Skip to content
Licensed Unlicensed Requires Authentication Published by De Gruyter April 14, 2016

Size corrected Significance Tests in Seemingly Unrelated Regressions with Autocorrelated Errors

  • Spyridon D. Symeonides , Yiannis Karavias ORCID logo EMAIL logo and Elias Tzavalis

Abstract

Refined asymptotic methods are used to produce degrees-of-freedom- adjusted Edgeworth and Cornish-Fisher size corrections of the t and F testing procedures for the parameters of a S.U.R. model with serially correlated errors. The corrected tests follow the Student-t and F distributions, respectively, with an approximation error of order Oτ3, where τ=1/T and T is the number of time observations. Monte Carlo simulations provide evidence that the size corrections suggested hereby have better finite sample properties, compared to the asymptotic testing procedures (either standard or Edgeworth corrected), which do not adjust for the degrees of freedom.

JEL Classification: CIO; C12; D24

Appendix

In this appendix, we provide proofs of the main results of the paper. To prove these results, we rely on a number of lemmas, which are presented with their proofs in a Technical Appendix, available at the . This appendix is structured as follows: First, we introduce the stochastic order ω(·), which measures the approximation error of the asymptotic expansions given in the paper. Then, using the lemmas in the Technical Appendix, we provide the proofs of the theorems.

The Order ω(·)

Following Magdalinos (1992, 344), let I be a given set of indexes which, without loss of generality, can be considered to belong to the open interval (0,1). For any collection of real-valued stochastic quantities (scalars, vectors, or matrices) Yτ (τI), we write Yτ = ω(τi), if for any given n > 0, there exists a 0 < ε < ∞ such that

[69]PrYτ/τi>lnτε=oτn,

as τ → 0, where the ||·|| is the Euclidean norm. If eq. [69] is valid for any n > 0, we write Yτ= (ω). The use of this order of magnitude is motivated by the fact that, if two stochastic quantities differ by a quantity of order ω(τi), then, under general conditions, the distribution function of the one provides an asymptotic approximation of the distribution function of the other, with an error of order O(τi). Furthermore, orders ω(·) and O(·) have similar operational properties (Magdalinos (1992).

Asymptotic Expansions of Size Corrected Tests: Proofs of Theorems

Given the lemmas in the Technical Appendix, next we derive the proofs of the theorems presented in the main text. These are based on known expansions of standard normal and chi-square distributed tests. We derive new expansions of the degrees- of-freedom-adjusted versions of these tests, by inverting their characteristic functions. These degrees-of-freedom-adjusted approximations of distribution functions are proved to be locally exact.

Proof of Theorems 1 and 2

Approximation eq. [42] of Theorem 1 can be proved following the steps of the proof in Rothenberg (1988). The quantities in eq. [40] can be obtained by expanding the corresponding quantities given by Rothenberg and retaining the first term in each of these expansions. The approximation eq. [44] of Theorem 2 follows from the approximation eq. [42] and the following asymptotic approximations of the Student-t distribution and density functions, which are given in terms of the standard normal distribution and density functions, respectively (see Fisher (1925)):

[70]ITnx=Ixτ2/41+x2xix+Oτ4,iTnx=ix+Oτ2.

Note that approximation eq. [44] of Theorem 2 is locally exact. This can be easily seen as follows: If parameter vector γ=e,ς is known to belong to a ball of radius v, then, as v → 0, γ becomes a fixed known vector. By using eqs [27], [29], [33] and [35] we can prove that

[71]Λ=0,λ=κ=0,λ0=2,κ0=0.

Then, the analytic formulas of pi and p2, given in eq. [43], become

[72]p1=p2=0.

This result implies that, with an error of order O3), approximation eq. [44] becomes the Student-t distribution function with MT – n degrees of freedom. □

Proof of Theorem 3: We begin the proof by noticing that, under null hypothesis eq. [36], the t statistic, given by eq. [37], admits a stochastic expansion of the form

[73]t=t0+τt1+τ2t2+ωτ3,

where the first term in the expansion is given as

t0=eb/eGe1/2=hb,whereb=GXΩu/T.

