Skip to content
Licensed Unlicensed Requires Authentication Published by De Gruyter August 3, 2016

On the Validity of Edgeworth Expansions and Moment Approximations for Three Indirect Inference Estimators

Stelios Arvanitis and Antonis Demos

Abstract

This paper deals with higher order asymptotic properties for three indirect inference estimators. We provide conditions that ensure the validity of locally uniform, with respect to the parameter, Edgeworth approximations. When these are of sufficiently high order they also form integrability conditions that validate locally uniform moment approximations. We derive the relevant second order bias and MSE approximations for the three estimators as functions of the respective approximations for the auxiliary estimator. We confirm that in the special case of deterministic weighting and affinity of the binding function, one of them is second order unbiased. The other two estimators do not have this property under the same conditions. Moreover, in this case, the second order approximate MSEs imply the superiority of the first estimator. We generalize to multistep procedures that provide recursive indirect inference estimators which are locally uniformly unbiased at any given order. Furthermore, in a particular case, we manage to validate locally uniform Edgeworth expansions for one of the estimators without any differentiability requirements for the estimating equations. We examine the bias-MSE results in a small Monte Carlo exercise.

JEL: C10; C13

Acknowledgments

We would like to thank, without implicating them, Gabriele Fiorentini, Enrique Sentana, Eric Renault, Peter Phillips, the seminar participants at the University of Piraeus, University of Cyprus, and AUEB, the conference participants at SSEE and at EEA-ESEM 2012. The comments of two anonymous referees and the associate editor have greatly improve the exposition of the paper. Financial support of BRFP II AUEB is gratefully acknowledged.

Appendix

Proofs

Proof of Lemma 2:

  1. Due to the triangle inequality and Assumption 3 we have that for ε>0

    supθΘPθ(supθΘ|||βnb(θ)||||b(θ)b(θ)|||>ε)supθΘPθ(||βnb(θ)||>ε)=o(na)

    Hence for qn(θ)=βnb(θ), q(θ*, θ)=b(θ*)–b(θ) and by Assumption 4.a) Lemma AL.3 applies. Hence for γ(θ)=θ due to Assumption 1.3 Lemma AL.1 also applies implying the result.

  2. Given i), we have that θn∈𝒪ϵ(θ) with Pθ-probability 1o(na) that is locally independent of θ for any ϵ>0. For some ϵ small enough, such that Oϵ(θ)Oε0(θ0) (which exists due to the fact that ε0>ε) due to Assumption 6, we have that condition FOC of the appendix Lemmas AL.4 and AL.5 is satisfied by the GMR1 estimator with Qnbθ. Furthermore Assumption 6 implies conditions HUB (γ (θ)=θ hence set δ=ε0) and RANK of the same lemma. Condition TIGHT follows from 5, as under this assumption there is C*>0 locally independent of θ such that

    (5)supθO¯ε(θ0)Pθ(n||βnb(θ)||>Cln1/2n)=o(na) (5)

    (see Lemma AL.2 of Arvanitis and Demos 2014). Hence Lemma AL.4 applies ensuring that

    supθO¯ε(θ0)Pθ(n||GMR1θ||>Cln1/2n)

    for some C>0 locally independent of θ. Hence condition UTIGHT of Lemma AL.5 holds. Moreover Assumption 5 implies condition UEDGE of the same lemma for Mn(θ)=nmn(θ). Due to Assumption 6 for any θ∈𝒪̅ε (θ0) and any θ* sufficiently close to θ, bθ(θ) admits a Taylor expansion of order s*–1 around θ of the form

    bθ(θ)=i=0s11i!Dibθ(θ)((θθ)i)+1(s1)!(Ds1bθ(θ+)Ds1bθ(θ))((θθ)s1)

    where θ+ lies between θ* and θ. This implies that for any θn=GMR1 due to condition UTIGHT we have that with Pθ-probability 1o(na) locally independent of θ

    bθ(θn)=i=0s11i!1ni/2Dibθ(θ)((n(θnθ))i)+Rn(θn,θ)

    where Rn(θn,θ)=1(s1)!1n(s1)/2(Ds1bθ(θn+)Ds1bθ(θ))((n(θnθ))s1), and θn+ lies between θn and θ. Now by Assumption 6 bθ(θ) has full rank for any θ∈𝒪̅ε (θ0) and by submultiplicativity, the relation of θn+ to θn and condition UTIGHT

    supθO¯ε(θ0)Pθ(1(s1)!1n(s1)/2(Ds1bθ(θn+)Ds1bθ(θ))×((n(θnθ))s1)>γn)supθO¯ε(θ0)Pθ(1(s1)!1n(s1)/2supθO¯ε0(θ0)Dsbθ(θ)||θn+θ||||n(θnθ)||s1>γn)supθO¯ε(θ0)Pθ(M(s1)!Csns/2lns/2n>γn)+o(na)

    which is of order o(na) for γn=M(s1)!Csns/2lns/2n=o(na) and locally independent of θ. Analogously, due to Assumption 6 for any θ∈𝒪̅ε (θ0) and any θ* sufficiently close to θ, b(θ*) admits a Taylor expansion of order s*–1 around θ of the form

    qn=βnb(θ)=βnb(θ)i=1s1i!Dib(θ)((θθ)i)1s!(Dsb(θ+)Dsb(θ))((θθ)s)

    where θ+ lies between θ* and θ. This implies that for θn we have that with Pθ-probability 1o(na)

    n(βnb(θn))=n(βnb(θ))+i=0s11(i+1)!1ni/2Di+1b(θ)((n(θnθ))i+1)+Rn#(θn,θ)

    where Rn#(θn,θ)=1s!1n(s1)/2(Dsb(θn+)Dsb(θ))((n(θnθ))s), and θn+ lies between θn and θ. Now by Assumption 6 bθ(θ) has full rank for any θ∈𝒪̅ε (θ0) and so does the identity matrix in front of n(βnb(θ)), and thereby due to submultiplicativity, the relation of θn+ to θn and condition UTIGHT

    supθO¯ε(θ0)Pθ(1s!1n(s1)/2(Dsb(θn+)Dsb(θ))×((n(θnθ))s)>γn#)supθO¯ε(θ0)Pθ(1s!1n(s1)/2supθO¯ε0(θ0)||Ds+1b(θ)||||θn+θ||×||n(θnθ)||s>γn#)supθO¯ε(θ0)Pθ(Ms!Cs+1ns/2ln(s+1)/2n>γn#)+o(na)

    which is of order o(na) for γn#=Ms!Cs+1ns/2ln(s+1)/2n=o(na) and locally independent of θ. Finally due to Lemma AL.6 which applies by Assumptions 5 and 4 condition EXPAND holds and the result follows by the same lemma.   ■

Proof of Lemma 3: The proof is in the same spirit as the proof of Lemma 2. For ε>0, let E(ε,θ)={ωΩ:||βnb(θ)||>ε2}, then

supθΘ||Eθβnb(θ)||supθΘEθ||βnb(θ)||1E(ε,θ)+ε2.

