene expression data set.

Model uncertainty is a prominent feature in many applied settings. This is certainty true in the efficiency analysis realm where concerns over the proper distributional specification of the error components of a stochastic frontier model is, generally, still open along with which variables influence inefficiency. Given the concern over the impact that model uncertainty is likely to have on the stochastic frontier model in practice, the present research proposes two distinct model averaging estimators, one which averages over nested classes of inefficiency distributions and another that has the ability to average over distinct distributions of inefficiency. Both of these estimators are shown to produce optimal weights when the aim is to uncover conditional inefficiency at the firm level. We study the finite-sample performance of the model average estimator via Monte Carlo experiments and compare with traditional model averaging estimators based on weights constructed from model selection criteria and present a short empirical application.

The generalized method of moments (GMM) and empirical likelihood (EL) are popular methods for combining sample and auxiliary information. These methods are used in very diverse fields of research, where competing theories often suggest variables satisfying different moment conditions. Results in the literature have shown that the efficient‐GMM (GMME) and maximum empirical likelihood (MEL) estimators have the same asymptotic distribution to order n−1/2 and that both estimators are asymptotically semiparametric efficient. In this paper, we demonstrate that when data are missing at random from the sample, the utilization of some well‐known missing‐data handling approaches proposed in the literature can yield GMME and MEL estimators with nonidentical properties; in particular, it is shown that the GMME estimator is semiparametric efficient under all the missing‐data handling approaches considered but that the MEL estimator is not always efficient. A thorough examination of the reason for the nonequivalence of the two estimators is presented. A particularly strong feature of our analysis is that we do not assume smoothness in the underlying moment conditions. Our results are thus relevant to situations involving nonsmooth estimating functions, including quantile and rank regressions, robust estimation, the estimation of receiver operating characteristic (ROC) curves, and so on.

In the last decade, significant theoretical advances have been made in the area of frequentist model averaging (FMA); however, the majority of this work has emphasized parametric model setups. This...

Often fixed-effects dynamic panel data model assumes parametric structures and an AR(1) dynamic order. The latter assumption is mainly for convenience and not consistent with many sampling processes especially when longer panels are available. A fixed-effects dynamic partially linear additive model with a finite autoregressive lag order is considered. Based on this setup, semiparametric Generalized Method of Moments (GMM) estimators of the unknown coefficients and functions using the B(asis)-spline approximation are developed. The asymptotic properties of these estimators are established. A procedure to identify the dynamic lag order and significant exogenous variables by employing the smoothly clipped absolute deviation (SCAD) penalty is developed. It is proven that the SCAD-based GMM estimators achieve the oracle property and are selection consistent. The usefulness of proposed procedure is further illustrated in Monte Carlo studies and a real data example.

In recent years, the suggestion of combining models as an alternative to selecting a single model from a frequentist prospective has been advanced in a number of studies. In this paper, we propose a new semi-parametric estimator of regression coe¢ cients, which is in the form of a feasible generalized ridge estimator by Hoerl and Kennard (1970b) but with di¤erent biasing factors. We prove that the generalized ridge estimator is algebraically identical to the model average estimator. Further, the biasing factors that determine the properties of both the generalized ridge and semi-parametric estimators are directly linked to the weights used in model averaging. These are interesting results for the interpretations and applications of both semi-parametric and ridge estimators. Furthermore, we demonstrate that these estimators based on model averaging weights can have properties superior to the well-known feasible generalized ridge estimator in a large region of the parameter space. Two empirical examples are presented.

Considerable intellectual progress has been made to the development of various semiparametric varying-coefficient models over the past ten to fifteen years. An important advantage of these models is that they avoid much of the curse of dimensionality problem as the nonparametric functions are restricted only to some variables. More recently, varying-coefficient methods have been applied to quantile regression modeling, but all previous studies assume that the data are fully observed. The main purpose of this paper is to develop a varying-coefficient approach to the estimation of regression quantiles under random data censoring. We use a weighted inverse probability approach to account for censoring, and propose a majorize-minimize type algorithm to optimize the non-smooth objective function. The asymptotic properties of the proposed estimator of the nonparametric functions are studied, and a resampling method is developed for obtaining the estimator of the sampling variance. An important aspect of our method is that it allows the censoring time to depend on the covariates. Additionally, we show that this varying-coefficient procedure can be further improved when implemented within a composite quantile regression framework. Composite quantile regression has recently gained considerable attention due to its ability to combine information across different quantile functions. We assess the finite sample properties of the proposed procedures in simulated studies. A real data application is also considered.