The result given by eq. [73] implies that the Cornish-Fisher corrected statistic t*, given by eq. [47], admits a stochastic expansion of the form

[74]t=t0+τt1+τ2t2t3+ωτ3,

where

t3=p1+p2t02t0/2.

Let s be an imaginary number, and ψ(s) and ϕ(s) denote the characteristic functions of the t statistic, given by eq. [37], and a standard normal random variable, respectively. Using eq. [74] and the relationships:

Eexpst0t0=sϕsandEexpst0t03=3s+s3ϕs,

we can show that the characteristic function of the Cornish-Fisher corrected statistic t*, denoted as ψ *(s), can be approximated as follows:

ψs=ψsτ2sEexpst0t3+Oτ3=ψsτ22sp1s+p23s+s3ϕs+Oτ3.

Dividing ψ*(s) by –s, applying the inverse Fourier transform, and using Theorem 2, we can show that

[75]Prtx=Prtx+τ22p1+p2x2xiTnx+Oτ3=ITnxτ22p1+p2x2xiTnx+τ22p1+p2x2xiTnx+Oτ3=ITnx+Oτ3.

The last result means that the Cornish-Fisher corrected statistic t* is distributed as a Student-t random variable with MT – n degrees of freedom. □

Proof of Theorems 4 and 5

Approximation eq. [58] of Theorem 4 can be proved following the steps of the proof in Rothenberg (1984b). The quantities in eq. [56] can be obtained by expanding the corresponding quantities given by Rothenberg and retaining the first term in each of these expansions. Approximation eq. [60] of Theorem 5 follows from approximation eq. [58] and the following asymptotic approximations of the F distribution and density functions, which are given in terms of the chi-square distribution and density functions, respectively:

[76]FTnmx=Fmmx+τ2/2m2mxmxfmmx+Oτ4,fTnmx=mfmmx+Oτ2.

Note that approximation eq. [60] of Theorem 5 can be easily seen to be locally exact. By using eqs [71], [59], and [61], we can show that

[77]ξ1=mm2/2andξ2=mm+2/2
[78]q1=q2=0.

This result means that, with an error of order O(τ3), approximation eq. [60] becomes the F distribution function with m and MT – n degrees of freedom. □

Proof of Theorem 6

To begin the proof, we first notice that, under null hypothesis eq. [48], the F statistic, given by eq. [50], admits a stochastic expansion of the form

[78]F=F0+τF1+τ2F2+ωτ3,

where the first term in the expansion is

F0=bQb/m,b=GXΩu/T.

Equation [78] Implies that the Cornish-Fisher corrected statistic F*, given by eq. [64], admits a stochastic expansion of the form

[79]F=F0+τF1+τ2(F2F3)+ω(τ3),

where

F3=q1+q2F0F0.

Let s be an imaginary number, and ψ(s) and ϕ(s) now denote the characteristic functions of the F statistic, given by eq. [50], and a chi-square random variable with m degrees of freedom, respectively. Using eq. [79] and the following relationships:

EexpsF0F0=ϕm+2s/mandEexpsF0F02=m+2mϕm+4s/m,

we can show that the characteristic function of the Cornish-Fisher corrected statistic F*, denoted as ψ*(s), can be approximated as follows:

[80]ψs=ψsτ2sEexpsF0F3+Oτ3=ψsτ2sq1ϕm+2s/m+q2m+2mϕm+4s/m+Oτ3.

For the chi-square density/m(x), the following results can be shown:

[81]mxfmmx=mfm+2mxandmx2fmmx=mm+2fm+4mx.

Dividing eq. [80] by –s, applying the inverse Fourier transform, and using Theorem 5 and the results of eqs [76] and [81], we can show that

[82]Pr{Fx}=Pr{Fx}+τ2[(q1mfm+2(mx)+q2m+2mmfm+4(mx)]+O(τ3)=Pr{Fx}+τ2[(q1mxfm(mx)+q2mx2fm(mx)]+O(τ3)=Pr{Fx}+τ2(q1+q2x)mxfm(mx)+O(τ3)=FTnm(x)τ2(q1+q2x)xfTnm(x)+τ2(q1+q2x)xfTnm(x)+O(τ3)=FTnm(x)+O(τ3).

