This paper shows what quantile regressions identify for general structural functions. Under fairly mild conditions, the quantile partial derivative identifies a weighted average of heterogeneous structural partial effects among the subpopulation of individuals at the conditional quantile of interest. This result justifies the use of quantile regressions as means of measuring heterogeneous causal effects for a general class of structural functions with multiple unobservables.

We provide novel characterizations of multivariate normality that incorporate both the characteristic function and the moment generating function, and we employ these results to construct a class of affine invariant, consistent and easy-to-use goodness-of-fit tests for normality. The test statistics are suitably weighted L2-statistics, and we provide their asymptotic behavior both for i.i.d. observations as well as in the context of testing that the innovation distribution of a multivariate GARCH model is Gaussian. We also study the finite-sample behavior of the new tests and compare the new criteria with alternative existing tests.

In an hypothesis testing problem involving nuisance parameters for which boundedly complete sufficient statistics exist under the null hypothesis, the class of all similar regions for the problem is characterized by the conditional distribution of the data given these sufficient statistics. If there exists a one-to-one transformation y → (t, u) of the data, y, to the sufficient statistic, t, and a second vector of statistics, u, that is independent of t under the null hypothesis, then the statistic u itself characterizes the class of similar regions. This paper applies this idea to five testing problems of interest in econometrics. In each case we obtain the density of the relevant statistic under the null hypothesis, when it is free of nuisance parameters, and under the alternative. Using the density under the alternative, we discuss the power properties of the class of similar tests for each problem. Other applications are also suggested.

We investigate the possible bimodality of the density of the two-stage least squares (TSLS) estimator in a just-identified/overidentified linear structural equation. By studying the interaction between weakness of instruments, degree of endogeneity, and degree of overidentification we are able to identify conditions for its existence.I thank Grant Hillier, Patrick Marsh, Don Poskitt, the editor Peter Phillips, and two anonymous referees for useful and encouraging comments.

Our subject is the notion of automated discovery in econometrics. Advances in computer power, electronic communication, and data collection processes have all changed the way econometrics is conducted. These advances have helped to elevate the status of empirical research within the economics profession in recent years, and they now open up new possibilities for empirical econometric practice. Of particular significance is the ability to build econometric models in an automated way according to an algorithm of decision rules that allow for (what we call here) heteroskedastic and autocorrelation robust (HAR) inference. Computerized search algorithms may be implemented to seek out suitable models, thousands of regressions and model evaluations may be performed in seconds, statistical inference may be automated according to the properties of the data, and policy decisions can be made and adjusted in real time with the arrival of new data. We discuss some aspects and implications of these exciting, emergent trends in econometrics.The first version of this paper was written in April 2004 for the 20th Anniversary Issue of Econometric Theory. Helpful comments by the co-editor, Oliver Linton, Benno Pötscher, Brendan Beare, and two referees on the first draft are gratefully acknowledged.

The contribution of this paper is two-fold. First, we derive the asymptotic null distribution of the familiar augmented Dickey-Fuller [ADF] statistics in the case where the shocks follow a linear process driven by infinite variance innovations. We show that these distributions are free of serial correlation nuisance parameters but depend on the tail index of the infinite variance process. These distributions are shown to coincide with the corresponding results for the case where the shocks follow a finite autoregression, provided the lag length in the ADF regression satisfies the same o(T1/3) rate condition as is required in the finite variance case. In addition, we establish the rates of consistency and (where they exist) the asymptotic distributions of the ordinary least squares sieve estimates from the ADF regression. Given the dependence of their null distributions on the unknown tail index, our second contribution is to explore sieve wild bootstrap implementations of the ADF tests. Under the assumption of symmetry, we demonstrate the asymptotic validity (bootstrap consistency) of the wild bootstrap ADF tests. This is done by establishing that (conditional on the data) the wild bootstrap ADF statistics attain the same limiting distribution as that of the original ADF statistics taken conditional on the magnitude of the innovations.

In this paper we develop nonparametric estimators of the joint time series data generating process (DGP) of (xt, yt) at different t-values, of conditional DGP, of the conditional mean of xt given the past values of x and y, and, more generally, the conditional mean of (xt, yt) given their past values (vector autoregression). We establish, among other results, the central limit theorems for these estimators under far weaker mixing conditions than those used in Robinson [23], where only the xt series is considered. Uniform consistency and rate results for the consistencies of various estimators are also obtained. The results of the paper are useful in light of the fact that often the functional form of the dynamic regression is not known and also the assumption of the Gaussian process is not true.

The invariant polynomials (Davis [8] and Chikuse [2] with r(r ≥ 2) symmetric matrix arguments have been defined, extending the zonal polynomials, and applied in multivariate distribution theory. The usefulness of the polynomials has attracted the attention of econometricians, and some recent papers have applied the methods to distribution theory in econometrics (e.g., Hillier [14] and Phillips [22]).

The ‘top order’ invariant polynomials , in which each of the partitions of ki 1 = 1,…,r, and has only one part, occur frequently in multivariate distribution theory (e.g., Hillier and Satchell [17] and Phillips [27]). In this paper we give three methods of constructing these polynomials, extending those of Ruben [28] for the top order zonal polynomials. The first two methods yield explicit formulae for the polynomials and then we give a recurrence procedure. It is shown that some of the expansions presented in Chikuse and Davis [4] are simplified for the top order invariant polynomials. A brief discussion is given on the ‘lowest order’ invariant polynomials.