This article examines the problem of estimation in a quantile regression model when observations are missing at random under independent and nonidentically distributed errors. We consider three approaches of handling this problem based on nonparametric inverse probability weighting, estimating equations projection, and a combination of both. An important distinguishing feature of our methods is their ability to handle missing response and/or partially missing covariates, whereas existing techniques can handle only one or the other, but not both. We prove that our methods yield asymptotically equivalent estimators that achieve the desirable asymptotic properties of unbiasedness, normality, and n-consistency. Because we do not assume that the errors are identically distributed, our theoretical results are valid under heteroscedasticity, a particularly strong feature of our methods. Under the special case of identical error distributions, all of our proposed estimators achieve the semiparametric efficiency b...

This article considers a nonparametric additive seemingly unrelated regression model with autoregressive errors, and develops estimation and inference procedures for this model. Our proposed method first estimates the unknown functions by combining polynomial spline series approximations with least squares, and then uses the fitted residuals together with the smoothly clipped absolute deviation (SCAD) penalty to identify the error structure and estimate the unknown autoregressive coefficients. Based on the polynomial spline series estimator and the fitted error structure, a two-stage local polynomial improved estimator for the unknown functions of the mean is further developed. Our procedure applies a prewhitening transformation of the dependent variable, and also takes into account the contemporaneous correlations across equations. We show that the resulting estimator possesses an oracle property, and is asymptotically more efficient than estimators that neglect the autocorrelation and/or contemporaneous...

Most of the literature on clustered data models emphasizes two-level clustering, and within-cluster correlation. While multi-level clustered data models can arise in practice, analysis of multi-level clustered data models poses additional difficulties owing to the existence of error correlations both within and across the clusters. It is perhaps for this reason that existing approaches to multi-level clustered data models have been mostly parametric. The purpose of this paper is to develop a varying-coefficient nonparametric approach to the analysis of three-level clustered data models. Because the nonparametric functions are restricted only to some of the variables, this approach has the appeal of avoiding many of the curse of dimensionality problems commonly associated with other nonparametric methods. By applying an undersmoothing technique, taking into account the correlations within and across clusters, we develop an efficient two-stage local polynomial estimation procedure for the unknown coefficient functions. The large and finite sample properties of the resultant estimators are examined; in particular, we show that the resultant estimators are asymptotically normal, and exhibit considerably smaller asymptotic variability than the traditional local polynomial estimators that neglect the correlations within and among clusters. An application example is presented based on a data set extracted from the World Bank’s STARS database.

Multinomial and ordered Logit models are quantitative techniques which are used in a range of disciplines nowadays. When applying these techniques, practitioners usually select a single model using either information-based criteria or pretesting. In this paper, we consider the alternative strategy of combining models rather than selecting a single model. Our strategy of weight choice for the candidate models is based on the minimization of a plug-in estimator of the asymptotic squared error risk of the model average estimator. Theoretical justifications of this model averaging strategy are provided, and a Monte Carlo study shows that the forecasts produced by the proposed strategy are often more accurate than those produced by other common model selection and model averaging strategies, especially when the regressors are only mildly to moderately correlated and the true model contains few zero coefficients. An empirical example based on credit rating data is used to illustrate the proposed method. To reduce the computational burden, we also consider a model screening step that eliminates some of the very poor models before averaging.

We study estimation and inference in a marginal proportional hazards model that can handle (1) linear effects, (2) non-linear effects and (3) interactions between covariates. The model under consideration is an amalgamation of three existing marginal proportional hazards models studied in the literature. Developing an estimation and inference procedure with desirable properties for the amalgamated model is rather challenging due to the co-existence of all three effects listed above. Much of the existing literature has avoided the problem by considering narrow versions of the model. The object of this paper is to show that an estimation and inference procedure that accommodates all three effects is within reach. We present a profile partial-likelihood approach for estimating the unknowns in the amalgamated model with the resultant estimators of the unknown parameters being root- $$n$$ consistent and the estimated functions achieving optimal convergence rates. Asymptotic normality is also established for the estimators.