As B is bounded, due to Assumption 1.2 and by Assumption 3 there exists an n* such that supθΘPθ(||βnb(θ)||>ε3)ε2M where M denotes the diameter of B. Hence

supθΘ||Eθβnb(θ)||ε for all nn

and since ε is arbitrary

(6)supθΘ||Eθβnb(θ)||=o(1) (6)

Due to the triangle inequality and Assumption 3 we have that for ε>0

supθΘPθ(supθΘ|||βnEθβn||||b(θ)b(θ)|||>ε)supθΘPθ(||βnb(θ)||+supθΘ||Eθβnb(θ)||>ε)=o(na)

For qn(θ)=Eθβn, q(θ*, θ)=b(θ*)–b(θ) and by Assumption 4.a) Lemma AL.3 applies. Hence for γ(θ)=θ due to Assumption 1.3 Lemma AL.1 also applies implying the result.

ii) Given i), we have that θn∈𝒪ϵ(θ) with Pθ-probability 1o(na) that is locally independent of θ for any ϵ>0. For some ϵ small enough, such that Oϵ(θ)Oϵ0(θ0) (which exists due to the fact that ε0>ε) due to Assumption 7, we have that condition FOC of the appendix Lemmas AL.4 and AL.5 is satisfied by the GMR2 estimator with QnEθ(βn)θ. Furthermore Assumption 7 and 6 imply conditions HUB (γ(θ)=θ hence set δ=ε0) and RANK of the same lemma due to the fact that since D2Eθβn is uniformly bounded on Oε0(θ0),DEθβn converges uniformly to Db(θ) due to Lemma AL.7 and therefore the rank condition is implied by 7 for large enough n. Condition TIGHT follows from 5 and Lemma AL.2 of Arvanitis and Demos (2014) and the fact that as a*>a≥0 we have that a*>0 and there exists a C2>0 locally independent of θ and for E={ωΩ:||βnb(θ)||>C2ln1/2nn1/2}

supθO¯ε(θ0)||Eθβnb(θ)||supθO¯ε(θ0)Eθ[||βnb(θ)||1E]+supθO¯ε(θ0)Eθ[||βnb(θ)||1E¯]MsupθO¯ε(θ0)Pθ(||βnb(θ)||>C2ln1/2nn1/2)+C2ln1/2nn1/2supθO¯ε(θ0)Pθ(||βnb(θ)||C2ln1/2nn1/2)=o(na)+C2ln1/2nn1/2(1o(na))=O(ln1/2nn1/2),

where the last line comes from equation 5, above, and therefore

supθO¯ε(θ0)Pθ(||βnEθβn||>C1ln1/2nn1/2)supθO¯ε(θ0)Pθ(||βnb(θ)||+||Eθβnb(θ)||>C1ln1/2nn1/2)supθO¯ε(θ0)Pθ(||βnb(θ)||+O(ln1/2nn1/2)>C1ln1/2nn1/2)=o(na)

Hence Lemma AL.4 applies ensuring that

supθO¯ε(θ0)Pθ(n||GMR2θ||>Cln1/2n)=o(na)

for some C>0 locally independent of θ, hence condition UTIGHT of Lemma AL.5 holds. Moreover Assumption 5 along with the fact that the support of βnb(θ) is uniformly bounded by 𝒪̅3η (0) for any η greater or equal than the diameter of B, and the fact that n(βnEθβn) admits a locally uniform Edgeworth expansion of order s*–1 (see Lemma 4.1 of Arvanitis and Demos 2014) imply condition UEDGE of Lemma 5 for Mn(θ)=n(mn(θ)βnEθβn) with order s*–1. Due to Assumption 7 for any θ∈𝒪̅ε (θ0) and any θ* sufficiently close to θ, Eθ(βn)θ(θ) admits a Taylor expansion of order s*–1 around θ and the rest follows as in the proof of Lemma 2 above with the difference that here Eθ(βn)θ is considered instead of bθ(θ) and Assumption 7 is employed instead of 6 (see Arvanitis and Demos 2013b, for details).■

Proof of Lemma 4: By Assumption 5, Lemma 8, below, adding subtracting n(b(θ)+i=1sV(kiβ(z,θ))ni2) and n(b(θn+)+i=1sV(kiβ(z,θn+))ni2), we get n(EθnβnEθβn)An(θ)=n(Eθnβnb(θn+)i=1sV(kjβ(z,θn+))ni2)n(Eθβnb(θ)i=1sV(kiβ(z,θ))ni2)+n(b(θn+)b(θ)i=1s1i!Dib(θ)((θn+θ)i))+Bn where Bn=i=1sV(kiβ(z,θn+))ni12i=1sV(kiβ(z,θ))ni12i=1s1i!j=1siDiV(kjβ(z,θ))nj12((θn+θ)i). Employing the mean value Theorem for V(kiβ(z,θn+)), and for θn++ such that ||θn++θ||<||θn+θ||, and collecting terms we get:

Bn=i=1s1ni121(si+1)!Dsi+1V(kiβ(z,θ))((θn++θ)si+1).

Taking into account that θn+O¯ε(θ0) with probability 1o(na) and employing the triangular inequality we have that, for s<s*,

supθO¯ε(θ0)Pθ(||n(Eθn+βnEθβn)An(θ)||>γn)supθO¯ε(θ0)Pθ(supθO¯ε(θ0)nEθβnb(θ)i=1sV(kiβ(z,θ))ni2>γn6)+i=1ssupθO¯ε(θ0)Pθ(1ni12||Bn||>γn3s)+supθO¯ε(θ0)Pθ(nb(θn+)b(θ)i=1s1i!Dib(θ)((θn+θ)i)>γn3)+o(na).