Two different methods for pooling time series of cross section data are used by economists. The first method, described by Kmenta, is based on the idea that pooled time series of cross sections are plagued with both heteroskedasticity and serial correlation.The second method, made popular by Balestra and Nerlove, is based on the error components procedure where the disturbance term is decomposed into a cross-section effect, a time-period effect, and a remainder.Although these two techniques can be easily implemented, they differ in the assumptions imposed on the disturbances and lead to different estimators of the regression coefficients. Not knowing what the true data generating process is, this article compares the performance of these two pooling techniques under two simple setting. The first is when the true disturbances have an error components structure and the second is where they are heteroskedastic and time-wise autocorrelated.

First, the strengths and weaknesses of the two techniques are discussed. Next, the loss from applying the wrong estimator is evaluated by means of Monte Carlo experiments. Finally, a Bartletfs test for homoskedasticity and the generalized Durbin-Watson test for serial correlation are recommended for distinguishing between the two error structures underlying the two pooling techniques.

Shrinkage estimation procedures such as ridge regression and the lasso have been proposed for stabilizing estimation in linear models when high collinearity exists in the design. In this paper, we consider asymptotic properties of shrinkage estimators in the case of “nearly singular” designs.I thank Hannes Leeb and Benedikt Pötscher and also the referees for their valuable comments. This research was supported by a grant from the Natural Sciences and Engineering Research Council of Canada.

In this paper we consider the small sample properties of the coefficient of determination in a linear regression model with multivariate t errors when proxy variables are used instead of unobservable regressors. The results show that if the unobservable variable is an important variable, the adjusted coefficient of determination can be more unreliable in small samples than the unadjusted coefficient of determination from both viewpoints of the bias and the MSE.

We propose additive functional-based nonstationarity tests that exploit the different divergence rates of the occupation times of a (possibly nonlinear) process under the null of nonstationarity (stationarity) versus the alternative of stationarity (nonstationarity). We consider both discrete-time series and continuous-time processes. The discrete-time case covers Harris recurrent Markov chains and integrated processes. The continuous-time case focuses on Harris recurrent diffusion processes. Notwithstanding finite-sample adjustments discussed in the paper, the proposed tests are simple to implement and rely on tabulated critical values. Simulations show that their size and power properties are satisfactory. Our robustness to nonlinear dynamics provides a solution to the typical inconsistency problem between assumed linearity of a time series for the purpose of nonstationarity testing and subsequent nonlinear inference.

The central limit theorem in Davidson [2] is extended to allow cases where the variances of sequence coordinates can be tending to zero. A trade-off is demonstrated between the degree of dependence and the rate of degeneration. For the martingale difference case, it is sufficient for the sum of the variances to diverge at the rate of log n.

We provide rankings of academic institutions according to the publication record of their faculty members in the specialization of econometrics. The rankings are based on standardized page counts of articles published by faculty members in 14 journals over the period 1980 to 1985. Separate rankings of the 200 leading institutions are provided for their theoretical contributions, as well as their contributions in theory and applied work. The rankings are compared with a ranking in the general field of economics.

This paper studies the estimation of a nonparametric function ϕ from the inverse problem r = Tϕ given estimates of the function r and of the linear transform T. We show that rates of convergence of the estimator are driven by two types of assumptions expressed in a single Hilbert scale. The two assumptions quantify the prior regularity of ϕ and the prior link existing between T and the Hilbert scale. The approach provides a unified framework that allows us to compare various sets of structural assumptions found in the econometric literature. Moreover, general upper bounds are also derived for the risk of the estimator of the structural function ϕ as well as that of its derivatives. It is shown that the bounds cover and extend known results given in the literature. Two important applications are also studied. The first is the blind nonparametric deconvolution on the real line, and the second is the estimation of the derivatives of the nonparametric instrumental regression function via an iterative Tikhonov regularization scheme.

In Perez-Amaral, Gallo, and White (2003, Oxford Bulletin of Economics and Statstics 65, 821–838), the authors proposed an automatic predictive modeling tool called relevant transformation of the inputs network approach (RETINA). It is designed to embody flexibility (using nonlinear transformations of the predictors of interest), selective search within the range of possible models, control of collinearity, out-of-sample forecasting ability, and computational simplicity. In this paper we compare the characteristics of RETINA with PcGets, a well-known automatic modeling method proposed by David Hendry. We point out similarities, differences, and complementarities of the two methods. In an example using U.S. telecommunications demand data we find that RETINA can improve both in- and out-of-sample over the usual linear regression model and over some models suggested by PcGets. Thus, both methods are useful components of the modern applied econometrician's automated modeling tool chest.

Adaptive maximum likelihood estimators of unit roots in autoregressive processes with possibly non-Gaussian innovations are considered. Unit root tests based on the adaptive estimators are constructed. Limiting distributions of the test statistics are derived, which are linear combinations of two functionals of Brownian motions. A Monte Carlo simulation reveals that the proposed tests have improved powers over the classical Dickey–Fuller tests when the distribution of the innovation is not close to normal. We also compare the proposed tests with those of Lucas (1995, Econometric Theory 11, 331–346) based on M-estimators.

This article deals with the derivation of the exact discrete model that corresponds to a closed linear first-order continuous-time system with mixed stock and flow data. This exact discrete model is (under appropriate additional conditions) a stationary autoregressive moving average time series model and may allow one to obtain asymptotically efficient estimators of the parameters describing the continuous-time system.

We provide a proof of the consistency and asymptotic normality of the estimator suggested by Heckman (1990, American Economic Review 80, 313–318) for the intercept of a semiparametrically estimated sample selection model. The estimator is based on “identification at infinity,” which leads to nonstandard convergence rate.