The past decade witnessed a literature on model averaging by frequentist methods. For the most part, the asymptotic optimality of various existing frequentist model averaging estimators has been established under i.i.d. errors. Recently, Hansen and Racine [Hansen, B.E., Racine, J., 2012. Jackknife model averaging. Journal of Econometrics 167, 38–46] developed a jackknife model averaging (JMA) estimator, which has an important advantage over its competitors in that it achieves the lowest possible asymptotic squared error under heteroscedastic errors. In this paper, we broaden Hansen and Racine’s scope of analysis to encompass models with (i) a non-diagonal error covariance structure, and (ii) lagged dependent variables, thus allowing for dependent data. We show that under these set-ups, the JMA estimator is asymptotically optimal by a criterion equivalent to that used by Hansen and Racine. A Monte Carlo study demonstrates the finite sample performance of the JMA estimator in a variety of model settings.

Claeskens and Hjort (2003) have developed a focused information criterion (FIC) for model selection that selects different models based on different focused functions with those functions tailored to the parameters singled out for interest. Hjort and Claeskens (2003) also have presented model averaging as an alternative to model selection, and suggested a local misspecification framework for studying the limiting distributions and asymptotic risk properties of post-model selection and model average estimators in parametric models. Despite the burgeoning literature on Tobit models, little work has been done on model selection explicitly in the Tobit context. In this article we propose FICs for variable selection allowing for such measures as mean absolute deviation, mean squared error, and expected expected linear exponential errors in a type I Tobit model with an unknown threshold. We also develop a model average Tobit estimator using values of a smoothed version of the FIC as weights. We study the finite...

Least-squares and quantile regressions are method of moments techniques that are typically used in isolation. A leading example where efficiency may be gained by combining least-squares and quantile regressions is one where some information on the error quantiles is available but the error distribution cannot be fully specified. This estimation problem may be cast in terms of solving an over-determined estimating equation (EE) system for which the generalized method of moments (GMM) and empirical likelihood (EL) are approaches of recognized importance. The major difficulty with implementing these techniques here is that the EEs associated with the quantiles are non-differentiable. In this paper, we develop a kernel-based smoothing technique for non-smooth EEs, and derive the asymptotic properties of the GMM and maximum smoothed EL (MSEL) estimators based on the smoothed EEs. Via a simulation study, we investigate the finite sample properties of the GMM and MSEL estimators that combine least-squares and quantile moment relationships. Applications to real datasets are also considered.

There has been increasing interest recently in model averaging within the frequentist paradigm. The main benefit of model averaging over model selection is that it incorporates rather than ignores the uncertainty inherent in the model selection process. One of the most important, yet challenging, aspects of model averaging is how to optimally combine estimates from different models. In this work, we suggest a procedure of weight choice for frequentist model average estimators that exhibits optimality properties with respect to the estimator’s mean squared error (MSE). As a basis for demonstrating our idea, we consider averaging over a sequence of linear regression models. Building on this base, we develop a model weighting mechanism that involves minimizing the trace of an unbiased estimator of the model average estimator’s MSE. We further obtain results that reflect the finite sample as well as asymptotic optimality of the proposed mechanism. A Monte Carlo study based on simulated and real data evaluates...

The recently proposed ‘weighted average least squares’ (WALS) estimator is a Bayesian combination of frequentist estimators. It has been shown that the WALS estimator possesses major advantages over standard Bayesian model averaging (BMA) estimators: the WALS estimator has bounded risk, allows a coherent treatment of ignorance and its computational effort is negligible. However, the sampling properties of the WALS estimator as compared to BMA estimators are heretofore unexamined. The WALS theory is further extended to allow for nonspherical disturbances, and the estimator is illustrated with data from the Hong Kong real estate market. Monte Carlo evidence shows that the WALS estimator performs significantly better than standard BMA and pretest alternatives.

Model averaging or combining is often considered as an alternative to model selection. Frequentist Model Averaging (FMA) is considered extensively and strategies for the application of FMA methods in the presence of missing data based on two distinct approaches are presented. The first approach combines estimates from a set of appropriate models which are weighted by scores of a missing data adjusted criterion developed in the recent literature of model selection. The second approach averages over the estimates of a set of models with weights based on conventional model selection criteria but with the missing data replaced by imputed values prior to estimating the models. For this purpose three easy-to-use imputation methods that have been programmed in currently available statistical software are considered, and a simple recursive algorithm is further adapted to implement a generalized regression imputation in a way such that the missing values are predicted successively. The latter algorithm is found to be quite useful when one is confronted with two or more missing values simultaneously in a given row of observations. Focusing on a binary logistic regression model, the properties of the FMA estimators resulting from these strategies are explored by means of a Monte Carlo study. The results show that in many situations, averaging after imputation is preferred to averaging using weights that adjust for the missing data, and model average estimators often provide better estimates than those resulting from any single model. As an illustration, the proposed methods are applied to a dataset from a study of Duchenne muscular dystrophy detection.