Now we have that

an=nEθβnb(θ)i=1sV(kiβ(z,θ))ni2=o(na)

independent of θ, due to Lemma 8.

Further, due to the continuity of Dsi+1V(kiβ(z,θ)), Assumption 8, and the assumption of the asymptotic behavior of θn+ we get

supθO¯ε(θ0)Pθ(1ni12||Bn||>γn3s)o(na)

provided that γnγn(1)=lnsi+12nns23ssupθO¯ε(θ0)sup||Dsi+1V(kiβ(z,θ))||(si+1)!.

Furthermore using the same reasoning as above

supθO¯ε(θ0)Pθ(n||b(θn+)b(θ)i=1s1i!Dib(θ)((θnθ)i)||>γn3)o(na)

when γnγn(2)=3supθO¯ε(θ0)||Ds+1b(θ)||(s+1)!lns+12nns2. Hence for γn=max(γn(2),6an,γn(1), i=1,,s) the result follows for large enough n.■

Proof of Lemma 5: Consider the random element θn defined by

θn=argminθO¯ε0(θ0)βnb(θ)i=1s11ni2V(kiβ(z,θ)),

where ε0<ε (defined in Assumption 5) Compactness and continuity imply the existence of θn. Compactness along with equicontinuity of V(kiβ(z,θ)) for all i along with Assumptions 1, 3 and Lemma AL.1 imply that

supθO¯ε0(θ0)Pθ(||θnθ||>δ)=o(na),δ>0.

Again the existence of equicontinuous derivatives implies that D(b(θ)+i=1s11ni2V(kiβ(z,θ))) converges uniformly over O¯ε0(θ0) which implies that rankD(b(θ)+i=1s11ni2V(kiβ(z,θ)))=p for all θ and large enough n. The last pair of assertions along with the equality of dimensions imply that βnb(θn)i=1s11ni2V(kiβ(z,θn))=0 with Pθ probability 1o(na) that is independent of θ∈𝒪̅ε (θ0). But then for any ε*<ε0 Lemma 8 also implies that ||βnEθnβn||ρn with supθO¯ε(θ0)Pθ(ρn>o(na))=o(na). But by definition

||βnEGMR2βn||||βnEθnβn||ρn

with Pθ probability 1o(na) that is independent of θO¯ε(θ0) and thereby there exists a random element vn such that supθO¯ε(θ0)Pθ(||vn||>o(na))=o(na) and βnEGMR2βn=vn with Pθ probability 1o(na) that is independent of θO¯ε(θ0). Notice that this necessarily holds even if the Euclidean norm is replaced by any other (possibly stochastic) norm. Consequently, this implies condition FOC of Lemma AL.5. Notice that the uniform consistency follow for the GMR1 and GMR2 as in the first parts of Lemmas 2, 3. Assumption 6 along with uniform consistency and as p=q we have that b(GMR1)=βn with probability 1o(na) uniformly on 𝒪̅ε (θ0). Furthermore by the second part of Lemma 2 we have that

(7)supθO¯ε(θ0)Pθ(n||GMR1θ||>Mln1/2n)=o(na) (7)

Hence with probability 1o(na) locally independent of θ, applying the mean value theorem we have that

b(GMR1)=b(GMR2)+b/(θn+)θ(GMR1GMR2),

where θn+ is such that ||θn+GMR2||<||GMR1GMR2||. It follows that with Pθ-probability 1o(na) locally independent of θ

GMR1GMR2=(b/(θn+)θ)1(b(GMR1)b(GMR2)).

Hence with probability 1o(na) uniformly on 𝒪̅ε (θ0)

||GMR1GMR2||M||βnb(GMR2)||M(||βnEGMR2βn||+||EGMR2βnb(GMR2)||)M||EGMR2βnb(GMR2)||=O(1n)

and the last equality is true (as βn has a uniform Edgeworth expansion on 𝒪̅ε (θ0), Assumption 5, and apply Lemma 8). Taking into account equation 7 we get that, for some C>0, locally independent of θ

supθO¯ε(θ0)Pθ(n||GMR2θ||>Cln1/2n)=o(na).

This implies condition UTIGHT of Lemma AL.5. It also, along with Lemmas 4 and 8, implies that for any ε*<ε

supθO¯ε(θ0)Pθ(||n(βnEGMR2βn)Γn(θ)||>γn)=o(na)

where γn=o(na) independent of θ and

Γn(θ)=n(βnEθβn)i=1s11ni2V(kiβ(z,θ))i=1s1ni12i!Di(b(θ)+j=1siV(kjβ(z,θ))nj2)(n(GMR2θ)i)

which validates condition EXPAND of lemma AL.5 for Gn=Wnj=Idp×p. Moreover Assumption 5 along with the fact that the support of βnb(θ) is uniformly bounded by 𝒪̅3η (0) for any η greater or equal than the diameter of B, and Lemma 4.1 of Arvanitis and Demos (2014) imply condition UEDGE of Lemma AL.5 for Mn(θ)=n(mn(θ)βnEθβn) with order s*–1. Hence the conditions of Lemma AL.5 are satisfied and the result follows.■

Proof of Lemma 6:A. It is easy to see that this special assumption implies that GMR1=GT with probability 1o(na) independent of θ. The rest are trivial consequences of Lemma 2. B. Similarly this special assumption implies that GMR2=GT (Pθ almost surely for all θ). The rest are trivial consequences of Lemmas 3 or 5 i) or ii).■

Proof of Lemma 7:

  1. By Assumption 9.2–4., uniform integrability and via the triangle inequality we get that

    supθΘPθ(supθΘ||Eθcn(βn)c(θ,b(θ))||>ε)=o(na)

    Hence for qn(θ)=Eθcn(βn), q(θ*, θ)=c(θ, b(θ*)) and by Assumptions 4.a) Lemma AL.3 applies. Hence for γ(θ)=θ due to Assumptions 1.3, 9 Lemma AL.1 also applies proving the result.

