GEL statistics under weak identification☆
Introduction
In many situations empirical researchers are confronted with instrumental variables only weakly correlated with the endogenous variables with the consequence that structural parameters are only weakly identified. This paper contributes to what is now a voluminous literature on the topic of partial and weak identification initiated in Sargan (1983), Phillips (1989), and Staiger and Stock (1997). Those papers demonstrate that the correspondence between the limiting and finite-sample behaviours of estimators and test statistics may be poor in such circumstances.1 The particular aim of this paper then is the practical recommendation of test statistics appropriate for nonlinear moment condition models with time series data that possess finite sample size properties relatively unaffected by the weakness or strength of identification and that also display good power. Consequently their acceptance regions may be inverted to provide reliable and robust confidence regions for parameters of interest.
The contributions of the paper are as follows. We propose new test statistics for a time series setting based on the implied probabilities generated by generalised empirical likelihood (GEL) estimation, see, e.g., Smith, 1997, Smith, 2011 and Newey and Smith (2004), appropriate for circumstances in which identification may be weak as discussed in Stock and Wright (2000), henceforth SW. These statistics generalise to the time series context versions of the Pearson-type GEL statistics suggested in Ramalho and Smith (2004) for independently and identically distributed (i.i.d.) observations and are defined in terms of smoothed moment indicators obtained using kernel function based weights which incorporate a bandwidth parameter; cf. Kitamura and Stutzer (1997) and Smith, 1997, Smith, 2000, Smith, 2011. We also reconsider the statistics developed in Guggenberger and Smith (2005), hereafter GSa, for i.i.d. observations and those proposed for time series data in Guggenberger and Smith (2008), henceforth GSb, that use a truncated kernel function for the smoothed moment indicators. In particular, since their finite sample behaviour may be affected by the choice of kernel function, we reformulate the GSb statistics for a general form of kernel function. Likewise, our Pearson-type GEL statistics also incorporate a general kernel function. The tests discussed here deal with both simple hypotheses on the full parameter vector and composite hypotheses on a subvector of parameters, the latter case, apart from one particular statistic, requiring that the parameters not under test be strongly identified, an assumption adopted in other papers, see, e.g., Kleibergen (2005), GSa, Otsu (2006) and GSb. In addition, and importantly for our conclusions, we propose GEL time series counterparts to the GMM conditional likelihood ratio (CLR) statistic in Kleibergen (2005) and the GEL CLR statistics developed in Smith (2007) for the i.i.d context; cf. Moreira (2003).
We investigate the large sample behaviour of the statistics proposed here at fixed alternative values for the weakly identified parameters, a scenario that permits a comparison of their relative asymptotic power properties in the weakly identified context. Importantly, under the null hypothesis, the statistics proposed here, in GSb and elsewhere, e.g., in Kleibergen (2005) and Otsu (2006), are asymptotically (conditionally) pivotal, unlike the corresponding Wald and likelihood ratio statistics. Indeed, with the notable exception of the GEL CLR statistics, whose large sample conditional null distribution may be straightforwardly simulated, our statistics are asymptotically chi-square under the null hypothesis. Thus, the empirical rejection probabilities under the null hypothesis (NRPs) of tests formed from these statistics may be expected to be relatively invariant to the strength or weakness of identification. Our elucidation of the asymptotic properties of the statistics also demonstrates certain asymptotic equivalences among classes of statistics. Hence, without recourse to higher order analysis, reliance must be placed on simulation experiments for statistical choice. Consequently, in order to inform our recommendations for econometric practice, we undertake a comprehensive Monte Carlo study to assess the finite sample size and power characteristics of the statistics proposed here based on a similar design to that in GSb. For comparative purposes we also include in this study a number of additional tests favourably reported in the recent weak instrument literature; see Kleibergen (2005), Otsu (2006) and GSb. To anticipate our conclusions, particular implementations of the GEL CLR and Pearson-type tests have satisfactory finite sample size properties and power characteristics that are very competitive with existing tests.