This paper is in response to a recent paper by Hansen (2007) who proposed an optimal model average estimator with weights selected by minimizing a Mallows criterion. The main contribution of Hansen’s paper is a demonstration that the Mallows criterion is asymptotically equivalent to the squared error, so the model average estimator that minimizes the Mallows criterion also minimizes the squared error in large samples. We are concerned with two assumptions that accompany Hansen’s approach. The first is the assumption that the approximating models are strictly nested in a way that depends on the ordering of regressors. Often there is no clear basis for the ordering and the approach does not permit non-nested models which are more realistic from a practical viewpoint. Second, for the optimality result to hold the model weights are required to lie within a special discrete set. In fact, Hansen noted both difficulties and called for extensions of the proof techniques. We provide an alternative proof which shows that the result on the optimality of the Mallows criterion in fact holds for continuous model weights and under a non-nested set-up that allows any linear combination of regressors in the approximating models that make up the model average estimator. These results provide a stronger theoretical basis for the use of the Mallows criterion in model averaging by strengthening existing findings.

The Stein-rule (SR) and positive-part Stein-rule (PSR) estimators are two popular shrinkage techniques used in linear regression, yet very little is known about the robustness of these estimators to the disturbances’ deviation from the white noise assumption. Recent studies have shown that the OLS estimator is quite robust, but whether this is so for the SR and PSR estimators is less clear as these estimators also depend on the F statistic which is highly susceptible to covariance misspecification. This study attempts to evaluate the effects of misspecifying the disturbances as white noise on the SR and PSR estimators by a sensitivity analysis. Sensitivity statistics of the SR and PSR estimators are derived and their properties are analyzed. We find that the sensitivity statistics of these estimators exhibit very similar properties and both estimators are extremely robust to MA(1) disturbances and reasonably robust to AR(1) disturbances except for the cases of severe autocorrelation. The results are useful in light of the rising interest of the SR and PSR techniques in the applied literature.

There is a large and growing body of literature on estimating equation (EE) as an estimation approach. One basic property of EE that has been universally adopted in practice is that of unbiasedness, and there are deep conceptual reasons why unbiasedness is a desirable EE characteristic. This article deals with inference from EEs when data are missing at random. The investigation is motivated by the observation that direct imputation of missing data in EEs generally leads to EEs that are biased and, thus, violates a basic assumption of the EE approach. The main contribution of this article is that it goes beyond existing imputation methods and proposes a procedure whereby one mitigates the effects of missing data through a reformulation of EEs imputed through a kernel regression method. These (modified) EEs then constitute a basis for inference by the generalized method of moments (GMM) and empirical likelihood (EL). Asymptotic properties of the GMM and EL estimators of the unknown parameters are derived a...

Summary Traditional econometrics has long stressed the serious consequences of non-spherical disturbances for the estimation and testing procedures under the spherical disturbance setting, that is, the procedures become invalid and can give rise to misleading results. In practice, it is not unusual, however, to find that the parameter estimates do not change much after fitting the more general structure. This suggests that the usual procedures may well be robust to covariance misspecification. Banerjee and Magnus (1999) proposed sensitivity statistics to decide if the Ordinary Least Squares estimators of the coefficients and the disturbance variance are sensitive to deviations from the spherical error assumption. This paper extends their work by investigating the sensitivity of the restricted least squares estimator to covariance misspecification where the restrictions may or may not be correct. Large sample results giving analytical evidence to some of the numerical findings reported in Banerjee and Magnus (1999) are also obtained.

This note considers the problem of estimating regression coefficients when some other coefficients in the model are of no interest. For the case of normal errors, Magnus and Durbin [1999. Estimation of regression coefficients of interest when other regression coefficients are of no interest. Econometrica 67, 639–643] and Danilov and Magnus [2004. On the harm that ignoring pretesting can cause. J. Econometrics 122, 27–46] studied this problem and established an equivalence theorem which states that the problem of estimating the coefficients of interest is equivalent to that of finding an optimal estimator of the vector of coefficients of no interest given a single observation from a normal distribution. The aim of this note is to generalize their findings to the large sample non-normal errors case. Some applications of our results are also given.