  2. Given i), we have that θn∈𝒪ϵ(θ) with Pθ-probability 1o(na) that is locally independent of θ for any ϵ>0. For some ϵ small enough, such that Oϵ(θ)Oε0(θ0) (which exists due to the fact that ε0>ε) due to Assumption 10, we have that condition FOC of Lemma AL.4 (in the Appendix) is satisfied by the GT estimator for Qn=Eθ(cn(βn))θ. Furthermore Assumption 10 implies conditions HUB (γ(θ)=θ hence set δ=ε0) and RANK of the same lemma. Condition TIGHT follows from 5 Lemma AL.2 of Arvanitis and Demos (2014) and as Eθcn(b(θ))=0 the fact that supθO¯ε(θ0)Pθ(||Eθcn(βn)||>C1ln1/2nn1/2)supθO¯ε(θ0)Pθ(||βnb(θ)||>C1supθO¯ε(θ0)Eθ(κn)ln1/2nn1/2) we get that there exist C1>0 large enough locally independent of θ for which the last term in the display is of order o(na). Hence Lemma AL.4 applies ensuring that

    supθO¯ε(θ0)Pθ(n||GTθ||>Cln1/2n)=o(na)

    for some C>0 independent of θ. Hence condition UTIGHT of Lemma AL.5 holds. Moreover Assumption 5 implies condition UEDGE of the same lemma for Mn(θ)=nmn(θ). Due to Assumption 10 for any φ=(θb(θ)) for any θO¯ε0(θ0) and as GT=θn is sufficiently close to φ, Eθncn(βn)θ admits with Pθ-probability 1o(na) that is independent of θ a Taylor expansion of order s*–1 around φ, with remainder

    Rn(θn,θ)=1(s1)!(Ds1(Eθn+cn(βn+)θ)Ds1(Eθcn(b(θ))θ))((φnφ)s1),

    and θn+,βn+ lie between θn and θ and βn and b(θ), respectively (see Arvanitis and Demos 2014, for details). Due to Assumptions 10, 5, Lemma AL.2 of Arvanitis and Demos (2014) and by submultiplicativity, the relation of θn+ to θn and condition UTIGHT

    supθO¯ε(θ0)Pθ(1n(s1)/2Rn(θn,θ)>γn)supθO¯ε(θ0)Pθ(2s1M(s1)!maxs(C,C+)ns/2lns/2n>γn)+o(na)

    which is of order o(na) for γn=2s1M(s1)!maxs(C,C+)ns/2lns/2n=o(na) and independent of θ. Furthermore, due to the same assumption and the fact that c(θ, b(θ))=0 we have that with Pθ-probability 1o(na) locally independent of θ the s*–1 Taylor expansion of nEθncn(βn) around φ has remainder

    Rn#(θn,θ)=1s!1n(s1)/2(DsEθn+cn(βn+)DsEθcn(b(θ)))((n(φnφ))s),

    and θn+ lies between θn and θ. Hence analogously to the previous

    supθO¯ε(θ0)Pθ(||Rn#(θn,θ)||>γn#)supθO¯ε(θ0)Pθ(2sMs!maxs+1(C,C+)ns/2ln(s+1)/2n>γn#)+o(na)

    which is of order o(na) for γn#=2sMs!maxs+1(C,C+)ns/2ln(s+1)/2n=o(na) and independent of θ (see Arvanitis and Demos 2013b, for details). Then due to Assumption 10 and the fact that Eθcn(β)=c(θ, β), Eθcn(b(θ))θ,Eθcn(b(θ))β are of full rank for any θ∈𝒪̅ε (θ0). Finally due to Lemma AL.6 which applies by Assumptions 5 and 4 condition EXPAND holds and the result follows by the same lemma.■

Proof of Lemma 8: Let Qn denote the measure with density (1+i=1sπi(z)ni2)φV(z). Since 2a+m+1>2a+1, we have that supAC|Pn(A)Qn(A)|=O(naη), where η>0. Hence

na|qK(xm)(dPndQn)|na|Oc(lnn)ϵ(0)K(xm)(dPndQn)|+na|q\Oc(lnn)ϵ(0)K(xm)dPn|+na|q\Oc(lnn)ϵ(0)K(xm)dQn|naM(lnn)mϵOc(lnn)ϵ(0)|dPndQn|+naq\Oc(lnn)ϵ(0)|K(xm)|(dPn+|dQn|)M(lnn)mϵsupACna|Pn(A)Qn(A)|+naq\Oc(lnn)ϵ(0)|K(xm)|(dPn+|dQn|)

Due to the hypothesis for the support of Pn

naq\Oc(lnn)ϵ(0)|K(xm)|dPn=na[q\Oc(lnn)ϵ(0)]Onρ(0)|K(xm)|dPnna+mβρmqmq1||x||>c(lnn)ϵdPn

Hence

na|qxm(dPndQn)|M(lnn)mϵsupACna|Pn(A)Qn(A)|+na+mβρmqmP(||ζn||>c(lnn)ϵ)+naq\Oc(lnn)ϵ(0)|K(xm)||dQn|.

As supACna|Pn(A)Qn(A)|=O(nη) for η>0, we have that

(lnn)2ϵsupACna|Pn(A)Qn(A)|=o(1)

and na+m2ρmqmP(||ζn||>c(lnn)ϵ)=o(1) if ϵ12 and c2a+m+1 by Lemma 2 of Magdalinos (1992). Finally naq\Oc(lnn)ϵ(0)|K(xm)||dQn|=o(1) due to Gradshteyn and Ryzhik (1994) formula 8.357. For the uniform case first notice that

supθO¯ε(θ0)Pθ(||ζn||>Mln1/2n)=o(na)

This is due to the fact that the set {x∈ℝq:‖x‖≤M ln1/2n} has boundary of Lebesgue measure zero and

supθO¯ϵ(θ0)||x||>Mln1/2n(1+i=1s1ni2|πi(x,θ)|)φV(θ)(x)dx||z||>Mλmax(θ)ln1/2n(1+i=1s1ni2|πi(V1/2(θi)z,θi)|)φ(x)dx

where λmax(θ) denotes the maximum absolute eigenvalue of V1/2(θ) and θiO¯ε(θ0) exist for all i=1, …, s* due to the continuity and are independent of z due to the positivity and the fact that πi are polynomials in x, and θ* exists due to continuity of V and the compactness of 𝒪̅ε (θ0). For Ms*λmax(θ*) the result follows from Lemma 2 of Magdalinos (1992). The rest follows in the same spirit of the first part.■