The paper is organised as follows. Section 2 reviews GEL estimation and associated constructs. Statistics appropriate for tests on the full vector of parameters are introduced in Section 3 and their limiting behaviour stated whereas those appropriate for the subvector case are discussed in Section 4. Section 5 describes the results of the Monte Carlo study. Section 6 concludes. The appendices contain the technical assumptions and proofs of results contained in the text.
The following notation is used in the paper. The symbols “”, “” and “” denote convergence in distribution, convergence in probability and weak convergence of empirical processes, respectively, with “” indicating that the limiting distribution of the left hand side expression is bounded above by that of the right hand side. Convergence “with probability approaching 1” is written as “w.p.a.1”. The space contains all functions that are times continuously differentiable on the set . The notation stands for the column vectorisation of the matrix , i.e., if then , “” denotes the transpose matrix of and the element in the -th row and -th column. Furthermore, for a symmetric matrix , “” means that is positive definite, and are the smallest and largest eigenvalues of , respectively, and . The vector inequality is to be interpreted as the inequality holding element-wise. For a full column rank matrix and , we denote by the oblique projection matrix and define , where is the -dimensional identity matrix. We abbreviate this notation to and if and, if , set . Finally and respectively denote central chi-square and noncentral chi-square random variables with degrees of freedom and noncentrality parameter .
Section snippets
Generalised empirical likelihood
This section outlines GEL estimation for time series observations based on kernel smoothed counterparts of the moment indicators; see Kitamura and Stutzer (1997) and Smith, 1997, Smith, 2011.2 For i.i.d. observations, Newey and Smith (2004) show that GEL estimators possess attractive higher order properties as compared to those based on GMM (Hansen, 1982, Newey, 1985, Newey and West, 1987).
Full vector test statistics
The primary concern of this section is the provision of robust confidence regions for in a weakly identified and time series setting by the consideration of tests of the simple null hypothesis against the composite alternative .
We obtain versions of the Pearson-type class of statistics due to Ramalho and Smith (2004) for these circumstances utilising the implied probabilities , of (2.10). We also adapt the statistics considered in GSb for the general kernel
Subvector test statistics
Our concern now is the construction of robust confidence regions for the subvector of where . Consequently our interest is in testing
As in much of the literature, see, e.g., Kleibergen (2005), GSa and GSb, a critical limitation of our analysis, with the exception of the GEL criterion function statistic (4.5) discussed below, is that we require the parameter vector not under test to be strongly identified. Thus, only the subvector (say) of
Monte Carlo study
This section focuses on tests for the null hypothesis of (3.1) against the alternative hypothesis ; see Section 3. In particular, it reports on a comprehensive Monte Carlo study that assesses the finite sample size and power characteristics of the new tests suggested in Section 3 whose asymptotic behaviour is detailed in Theorem 3.1, Theorem 3.2, Theorem 3.3 above. For comparative purposes we include a number of additional tests that have been shown to perform favourably in the
Concluding remarks
The main purpose of this paper is the recommendation of various tests appropriate for inference in a weakly identified time series environment.
A comprehensive Monte Carlo study compares the tests proposed here with a number of other candidate tests that the literature has suggested possess satisfactory finite sample size and power properties. On the basis of these experiments, in terms of results for both size and power for nonrandom choices of bandwidth and partition number , the GEL test
References (62)
Chi-square diagnostic tests for econometric models: introduction and applications
Journal of Econometrics
(1988)- et al.
Applications of subsampling, hybrid, and size-correction methods
Journal of Econometrics
(2010) - et al.
Exactly distribution-free inference in instrumental variables regression with possibly weak instruments
Journal of Econometrics
(2008) - et al.
On the efficient use of the informational content of estimating equations: implied probabilities and euclidean empirical likelihood
Journal of Econometrics
(2007) - et al.
A new method of projection-based inference in GMM with weakly identified nuisance parameters
Journal of Econometrics
(2011) - et al.
Inferring the rank of a matrix
Journal of Econometrics
(1997) - et al.
