The exact discrete model satisfied by equispaced data generated by a linear stochastic differential equations system is derived by a method that does not imply restrictions on observed discrete data per se. The method involves integrating the solution of the continuous time model in state space form and a nonstandard change in the order of three types of integration, facilitating the representation of the exact discrete model as an asymptotically time-invariant vector autoregressive moving average model. The method applying to the state space form is general and is illustrated using the prototypical higher order model for mixed stock and flow data discussed by Bergstrom (1986, Econometric Theory 2, 350–373).

An important reason for analyzing panel data is to observe the dynamic nature of an economic variable separately from its time-invariant unobserved heterogeneity. This paper examines how to estimate the autocovariances of a variable separately from its time-invariant unobserved heterogeneity. When both cross-sectional and time series sample sizes tend to infinity, we show that the within-group autocovariances are consistent, although they are severely biased when the time series length is short. The biases have the leading term that converges to the long-run variance of the individual dynamics. This paper develops methods to estimate the long-run variance in panel data settings and to alleviate the biases of the within-group autocovariances based on the proposed long-run variance estimators. Monte Carlo simulations reveal that the procedures developed in this paper effectively reduce the biases of the estimators for small samples.

Following the approach proposed by Gregoir and Laroque (1993, Econometric Theory 9, 329–342), we consider a class of multivariate processes that, when differenced enough, yields covariance stationary processes whose determinant of the matrix series associated with their Wold representation has various unit roots with various orders of multiplicity we restrict to be integers. A representation theorem is provided that involves different polynomial error correction terms at each frequency associated with each unit root. An identification criterion for each set of error correction terms is proposed.

We investigate the finite-sample bias of the quasi-maximum likelihood estimator (QMLE) in spatial autoregressive models with possible exogenous regressors. We derive the approximate bias result of the QMLE in terms of model parameters and also the moments (up to order 4) of the error distribution, and thus a feasible bias-correction procedure is directly applicable. In some special cases, the analytical bias result can be significantly simplified. Our Monte Carlo results demonstrate that the feasible bias-correction procedure works remarkably well.

Commonly used tests to assess evidence for the absence of autocorrelation in a univariate time series or serial cross-correlation between time series rely on procedures whose validity holds for i.i.d. data. When the series are not i.i.d., the size of correlogram and cumulative Ljung–Box tests can be significantly distorted. This paper adapts standard correlogram and portmanteau tests to accommodate hidden dependence and nonstationarities involving heteroskedasticity, thereby uncoupling these tests from limiting assumptions that reduce their applicability in empirical work. To enhance the Ljung–Box test for non-i.i.d. data, a new cumulative test is introduced. Asymptotic size of these tests is unaffected by hidden dependence and heteroskedasticity in the series. Related extensions are provided for testing cross-correlation at various lags in bivariate time series. Tests for the i.i.d. property of a time series are also developed. An extensive Monte Carlo study confirms good performance in both size and power for the new tests. Applications to real data reveal that standard tests frequently produce spurious evidence of serial correlation.

Mundlak (1978,Econometrica 46, 69–85) showed that the fixedeffects estimator can be obtained as generalized least squares (GLS)for a panel regression model where the individual effects are randombut are all hopelessly correlated with theregressors. This result was obtained by partitioned inversion aftersubstituting the reduced form expression for the individual effectsas a function of the means of all the regressors.This note shows that Mundlak's result can be obtained usingsystem estimation without using partitionedinversion. System estimation has proved useful for derivingtwo-stage least squares (2SLS) and three-stage least squares (3SLS)counterparts for the random effects panel models by Baltagi (1981, Journal ofEconometrics 17, 189–200). It also has been used forobtaining an alternative derivation of the Hausman tests that isrobust to heteroskedasticity of unknown form (see Arellano, 1993, Journal ofEconometrics 59, 87–97) and more recently, forobtaining generalized method of moments (GMM) estimators for dynamicpanel models (see Arellano and Bover, 1995, Journal of Econometrics 68, 29–51;and Blundell and Bond, 1998,Journal of Econometrics 87, 115–143, to mentiona few). We also show that a necessary and sufficient condition forordinary least squares (OLS) to be equivalent to GLS is satisfiedfor this model.

We propose a model-free test for structural changes in factor models. The basic idea is to regress the data on commonly estimated factors by local smoothing and compare the fitted values of time-varying factor loadings with those of time-invariant factor loadings estimated via principal component analysis. By construction, the test is designed to be powerful against both smooth structural changes and sudden structural breaks with a possibly unknown number of breaks and unknown break dates in the factor loadings. No restrictions on the form of alternatives or trimming of boundary regions near the beginning or end of the sample period is required for the test. The test has power to detect the usual nonparametric rate of local alternatives. Monte Carlo studies demonstrate excellent power of the test in detecting both smooth and sudden structural changes in the factor loadings. In an application using U.S. asset returns, we find significant evidence against time-invariant factor loadings.

This paper uses the concept of dual likelihood to develop some higher order asymptotic theory for the empirical likelihood ratio test for parameters defined implicitly by a set of estimating equations. The resulting theory is likelihood based in the sense that it relies on methods developed for ordinary parametric likelihood models to obtain valid Edgeworth expansions for the maximum dual likelihood estimator and for the dual/empirical likelihood ratio statistic. In particular, the theory relies on certain Bartlett-type identities that can be used to produce a simple proof of the existence of a Bartlett correction for the dual/empirical likelihood ratio. The paper also shows that a bootstrap version of the dual/empirical likelihood ratio achieves the same higher order accuracy as the Bartlett-corrected dual/empirical likelihood ratio.This paper is based on Chapter 2 of my Ph.D. dissertation at the University of Southampton. Partial financial support under E.S.R.C. grant R00429634019 is gratefully acknowledged. I thank my supervisor, Grant Hillier, for many stimulating conversations and Peter Phillips, Andrew Chesher, and Jan Podivisnky for some useful suggestions. In addition, I am very grateful to the co-editor Donald Andrews and two referees for many valuable comments that have improved noticeably the original draft. All remaining errors are my own responsibility.