Proof of Lemma 9: Our Assumptions and Lemmas 2, 8 ensure the validity of the mean approximation. Then from Theorem 3.1 of Arvanitis and Demos (2014) we have that the relevant moment approximation can be obtained if we employ the appropriate approximations of n(βnb(θn)),Wn(θn), and b(θn)θ in terms of n(θnθ) and n(θnθ). Therefore an appropriate approximation for n(θnθ) is obtained by inverting the product of these terms. Finally, integrating with respect to (1+π1(z,θ)n)φV(θ)(z), noting that k1β(z,θ)=z,k2β(z,θ)=zπ1(z,θ) we obtain result (see Arvanitis and Demos 2013b, for details).■

Proof of Lemma 10: The Assumptions and Lemmas 2, 8 ensure the validity of the mean approximation uniformly over O¯ε(θ0). Furthermore from Lemma AL.7 we have that supθO¯ε(θ0)DEθβnDb(θ)1nDφV(k2β)(θ)=o(1) (recall that φV(k1β)=0). Then the proof proceeds as in the proof of Lemma 9 (see Arvanitis and Demos 2013b, for details).■

Proof of Lemma 11: First, notice that our Assumptions and Lemmas 2, 8 ensure the validity of the mean approximation. Then employing the implicit function Theorem, as c(θ, b(θ))=0l and c(θ,b(θ))θ=0lxp, appropriate order Taylor expansions of nc(θn,βn),Wn(θn) and c(θn,βn)θ, and inverting the product of these terms we get an appropriate approximation of n(θnθ). Integrating this approximation w.r.t. (1+π1(z,θ)n)φV(θ)(z) and employing Theorem 3.1 of Arvanitis and Demos (2014) we get the result (see Arvanitis and Demos 2013b, for details).■

Proof of Corollary 3: When W** is independent of x and θ and b(θ) is affine then 𝒥=0 and j=bθ2cj(θ,b)ββbθ. Substituting these expressions in the expansion of n(θnθ) and integrating w.r.t. (1+π1(z,θ)n)φV(θ)(z) we get the result. On the other hand, when l p=q=l and b(θ) is affine then D1=(c(θ,b)βbθ)1,J=0,q1β=(bθ)1k1β,q1β=(bθ)1k1β, and j=bθ2cj(θ,b)ββbθ. The result follows by integrating, again, w.r.t. (1+π1(z,θ)n)φV(θ)(z) (see Arvanitis and Demos 2013b, for details).■

Proof of Lemma 12:

i) The result follows from Lemmas 6.A and 9. Notice that as p=q=l, we have that C1bθW=(bθ)1.

ii) The result follows from Lemmas 6.B and 10.■

Proof of Lemma 13: For both estimators we have that due to Lemma 8, Theorem 3.1 of Arvanitis and Demos (2014) along with the approximations employed in Lemmas 9, 10 we have Eθ(n(θnθ)(θnθ))=qC1bθW(k1β(z,θ)+k2β(z,θ)n)(k1β(z,θ)+k2β(z,θ)n)WbθC1φV(θ)(z)dz, where k1β(z,θ)=z,k2β(z,θ)=zπ1(z,θ). Keeping the relevant order terms, the result follows.■

Proof of Lemma 14: Again we have that due to Lemma 8, Theorem 3.1 of Arvanitis and Demos (2014) along with the approximations used in Lemma 3 when W** is independent of x and θ and b(θ) is affine, we get the appropriate expression from the proof of Lemma 3. Integrating w.r.t. (1+π1(z,θ)n)φV(θ)(z) we get the result (see Arvanitis and Demos 2013b, for details).■

Proof of Lemma 15: First notice that in any step of the procedure the binding function is the identity. Next the o(na) uniform consistency of θn ensures the analogous for any step of the recursion. Then Lemma 5 applies and accordingly θn admits a locally uniform Edgeworth expansion of order s*. Given this the exact same reasoning, implies the same result for θn for any k. Moreover Assumption 8 follows for the expansions in every step of the procedure due to the previous. The proof for the moment approximations for the case k=1 follows easily. Using induction, let us assume that the result holds for some k=h, i.e. assume that the appropriate expression for n(θn(hθ) is given by:

Eθn(θn(hθ)=1n2h+12V(k2h+2)+1n2h+22V(k2h+3)+o(n2h+22).

uniformly over 𝒪̅ε (θ0). Hence for θ∈𝒪̅ε (θ0), by Lemma 4 it follows that n(Eθn(h+1θn(hEθθn(h)(Idp+1n2h+22V(k2h+2)θ)n(θn(h+1θ) is bounded by a real sequence of order o(n2h+32) that is independent of θ, with Pθ-probability 1o(n2h+32) independent of θ. The h+1st-step GMR2 estimator satisfies with Pθ-probability 1o(n2h+32) independent of θ, θn(h=Eθn(h+1θn(h. Hence Lemma 8 and Theorem 3.1 of Arvanitis and Demos (2014) imply that the required approximation would be given by the integration of the Edgeworth density in the hth step of the following approximation

n(θn(hθ)(1n2h+12V(k2h+2)+1n2h+22V(k2h+3)).

This integration gives an error of o(n2h+22) as p(1+i=12h+2πi(z,θ)ni/2)φV(θ)(z)dz=1+o(n2h+22) due to the validity of the Edgeworth approximation of the distribution of n(θnθ) and the result follows. For the MSE approximation the result follows analogously, by simply noticing that (Idp+1n2h+22V(k2h+2/)θ)1=Idp+o(1).

General Results

The following results essentially justify the methodologies used for the establishment of the validity of the locally uniform Edgeworth expansions in all the previous cases. They are cited without proofs. Their analytical proofs can be found in Arvanitis and Demos (2013b).