This paper compares the Stein and the usual estimators of the error variance under the Pitman nearness (PN) criterion in a regression model which is mis-specified due to missing relevant explanatory variables. The exact expression of the PN-probability is derived and numerically evaluated. Contrary to the well-known result under mean squared errors (MSE), with the PN criterion the Stein variance estimator is uniformly dominated by the usual estimator when no relevant variables are excluded from the model. With an increased degree of model mis-specification, neither estimator strictly dominates the other.

In hedonic housing price modeling, real estate researchers and practitioners are often not completely ignorant about the parameters to be estimated. Experience and expertise usually provide them with tacit understanding of the likely values of the true parameters. Under this scenario the subjective knowledge about the parameter value can be incorporated as non-sample information in the hedonic price model. In this paper, we consider a class of Generalized Stein Variance Double k-class (GSVKK) estimators which allows real estate practitioners to introduce potentially useful information about the parameter values in the estimation of hedonic pricing models. The GSVKK estimator is a generalization of a family of shrinkage estimators introduced by Ohtani and Wan (2002, Econometric Reviews). Data from the Hong Kong real estate market are used to investigate the estimators' performance empirically. Compared with the traditional Ordinary Lease Squares approach, the GSVKK estimators have smaller predictive mean squared errors and lead to more precise parameter estimates. Some results on the theoretical properties of the GSVKK estimators are also presented.

Summary This paper enlarges on results of Wan and Zou [Journal of Econometrics 114 (2003), 165–96] on the choice of critical values for pre-test procedures based on the minimum risk criterion. We consider a modification of the general theorem given in Wan and Zou (2003) to obtain the optimal critical value that minimizes the risks of various inequality pre-test estimators of the regression error variance under a general class of first-order differentiable loss functions. Theoretical proofs of earlier numerical results are provided. This paper also presents results on the optimal pre-test critical values for the simultaneous estimation of the error variance and coefficient vector.

In this paper, we derive the exact general expressions for the moments of the Lawless-Wang's operational almost unbiased ridge regression (AUGRR) estimator for individual regression coefficients. © 2003 Elsevier Inc. All rights reserved.

There is now a sizeable literature dealing with point estimation using Stein-type estimators. As discussed in Rukhin (In: Gupta, S.S., Berger, J.O. (Eds.), Statistical Decision Theory and Related Topics, Vol. IV, Springer, New York, pp. 409–418), instances arise in practice in which an estimation rule is to be accompanied by an estimate of its loss, which is unobservable. In the context of estimating the mean vector of a multi-normal distribution assuming a known population variance, Johnstone (In: Gupta, S.S., Berger, J.O. (Eds.), Statistical Decision Theory and Related Topics, Vol. IV, Springer, New York, pp. 361–379) derived an estimator that dominates the unbiased estimator of the quadratic loss incurred by the James–Stein estimator. By applying the Stein's lemma, this note generalizes Johnstone's (In: Gupta, S.S., Berger, J.O. (Eds.), Statistical Decision Theory and Related Topics, Vol. IV, Springer, New York, pp. 361–379) analysis to the setting of the unknown population variance. Computational evidence is provided about the risk magnitude of loss estimators associated with the James–Stein point estimator and its positive-part version.

This paper re-visits the problem of estimating the regression error variance in a linear multiple regression model after preliminary hypothesis tests for either linear restrictions on the coefficients or homogeneity of variances. There is an extensive literature that discusses these problems, particularly in terms of the sampling properties of the pre-test estimators using various loss functions as the basis for risk analysis. In this paper, a unified framework for analysing the risk properties of these estimators is developed under a general class of loss structures that incorporates virtually all first-order differentiable losses. Particular consideration is given to the choice of critical values for the pre-tests. Analytical results indicate that an α-level substantially higher than those normally used may be appropriate for optimal risk properties under a wide range of loss functions. The paper also generalizes some known analytical results in the pre-test literature and proves other results only previously shown numerically.

In this paper, the problem of estimating the parameter n of the binomial distribution is considered under the assumption of both infinite and finite parameter spaces. A class of estimators, which include as a special case Sadooghi-Alvandi's [Ann. Statist. 14 (1986) 1634] estimator is considered; and the necessary and sufficient conditions for this class of estimators to be admissible are obtained, along with a new proof on the admissibility of Sadooghi-Alvandi's estimator. Furthermore, the minimax property of some estimators is investigated.