The last result implies that the Cornish-Fisher corrected statistic F is distributed as an F random variable with m and MT – n degrees of freedom. □

References

Beach, C. M., and J. G. MacKinnon. 1978. “A Maximum Likelihood Procedure for Regression with Autocorrelated Errors.” Econometrica 46 (l):51–8.10.2307/1913644Search in Google Scholar

Breusch, T. S. 1980. “Useful Invariance Results for Generalized Regression Models.” Journal of Econometrics 13:327–40.10.1016/0304-4076(80)90083-4Search in Google Scholar

Cornish, E. A., and R. A. Fisher. 1937. “Moments and Cumulants in the Specification of Distributions.” Revue Del’ Institute International De Statistic 4:1–14.10.2307/1400905Search in Google Scholar

Dhrymes, P. J. 1971. “Equivalence of Iterative Aitken and Maximum Likelihood Estimators for a System of Regression Equations.” Australian Economic Papers 10:2–24.10.1111/j.1467-8454.1971.tb00165.xSearch in Google Scholar

Dhrymes, P. J. 1978. Introductory Econometrics. New York: Springer Verlang.10.1007/978-1-4612-6292-3Search in Google Scholar

Fisher, R. A. 1925. “Expansion of Student’s Integral in Powers of n−1.” Metron 5:109–20.Search in Google Scholar

Fisher, R. A., and E. A. Cornish. 1960. “The Percentile of Distributions Having Known Cumulants.” Technometrics 2:209–25.10.1080/00401706.1960.10489895Search in Google Scholar

Hill, G. W., and A. W. Davis. 1968. “Generalized Asymptotic Expansions of Cornish- Fisher Type.” The Annals of Mathematical Statistics 39 (4):1264–73.10.1214/aoms/1177698251Search in Google Scholar

Iglesias, E. M., and G. D. Phillips. 2010. “The Bias to Order T 2 for the General k- Class Estimator in A Simultaneous Equation Model.” Economics Letters 109:42–5.10.1016/j.econlet.2010.07.011Search in Google Scholar

Iglesias, E. M., and G. D. Phillips. 2012. “Almost Unbiased Estimation in Simultaneous Equations Models with Strong and/or Weak Instruments.” Journal of Business and Economic Statistics 30 (4):505–20.10.1080/07350015.2012.715959Search in Google Scholar

Kiviet, J., and G. D. Phillips. 1996. “The Bias of the Ordinary Least Squares Estimator in Simultaneous Equation Models.” Economics Letters 53:161–7.10.1016/S0165-1765(96)00908-1Search in Google Scholar

Kiviet, J., G. D. Phillips, and B. Schipp. 1995. “The Bias of OLS, GLS, and ZEF Estimators in Dynamic Seemingly Unrelated Regression Models.” Journal of Econometrics 69:241–66.10.1016/0304-4076(94)01670-USearch in Google Scholar

Magdalinos, M. A. 1985. “Improving Some Instrumental Variables Test Procedures.” Econometric Theory 1:240–61.10.1017/S0266466600011166Search in Google Scholar

Magdalinos, M. A. 1992. “Stochastic Expansions and Asymptotic Approximations.” Econometric Theory 8:343–67.10.1017/S0266466600012974Search in Google Scholar

Magdalinos, M. A., and S. D. Symeonides. 1995. “Alternative Size Corrections for Some GLS Test Statistics: The Case of the AR(1) Model.” Journal of Econometrics 66:35–59.10.1016/0304-4076(94)01607-2Search in Google Scholar

Magee, L. 1985. “Efficiency of Iterative Estimators in the Regression Model with AR(1) Disturbances.” Journal of Econometrics 29:275–87.10.1016/0304-4076(85)90156-3Search in Google Scholar

McDonald, G. C., and D. I. Galarneau. 1975. “A Monte Carlo Evaluation of Some Ridge-Type Estimators.” Journal of the American Statistical Association 70 (350):407–16.10.1080/01621459.1975.10479882Search in Google Scholar