Lemma AL.1:Suppose that:

  • UUC

    supθΘPθ(supβB|cn(β)c(θ,β)|>ε)=o(na),ε>0

  • AB c(θ, β) is jointly continuous and γ(θ)=arg minβBc(θ, β), then

supθΘPθ(||βnγ(θ)||>ε)=o(na),ε>0

where βn∈arg minβBcn(β).

Lemma AL.2:Let Assumptions 4.a) and 3 hold. Then for j=*,**

supθΘPθ(||Wnj(θn)EθWj(θ)||>ε)=o(na),ε>0

Furthermore, there exists K>0 for which

supθΘPθ(||Wnj(θn)||>K)=o(na)

Lemma AL.2:Suppose that

cn(β)=qn(β)Wnj(θn)qn(β)

for some appropriate random element qnwhere Wnj,θn satisfy Assumptions 4.a), 3 and for q an appropriate jointly continuous function on Θ×B

supθΘPθ(supβB||qn(β)q(θ,β)||>ε)=o(na),ε>0

Then AL.1.UUC holds for c(θ,β)=q(θ,β)EθWj(θ)q(θ,β) which is jointly continuous.

Lemma AL.4:Suppose Wnj,θn˜ satisfy Assumptions 4, 3, βnand γ(θ) are as in lemma AL.1, γ is continuous on 𝒪̅ε (θ0) and that:

  • FOC βnsatisfies

    qn(βn)βWnj(θn)qn(βn)=0

    with Pθ-probability1o(na)that is independent of θ,

  • HUB for some δ, M>0 independent of θ such thatγ(O¯ε(θ0))O¯δ(γ(θ0))and for all i, supθO¯ε(θ0)Pθ(supβO¯δ(γ(θ0))2qn(βn)ββi>M)=o(na),

  • RANK for anyβO¯δ(γ(θ0)),qn(β)βis of full rank with Pθ-probability1o(na)that is independent of θ and,

  • TIGHT for some C>0 independent of θ, supθO¯ε(θ0)Pθ(n||qn(γ(θ))||>Cln1/2n)=o(na), then

supθO¯ε(θ0)Pθ(n||βnγ(θ)||>C+ln1/2n)=o(na)

for some C+>0 independent of θ.

Lemma AL.5:Suppose that:

  • FOC βnsatisfies

    Gn(βn)Wnj(θn)qn(βn)=vn

    with Pθ-probability1o(na)andPθ(||vn||>o(na))=o(na)that are independent of θ and,

  • UTIGHT There exists a C+>0 independent of θ for which

    supθO¯ε(θ0)Pθ(n||βnγ(θ)||>C+ln1/2n)=o(na)

  • UEDGE There exists a random element Mn(θ) with values in an Euclidean space, containing the elements of n(θnθ), the distribution of which admits a uniform over 𝒪̅ε (θ0) Edgeworth expansion Ψn,s(θ). The ith polynomial, say, πi(z, θ) of Ψn,s(θ) is equicontinuous on 𝒪̅ε (θ0) ∀z∈ℝq, for i=1, …, s–2, and if Σ(θ) denotes the variance matrix in the density of Ψn,s(θ) then it is continuous on 𝒪̅ε (θ0) and positive definite.

  • EXPAND The following hold with Pθ-probability 1o(na) that is independent of θ

Gn(βn)=i=0s11ni/2j=0iCi,j,n(θ)(Mn(θ)j,Sn(θ)ij)+Rn(βn,θ)Wnj(θn)=i=0s11ni/2Ci,n(θ)(Mn(θ)i)+Rn(θn,θ)nqn(βn)=i=0s11ni/2j=0i+1Ci,j,n#(θ)(Mn(θ)j,Sn(θ)i+1j)+Rn#(βn,θ)

whereSn(θ)=n(βnγ(θ)),Ci,j,n:O¯ε(θ0)×qip,Ci,n:O¯ε(θ0)×qipare i-linear, Ci,j,n:Θ×qi+1pis (i+1) -linearθO¯ε(θ0),C0,0,n,C0,n,C0,0,n#(θ),C0,1,n#(θ)are independent of n and have full rankθO¯ε(θ0),Ci,n,Ci,n,Ci,j,n#are equicontinuous on 𝒪̅ε (θ0), and

supθO¯ε(θ0)Pθ(||Rnl||>γnl)=o(na), l=,,#

for real sequence γnl=o(na) independent of θ, for, l=*, **, #.

Thenn(βnγ(θ))admits a locally uniform Edgeworth expansion,Ψn,s(θ),over 𝒪̅ε (θ0). The ith polynomial, say, πi(z,θ)of the density ofΨn,s(θ)is equicontinuous onO¯ε(θ0)zq,for i=1, …, s–2, and if Σ*(θ) denotes the variance matrix in the density ofΨn,s(θ)then it is continuous on 𝒪̅ε (θ0) and positive definite.

Lemma AL.6: Under Assumptions 4 and 5 condition EXPAND hold for Wnj(θn) where Mn(θ)=nmn(θ).

Lemma AL.7: For real valued functions fn, f defined on Θ′⊇Θ, suppose that: supθ∈Θ|fnf|=o(1), and supθ∈Θ D2fn‖, supθ∈ΘD2f‖<M. Then supθ∈ΘDfnDf‖=o(1).

References

Andersen, T. G., L. Benzoni, and J. Lund. 2002. “An Empirical Investigation of Continuous-Time Equity Return Models.” The Journal of Finance LVII: 1239–1284.10.3386/w8510Search in Google Scholar

Andrews, D. W. K. 2002. “Equivalence of the Higher-order Asymptotic Efficiency of k-step and Extremum Statistics.” Econometric Theory 18: 1040–1085.10.1017/S0266466602185021Search in Google Scholar

Andrews, D. W. K., and O. Lieberman. 2005. “Valid Edgeworth Expansions for the Whittle Maximum Likelihood Estimators for Stationary Long-Memory Gaussian Time Series.” Econometric Theory 21: 710–734.10.1017/S0266466605050383Search in Google Scholar

Andrews D. W. K., and P. Guggenberger. 2009. “Hybrid and Size-Corrected Subsampling Methods.” Econometrica 77: 721–762.10.3982/ECTA7015Search in Google Scholar