Stein-rule and other improved estimators have scarcely been used in empirical work. One major reason is that it is not easy to obtain precision measures for these estimators. In this paper, we derive unbiased estimators for both the mean squared error (MSE) and the scaled MSE matrices of a class of Stein-type estimators. Our derivation provides the basis for measuring the estimators' precision and constructing confidence bands. Comparisons are made between these MSE estimators and the least squares covariance estimator. For illustration, the methodology is applied to data on energy consumption.

In regression analysis, ridge estimators are often used to alleviate the problem of multicollinearity. Ridge estimators have traditionally been evaluated using the risk under quadratic loss criterion, which places sole emphasis on estimators' precision. Here, we consider the balanced loss function (A. Zellner, in: S.S. Gupta, J.O. Berger (Eds.), Statistical Decision Theory and Related Topics, vol. V, Springer, New York, 1994, p. 377) which incorporates a measure for the goodness of fit of the model as well as estimation precision. By adopting this loss we derive and numerically evaluate the risks of the feasible generalized ridge and the almost unbiased feasible generalized ridge estimators. We show that in the case of severe multicollinearity, the feasible generalized ridge estimator often produces the greatest risk reductions, even if a relatively heavy weight is given to goodness of fit in the balanced loss function.

Despite the sizeable literature associated with the seemingly unrelated regression models, not much is known about the use of Stein-rule estimators in these models. This gap is remedied in this paper, in which two families of Stein-rule estimators in seemingly unrelated regression equations are presented and their large sample asymptotic properties explored and evaluated. One family of estimators uses a shrinkage factor obtained solely from the equation under study while the other has a shrinkage factor based on all the equations of the model. Using a quadratic loss measure and Monte-Carlo sampling experiments, the finite sample risk performance of these estimators is also evaluated and compared with the traditional feasible generalized least squares estimator.

Recent years have seen a growing body of literature on the analysis of length-biased data. Much of this literature adopts the accelerated failure time or proportional hazards models as the basis of study. The overwhelming majority of the existing work also assumes independence between the censoring variable and covariates. In this paper, we develop a varying-coefficient quantile regression approach to model length-biased data. Our approach does not only allow the direct estimation of the conditional quantiles of survival times based on a flexible model structure, but also has the important appeal of permitting dependence between the censoring variable and the covariates. We develop local linear estimators of the coefficients using a local inverse probability weighted estimating equation approach, and examine these estimators’ asymptotic properties. Moreover, we develop a resampling method for computing the estimators’ covariances. The small sample properties of the proposed methods are investigated in a simulation study. A real data example illustrates the application of the methods in practice.

In this paper, we consider a family of feasible generalised double k-class estimators in a linear regression model with non-spherical disturbances. We derive the large sample asymptotic distribution of the proposed family of estimators and compare its performance with the feasible generalized least squares and Stein-rule estimators using the mean squared error matrix and risk under quadratic loss criteria. A Monte-Carlo experiment investigates the finite sample behaviour of the proposed family of estimators.

We consider the estimation of the error variance of a linear regression model where prior information is available in the form of an (uncertain) inequality constraint on the coefficients. Previous studies on this and other related problems use the squared error loss in comparing estimator’s performance. Here, by adopting the asymmetric LINEX loss function, we derive and numerically evaluate the exact risks of the inequality constrained estimator and the inequality pre-test estimator which results after a preliminary test for an inequality constraint on the coefficients. The risks based on squared error loss are special cases of our results, and we draw appropriate comparisons.

In various guises, feasible generalized least squares (FGLS) estimation has occupied an important place in regression analysis for more than 35 years. Past studies on the characteristics of the FGLS estimators are largely based on large sample evaluations, and the important issue of admissibility remains unexplored in the case of the FGLS estimator. In this paper, an exact sufficient condition for the dominance of a Stein-type shrinkage estimator over the FGLS estimator in finite samples based on squared error loss is given. In deriving the condition, we assume that the model's disturbance covariance matrix is unknown except for a scalar multiple. Further, for models with AR(1) disturbances, it is observed that the dominance condition reduces to one that involves no unknown parameter. In other words, in the case of AR( I ) disturbances and where the condition for risk dominance is met, the FGLS estimator is rendered inadmissible under squared error loss.