In this paper we extend fixed-b asymptotic theory to the nonparametric Phillips–Perron (PP) unit root tests. We show that the fixed-b limits depend on nuisance parameters in a complicated way. These nonpivotal limits provide an alternative theoretical explanation for the well-known finite-sample problems of the PP tests. We also show that the fixed-b limits depend on whether deterministic trends are removed using one-step or two-step detrending approaches. This is in contrast to the asymptotic equivalence of the one- and two-step approaches under a consistency approximation for the long-run variance estimator. Based on these results we introduce modified PP tests that allow for asymptotically pivotal fixed-b inference. The theoretical analysis is cast in the framework of near-integrated processes, which allows us to study the asymptotic behavior both under the unit root null hypothesis and for local alternatives. The performance of the original and modified PP tests is compared by means of local asymptotic power and a small finite-sample simulation study.

We obtain uniform consistency results for kernel-weighted sample covariances in a nonstationary multiple regression framework that allows for both fixed design and random design coefficient variation. In the fixed design case these nonparametric sample covariances have different uniform asymptotic rates depending on direction, a result that differs fundamentally from the random design and stationary cases. The uniform asymptotic rates derived exceed the corresponding rates in the stationary case and confirm the existence of uniform super-consistency. The modelling framework and convergence rates allow for endogeneity and thus broaden the practical econometric import of these results. As a specific application, we establish uniform consistency of nonparametric kernel estimators of the coefficient functions in nonlinear cointegration models with time varying coefficients or functional coefficients, and provide sharp convergence rates. For the fixed design models, in particular, there are two uniform convergence rates that apply in two different directions, both rates exceeding the usual rate in the stationary case.

This paper presents a consistent specification test of conditional symmetry using a kernel method. The test statistic is shown to be asymptotically distributed as standard normal under the null hypothesis of conditional symmetry and consistent against any conditional asymmetric distribution. Power against local alternatives is also investigated. A Monte Carlo simulation is provided to evaluate the finite-sample performance of the test.

In this paper I propose estimating distributions on the unit interval semi-nonparametrically using orthonormal Legendre polynomials. This approach will be applied to the interval-censored mixed proportional hazard (ICMPH) model, where the distribution of the unobserved heterogeneity is modeled semi-nonparametrically. Various conditions for the nonparametric identification of the ICMPH model are derived. I will prove general consistency results for M-estimators of (partly) non-euclidean parameters under weak and easy-to-verify conditions and specialize these results to sieve estimators. Special attention is paid to the case where the support of the covariates is finite.

This paper provides weak and strong laws of largenumbers for weakly dependent heterogeneous randomvariables. The weak laws of large numbers presentedextend known results to the case of trended randomvariables. The main feature of our strong law oflarge numbers for mixingale sequences is the lessstrict decay rate that is imposed on the mixingalenumbers as compared to previous results.

The efficiency of the generalized method of moment (GMM) estimator is addressed by using a characterization of its variance as an inner product in a reproducing kernel Hilbert space. We show that the GMM estimator is asymptotically as efficient as the maximum likelihood estimator if and only if the true score belongs to the closure of the linear space spanned by the moment conditions. This result generalizes former ones to autocorrelated moments and possibly infinite number of moment restrictions. Second, we derive the semiparametric efficiency bound when the observations are known to be Markov and satisfy a conditional moment restriction. We show that it coincides with the asymptotic variance of the optimal GMM estimator, thus extending results by Chamberlain (1987, Journal of Econometrics 34, 305–33) to a dynamic setting. Moreover, this bound is attainable using a continuum of moment conditions.

Forecasts from a univariate autoregressive model estimated by OLS are unbiased, irrespective of whether the model fitted has the correct order; this property only requires symmetry of the distribution of the innovations. In this paper, this result is generalized to vector autoregressions and a wide class of multivariate stochastic processes (which include Gaussian stationary multivariate stochastic processes) is described for which unbiasedness of predictions holds: specifically, if a vector autoregression of arbitrary finite order is fitted to a sample from any process in this class, the fitted model will produce unbiased forecasts, in the sense that the prediction errors have distributions symmetric about zero. Different numbers of lags may be used for each variable in each autoregression and variables may even be missing, without unbiasedness being affected. This property is exact in finite samples. Similarly, the residuals from the same autoregressions have distributions symmetric about zero.

Invariant polynomials with matrix arguments have been defined by the theory of group representations, generalizing the zonal polynomials. They have developed as a useful tool to evaluate certain integrals arising in multivariate distribution theory, which were expanded as power series in terms of the invariant polynomials. Some interest in the polynomials has been shown by people working in the field of econometric theory. In this paper, we shall survey the properties of the invariant polynomials and their applications in multivariate distribution theory including related developments in econometrics.

There are many instances in the statistical literature in which inference is based on a normalized quadratic form in a standard normal vector, normalized by the squared length of that vector. Examples include both test statistics (the Durbin–Watson statistic) and estimators (serial correlation coefficients). Although much studied, no general closed-form expression for the density function of such a statistic is known. This paper gives general formulae for the density in each open interval between the characteristic roots of the matrix involved. Results are given for the case of distinct roots, which need not be assumed positive, and when the roots occur with multiplicities greater than one. Starting from a representation of the density as a surface integral over an (n − 2)-dimensional hyperplane, the density is expressed in terms of top-order zonal polynomials involving difference quotients of the characteristic roots of the matrix in the numerator quadratic form.