Arvanitis, S., and A. Demos. 2013a. “On the Validity of Edgeworth Expansions and Moment Approximations for Three Indirect Inference Estimators.” Working Paper No. 1406, Dept. of IEES, AUEB (http://wpa.deos.aueb.gr/docs/bias-ii-ee.pdf).Search in Google Scholar

Arvanitis, S., and A. Demos. 2013b. “On the Validity of Edgeworth Expansions and Moment Approximations for Three Indirect Estimators (Extended Revised Appendix).” Working Paper No. 1330, Dept. of IEES, AUEB (http://wpa.deos.aueb.gr/docs/bias-ii-revis-jspi-ta.pdf).Search in Google Scholar

Arvanitis, S., and A. Demos. 2014. “Valid Locally Uniform Edgeworth Expansions Under Weak Dependence and Sequences of Smooth Transformations.” Journal of Time Series Econometrics 6: 183–235.10.1515/jtse-2012-0003Search in Google Scholar

Arvanitis, S., and A. Demos. 2015. “A Class of Indirect Estimators: Higher-order Asymptotics and Approximate Bias Correction.” Econometrics Journal 18: 200–241.10.1111/ectj.12045Search in Google Scholar

Bansal, R. P., A. R. Gallant, R. Hussey, and G. Tauchen. 1995. “Nonparametric Estimation of Structural Models for High-Frequency Currency Market Data.” Journal of Econometrics 66: 251–287.10.1016/0304-4076(94)01618-ASearch in Google Scholar

Bao, Y. 2013. “Finite-Sample Bias of the QMLE in Spatial Autoregressive Model.” Econometric Theory 29: 68–88.10.1017/S0266466612000229Search in Google Scholar

Bao, Y., and A. Ullah. 2007. “The Second-Order Bias and Mean Squared Error of Estimators in Time-Series Models.” Journal of Econometrics 140: 650–669.10.1016/j.jeconom.2006.07.007Search in Google Scholar

Bhattacharya, R. N. 1987. “Some Aspects of Edgeworth Expansions in Statistics and Probability.” In New Perspectives in Theoretical and Applied Statistics, edited by M. L. Puri, J. P. Vilaplana, and W. Wertz, 157–170. N. York: J. Wiley and Sons.Search in Google Scholar

Bhattacharya, R. N., and J. K. Ghosh. 1978. “On the Validity of the Formal Edgeworth Expansion.” The Annals of Statistics 6.2: 434–451.10.1214/aos/1176344134Search in Google Scholar

Bollerslev, T. 1986. “Generalized Autoregressive Conditional Heteroskedasticity.” Journal of Econometrics 31: 307–328.10.1016/0304-4076(86)90063-1Search in Google Scholar

Calzolari, C., G. Fiorentini, and E. Sentana. 2004. “Constrained Indirect Inference Estimation.” Review of Economic Studies 71: 945–973.10.1111/0034-6527.00310Search in Google Scholar

Chernick, M. R. 2008. Bootstrap Methods:A Guide for Practitionersand Researchers. New Jersey: John Wiley & Sons.Search in Google Scholar

Chumacero, R. A. 2001. “Estimating ARMA Models Efficiently.” Studies in Nonlinear Dynamics and Econometrics 5: 103–114.10.1162/108118201317283626Search in Google Scholar

Chung, C., and G. Tauchen. 2001. “Testing Target-Zone Models using Efficient Method of Moments.” Journal of Business and Economic Statistics 19: 255–269.10.1198/073500101681019891Search in Google Scholar

Corradi, V., and E. M. Iglesias. 2008. “Bootstrap Refinements for QML Estimators of the GARCH(1, 1) Parameters.” Journal of Econometrics 144: 500–510.10.1016/j.jeconom.2008.03.003Search in Google Scholar

Demos, A., and D. Kyriakopoulou. 2013. “Edgeworth and Moment Approximations: The Case of MM and QML Estimators for the MA(1) Models.” Communications in Statistics-Theory and Methods 42: 1713–1747.10.1080/03610926.2011.597919Search in Google Scholar

Durbin, J. 1980. “Approximations for Densities of Sufficient Estimators.” Biometrika 67-2: 311–333.10.1093/biomet/67.2.311Search in Google Scholar

Gallant, R. A., and G. E. Tauchen. 1996. “Which Moments to Match.” Econometric Theory 12, 657–681.10.1017/S0266466600006976Search in Google Scholar

Gallant, R. A., and J. R. Long. 1997. “Estimating Stochastic Differential Equations Efficiently by Minimum Chi- Squared.” Biometrika 84: 125–141.10.1093/biomet/84.1.125Search in Google Scholar

Gallant, R. A., D. Hsieh, and G. Tauchen. 1997. “Estimation of Stochastic Volatility Models with Diagnostics.” Journal of Econometrics 81: 159–192.10.1016/S0304-4076(97)00039-0Search in Google Scholar

Garcia, R., E. Renault, and D. Veredas. 2011. “Estimation of Stable Distributions by Indirect Inference.” Journal of Econometrics 161: 325–337.10.1016/j.jeconom.2010.12.007Search in Google Scholar

Ghysels, I., L. Khalaf, and C. Vodounou. 2003. “Simulation Based Inference in Moving Average Models.” Annales d’ Econommie et de Statistique 69: 85–99.10.2307/20076364Search in Google Scholar

Gotze, F., and C. Hipp. 1983. “Asymptotic Expansions for Sums of Weakly Dependent Random Vectors.” Zeitschrift für Wahrscheinlichkeitstheorie und verwandie Gebiete 64: 211–239.10.1007/BF01844607Search in Google Scholar

Gotze, F., and C. Hipp. 1994. “Asymptotic Distribution of Statistics in Time Series.” The Annals of Statistics 22: 2062–2088.10.1214/aos/1176325772Search in Google Scholar

Gourieroux, C., and A. Monfort. 1996. “Simulation-Based Econometric Methods.” CORE Lectures, Ox. Un. Press.10.1093/0198774753.001.0001Search in Google Scholar

Gourieroux,C., A. Monfort, and E. Renault. 1993. “Indirect Inference.” Journal of Applied Econometrics 8 (1): 85–118.10.1002/jae.3950080507Search in Google Scholar