Generalized empirical likelihood tests in time series models with potential identification failure
Journal of Econometrics
(2008) - et al.
Generalized reduced rank tests using the singular value decomposition
Journal of Econometrics
(2006) - et al.
Bootstrap validity for the score test when instruments may be weak
Journal of Econometrics
(2009) GMM, GEL, serial correlation and asymptotic bias
Econometrica
(2005)
Estimators of the parameters of a single equation in a complete set of stochastic equations
Annals of Mathematical Statistics
Chi-square diagnostic tests for econometric models: theory
Econometrica
Heteroskedasticity and autocorrelation consistent covariance matrix estimation
Econometrica
Hybrid and size-corrected subsampling methods
Econometrica
An improved heteroskedasticity and autocorrelation consistent covariance matrix estimator
Econometrica
Optimal invariant similar tests for instrumental variables regression
Econometrica
Inference with weak instruments
Implied probabilities in GMM estimators
Econometrica
Efficient semiparametric estimation of expectations
Econometrica
Generalized method of moments, efficient bootstrapping, and improved inference
Journal of Business and Economic Statistics
Exponential tilting with weak instruments: estimation and testing
Oxford Bulletin of Economics and Statistics
Consistent estimation with a large number of weak instruments
Econometrica
On the asymptotic properties of ldu-based tests of the rank of a matrix
Journal of the American Statistical Association
Some impossibility theorems in econometrics with applications to structural and dynamic models
Econometrica
Identification, weak instruments and statistical inference in econometrics. Presidential Address to the Canadian Economics Association
Canadian Journal of Economics
Projection–based statistical inference in linear structural models with possibly weak instruments
Econometrica
Generalized empirical likelihood estimators and tests under partial, weak and strong identification
Econometric Theory
Large sample properties of generalized method of moments estimators
Econometrica
Cited by (14)
Generic results for establishing the asymptotic size of confidence sets and tests
2020, Journal of EconometricsCitation Excerpt :This arises in tests of stochastic dominance and CS’s based on conditional moment inequalities, e.g., see Andrews and Shi (2013, 2014) and Linton et al. (2010). Selected references in the literature regarding uniformity issues in the models discussed above include the following: for unit roots, Bobkowski (1983), Cavanagh (1985), Chan and Wei (1987), Phillips (1987), Stock (1991), Park (2002), Giraitis and Phillips (2006), Phillips and Magdalinos (2007), and Andrews and Guggenberger (2012); for weak identification due to weak IV’s, Staiger and Stock (1997), Stock and Wright (2000), Moreira (2003), Kleibergen (2005), Guggenberger et al. (2012a), Guggenberger et al. (2012b, 2019); for weak identification in other models, Andrews and Cheng (2012a, 2013, 2014), Qu (2014), Andrews and Mikusheva (2015, 2016), Cox (2016), and Han and McCloskey (2019); for parameters near a boundary, Chernoff (1954), Self and Liang (1987), Shapiro (1989), Geyer (1994), Andrews (1999, 2001, 2002), Andrews and Guggenberger (2010b), and McCloskey (2017); and for post-model selection inference, Kabaila (1995), Leeb and Pötscher (2005), Leeb (2006), Andrews and Guggenberger (2009a, b) and McCloskey (2017). This paper is organized as follows.
Asymptotic size of kleibergen's LM and conditional LR tests for moment condition models
2017, Econometric TheoryIdentification- and singularity-robust inference for moment condition models
2019, Quantitative Economics
- ☆
We are grateful for the helpful and constructive comments on and criticisms of previous versions of the paper by the Editors, a Co-Editor, two referees and S. Chaudhuri. Earlier versions of this paper were presented at the Tinbergen Institute, Amsterdam, the CIREQ Conference on Generalized Method of Moments, Université de Montréal, 2007, the Econometric Society European Meetings, Milan, 2008, the Conference in Honour of M.R. Wickens, University of York, 2008, the University of Hong Kong and the New Zealand Econometric Study Group Conference 2011, University of Otago, Dunedin.