Ogasawara, H. 2012. “Cornish-Fisher Expansions Using Sample Cumulants and Monotoning Transformations.” Journal of Multivariate Analysis 103:1–18.10.1016/j.jmva.2011.05.013Search in Google Scholar

Parks, R. W. 1967. “Efficient Estimation of a System of Regression Equations When Disturbances Are Both Serially and Contemporaneously Correlated.” Journal of the American Statistical Association 62:500–9.10.1080/01621459.1967.10482923Search in Google Scholar

Phillips, G. D. 2000. “An Alternative Approach to Obtaining Nagar-Type Moment Approximations in Simultaneous Equation Models.” Journal of Econometrics 97:345–64.10.1016/S0304-4076(99)00075-5Search in Google Scholar

Phillips, G. D. 2007. “Nagar-Type Moment Approximations in Simultaneous Equation Models: Some Further Results.” In The Refinement of Econometric Estimation and Test Procedures: Finite Sam,Pie and Asymptotic Analysis, edited by G. D. A. Phillips and E. Tzavalis, 1–1. Cambridge: Cambridge University Press.10.1017/CBO9780511493157Search in Google Scholar

Prais, S. J., and C. I. Winsten. 1954. “Trend Estimators and Serial Correlation.” Cowles Commission Discussion Paper: Statistics No. 383.Search in Google Scholar

Rothenberg, T. J. 1984a. “Approximate Normality of Generalized Least Squares Estimates.” Econometrica 52 (4):811–25.10.2307/1911185Search in Google Scholar

Rothenberg, T. J. 1984b. “Hypothesis Testing in Linear Models When the Error Covariance Matrix Is Nonscalar.” Econometrica 52 (4):827–42.10.2307/1911186Search in Google Scholar

Rothenberg, T. J. 1988. “Approximate Power Functions for Some Robust Tests of Regression Coefficients.” Econometrica 56 (5):997–1019.10.2307/1911356Search in Google Scholar

Symeonides, S. D., E. Tzavalis, and H. Kandilorou. 2007. “Cornish-Fisher Size Corrected T and F Statistics for the Linear Regression Model with Heteroskedastic Errors.” In The Refinement of Econometric Estimation and Test Procedures: Finite Sample and Asymptotic Analysis, edited by G. D. A. Philips and E. Tzavalis, 173–204. Cambridge: Cambridge University Press.10.1017/CBO9780511493157.010Search in Google Scholar

Turkington, D. A.. 2000. “Generalised Vec Operators and the Seemingly Unrelated Regression Equations Model with Vector Correlated Disturbances.” Journal of Econometrics 99 (2):225–53.10.1016/S0304-4076(00)00025-7Search in Google Scholar

Ullah, A. 2004. Finite Sample Econometrics. Oxford: Oxford University Press.10.1093/0198774478.001.0001Search in Google Scholar

Zellner, A. 1962. “An Efficient Method of Estimating Seemingly Unrelated Regressions and Tests for Aggregation Bias.” Journal of the American Statistical Association 57 (298):348–68.10.1080/01621459.1962.10480664Search in Google Scholar

Zellner, A. 1963. “Estimators for Seemingly Unrelated Regression Equations: Some Exact Finite Sample Results.” Journal of the American Statistical Association 58 (304):977–92.10.1080/01621459.1963.10480681Search in Google Scholar

Zellner, A., and D. S. Huang. 1962. “Further Properties of Efficient Estimators for Seemingly Unrelated Regreesion Equations.” International Economic Review 3 (3):300–13.10.2307/2525396Search in Google Scholar

Zellner, A., and H. Theil. 1962. “Three-Stage Least Squares: Simultaneous Estimation of Simultaneous Equations.” Econometrica 30 (l):54–78.10.1007/978-94-011-2546-8_10Search in Google Scholar

Published Online: 2016-4-14
Published in Print: 2017-1-1

©2017 by De Gruyter

Downloaded on 29.3.2024 from https://www.degruyter.com/document/doi/10.1515/jtse-2015-0014/html
Scroll to top button