Gourieroux, C., E. Renault, and N. Touzi. 2000. “Calibration by Simulation for Small Sample Bias Correction.” In Simulation-Based Inference in Econometrics: Methods and Applications, edited by R. Mariano, T. Schuermann, and M. J. Weeks. Cambridge: Cambridge University Press,10.1017/CBO9780511751981.018Search in Google Scholar

Gourieroux, C., P. C. B. Phillips, and J. Yu. 2010. “Indirect Inference for Dynamic Panel Models.” Journal of Econometrics 157: 68–77.10.1016/j.jeconom.2009.10.024Search in Google Scholar

Gradshteyn, I. S., and I. M. Ryzhik. 1994. “Table of Integrals, Series, and Products.” In edited by A. Jeffrey. San Diego: Academic Press.Search in Google Scholar

Hall, P. 1992. The Bootstrap and Edgeworth Expansion. New York: Springer-Verlag,.10.1007/978-1-4612-4384-7Search in Google Scholar

Hall, P., and J. L. Horowitz. 1996. “Bootstrap Critical Values for Tests based on Generalized-Method-of-Moment Estimators.” Econometrica 64: 891–916.10.2307/2171849Search in Google Scholar

Hall, P., and Q. Yao. 2003. “Inference in ARCH and GARCH Models with Heavy-Tailed Errors.” Econometrica 71: 285–317.10.1111/1468-0262.00396Search in Google Scholar

Koenker, R., J. A. F. Machado, C. L. Skeels, and A. H. Welsh. 1994. “Momentary Lapses: Moment Expansions and the Robustness of Minimum Distance Estimation.” Econometric Theory 10: 172–197.10.1017/S0266466600008288Search in Google Scholar

Kristensen, D., and O. Linton. 2006. “A Closed-Form Estimator for the GARCH(1,1) Model.” Econometric Theory 22: 323–337.10.1017/S0266466606060142Search in Google Scholar

Kyriacou, M., P. C. B. Phillips, and F. Rossi. 2014. “Indorect Inference in Spatial Autoregression.” Discussion Paper No. 1418, Economics Devision, Univesrity of Southampton (http://www.southampton.ac.uk/socsci/economics/research/papers).Search in Google Scholar

Lieberman, O., J. Rousseau, and D. M. Zucker. 2003. “Valid Asymptotic Expansions for the Maximum Likelihood Estimator of the Parameter of a Stationary, Gaussian, Strongly Dependent Process.” Annals of Statistics 31: 586–612.10.1214/aos/1051027882Search in Google Scholar

Linton, O. 1997. “An Asymptotic Expansion in the GARCH(1,1) Model.” Econometric Theory 13: 558–581.10.1017/S0266466600006009Search in Google Scholar

Magdalinos, M. A. 1992. “Stochastic Expansions and Asymptotic Approximations.” Econometric Theory 8 (3): 343–367.10.1017/S0266466600012974Search in Google Scholar

McCullagh, P. 1987. Tensor Methods in Statistics. London, New York: Chapman and Hall.Search in Google Scholar

Michaelides, A., and S. Ng. 2000. “Estimating the Rational Expectations Model of Speculative Storage: A Monte Carlo Comparison of Three Simulation Estimators.” Journal of Econometrics 96: 231–266.10.1016/S0304-4076(99)00058-5Search in Google Scholar

Mikusheva, A. 2007. “Uniform Inference in Autoregressive Models.” Econometrica 75: 1411–1452.10.1111/j.1468-0262.2007.00798.xSearch in Google Scholar

Newey, W. K. 1991. “Uniform Convergence in Probability and Stochastic Equicontinuity.” Econometrica 59: 1161–1167.10.2307/2938179Search in Google Scholar

Newey, W. K., and R. J. Smith. 2001. “Asymptotic Bias and Equivalence of GMM and GEL Estimators.” mimeo, Department of Economics, MIT.Search in Google Scholar

Phillips, P. C. B. 1980. “Finite Sample Theory and the Distribution of Alternative Estimators of the Marginal Propensity to Consume.” Review of Economic Studies 47: 183–224.10.2307/2297109Search in Google Scholar

Phillips, P. C. B. 2012. “Folklore Theorems, Implicit Maps and Indirect Inference.” Econometrica 80: 425–454.10.3982/ECTA9350Search in Google Scholar

Robinson, P. M. 1988. “The Stochastic Difference between Econometric Statistics.” Econometrica 56: 531–548.10.2307/1911699Search in Google Scholar

Sargan, J. D. 1976. “Econometric Estimators and the Edgeworth Approximation.” Econometrica 44: 421–448.10.2307/1913972Search in Google Scholar

Sentana, E., G. Calzolari, and G. Fiorentini. 2008. “Indirect Estimation of Large Conditionally Heteroskedastic Factor Models, with an Application to the Dow 30 Stocks.” Journal of Econometrics 146: 10–25.10.1016/j.jeconom.2008.06.001Search in Google Scholar

Skovgaard, I. M. 1981. “Transformation of an Edgeworth Expansion by a Sequence of Smooth Functions.” Scandinavian Journal of Statistics 8: 207–217.Search in Google Scholar

Smith, A. A. 1993. “Estimating Nonlinear Time-Series Models using Simulated Vector Autoregressions.” Journal of Applied Econometrics 8: 63–84.10.1002/jae.3950080506Search in Google Scholar

Spivak, M. 1999. A Comprehensive Introduction to Differential Geometry. Vol. 1, Texas: Publish or Perish, Inc.Search in Google Scholar

Taniguchi, M. 1987. “Validity of Edgeworth Expansions of Minimum Contrast Estimators for Gaussian ARMA Processes.” Journal of Multivariate Analysis 21: 1–28.10.1016/0047-259X(87)90096-0Search in Google Scholar

Taniguchi, M., and Y. Kakizawa. 2000. Asymptotic Theory of Statistical Inference for Time Series. New York: Springer-Verlag.10.1007/978-1-4612-1162-4Search in Google Scholar

van der Vaart, A. W., and J. A. Wellner. 2000. Weak Convergence and Empirical Processes. Springer.Search in Google Scholar

Published Online: 2016-8-3

©2018 Walter de Gruyter GmbH, Berlin/Boston

Scroll Up Arrow