There is a good deal of literature that investigates the properties of various operational variants of Theil's (1971, Principles of Econometrics, Wiley, New York) minimum mean squared error estimator. It is interesting that virtually all of the existing analysis to date is based on the premise that the model's disturbances are i.i.d., an assumption which is not satisfied in many practical situations. In this paper, we consider a model with non-spherical errors and derive the asymptotic distribution, bias and mean squared error of a general class of feasible minimum mean squared error estimators. A Monte-Carlo experiment is conducted to examine the performance of this class of estimators in finite samples.

This paper considers simultaneous estimation of means from several strata under error-in-variables superpopulation models. Necessary and sufficient conditions for an estimator to be admissible in the class of linear estimators are obtained.

This paper considers adaptive versions of the minimum mean-squared error estimators in models with an inequality constraint. We derive a sufficient condition under which the proposed class of estimators dominates the traditional inequality constrained least-squares estimator in terms of risk under quadratic loss. Numerical calculations of the risks show that over much of the parameter space, the proposed estimators are superior to the inequality constrained estimator, even if the sufficient condition is not satisfied, and some members of this class have risk advantage over the inequality constrained Stein-rule estimator proposed by Judge et al. (1984, J. Econometrics 25, 165–177) over a wide range of parameter values.

In this article, we consider the risk performance of an iterative feasible minimum mean squared error estimator of the regression disturbance variance under the LINEX loss function. This loss is a generalisation of the quadratic loss function allowing for asymmetry. Notwithstanding the justification for using the feasible minimum mean squared error estimator in estimating the regression coefficients, it is found that the corresponding estimator of the disturbance variance does not, in general, improve over a class of conventional estimators commonly used in practice.

Using the asymmetric LINEX loss function, we derive and numerically evaluate the exact risk function of the almost unbiased feasible generalized ridge regression estimator. Contrary to the properties of the (biased) feasible generalized ridge estimator, it is found that regardless of the loss asymmetry,the almost unbiased feasible generalized ridge estimator does not strictly dominate the traditional least squares estimator. Our numerical results show that over a wide range of parameter values, the almost unbiased feasible generalized ridge estimator is inferior to either the least squares or the feasible generalized ridge estimators.

Using the Stein (1964) variance estimator, this paper defines a modified Stein inequality constrained estimator and derives its exact risk under quadratic loss. Numerical evaluations show that over a wide range of the parameter space, the modified Stein inequality constrained estimator has lower risk than the traditional Stein inequality constrained estimator introduced by Judge et al. (1984).

We consider a bivariate normal linear regression model with an inequality restriction imposed on one of the regression coefficients. The exact analytical expressions for the density and distribution functions of the inequality constrained and pre-test estimators are derived and numerically evaluated. The implications of using the inequality constrained and pre-test estimators in confidence interval construction are also discussed and explored.

Estimation of the regression error variance after a preliminary test of an inequality constraint on the coefficient vector is considered. We derive the exact finite sample risk of several inequality restricted and pre-test estimators of σ2. These estimators are associated with the maximum likelihood, least squares and minimum mean squared error component estimators. Optimal critical values for the pre-test according to a mini-max regret criterion are numerically calculated. Furthermore, we examine the robustness of the optimal choice of critical values and the risk properties of the estimators of σ2 to model mis-specification through the exclusion of relevant regressors.

We derive and numerically evaluate the bias and mean square error of the inequality constrained least squares estimator in a model with two inequality constraints and multivariate terror terms. Our results suggest that qualitatively, the estimator properties found for models with normal errors carry over to the case of multivariate terrors.

The size and power of the F test for testing the joint significance of regression coefficients is considered in the context of an omitted variable model with an interval constraint. It is assumed that the F test is based on the interval constrained least squars (INCLS) estimator, and exact power of the test is computed by numerical integrations. Our results show that when the moddle is correctly specified, the test based on the INCLS estimator is not necessarily more powerful than the test by the OLS estimator even if the constraint is valid. On the contrary, over a large region in the parameter space, the test based on the OLS estimator has considerably higher power than the test based on the INCLS estimator. When the model is sufficiently mis-specified, we find that both tests, by the OLS as well as the INCLS estimators, always reject the null hypothesis even if it is true.

Summary Optimal critical values are derived for a pre-test of an inequality restriction in a model where relevant regressors are unwittingly omitted. The criterion adopted is either that of the minimum average relative risk, or that of the mini-max regret. The latter approach yields an optimal critical value which is sensitive to the degree of model mis-specification, while the former criterion always leads to the choice of the unrestricted estimator.