GMM and 2SLS Estimation of Mixed Regressive, Spatial Autoregressive Models. Lung-fei Lee* Department of Economics. Ohio State University, Columbus, OH

Size: px
Start display at page:

Download "GMM and 2SLS Estimation of Mixed Regressive, Spatial Autoregressive Models. Lung-fei Lee* Department of Economics. Ohio State University, Columbus, OH"

Transcription

1 GMM ad 2SLS Estimatio of Mixed Regressive, Spatial Autoregressive Models by Lug-fei Lee* Departmet of Ecoomics Ohio State Uiversity, Columbus, OH October 200, October 2004; curret revisio July 2005 Abstract The GMM method ad the classical 2SLS method are cosidered for the estimatio of mixed regressive, spatial autoregressive models. These methods have computatioal advatage over the covetioal maximum likelihood method. The proposed GMM estimators are show to be cosistet ad asymptotically ormal. Withi certai classes of GMM estimators, best oes are derived. The proposed GMM estimators improve upo the 2SLS estimators ad are applicable eve if all regressors are irrelevat. A best GMM estimator may have the same limitig distributio as the ML estimator with ormal disturbaces). Key Words: Spatial ecoometrics, spatial autoregressive models, regressio, 2SLS, GMM, ML, efficiecy. Classificatio: C3, C2, R5 Correspodig address: Lug-fei Lee, 40 Arps Hall, Departmet of Ecoomics, Ohio State Uiversity, 945 N. High Street, Columbus, OH lflee@eco.ohio-state.edu * I am grateful to the Natioal Sciece Foudatio, Ecoomics Program for fiacial support for my research uder the grat umber SES-0380 ad SES I thak Mr. Ji Tao, who has worked as my RA, to coduct the Mote Carlo study i this paper. I appreciate havig valuable commets ad suggestios from Professor Christopher R. Bolliger, a aoymous referee ad a associate editor of this joural. i

2 . Itroductio I this paper, we propose a geeral GMM framework for the estimatio of mixed regressive, spatial autoregressive MRSAR) models. The GMM estimatio for those models ca be computatioally simpler tha the maximum likelihood ML) or quasi maximum likelihood QML) methods i a geeral settig. The GMM estimator GMME) may be asymptotically more efficiet tha the two-stage least squares 2SLS) estimator 2SLSE) ad may be asymptotically efficiet as the ML estimator MLE). The 2SLS method has bee oted for the estimatio of the MRSAR model i Aseli 988, 990), Lad ad Deae 992), Kelejia ad Robiso 993), Kelejia ad Prucha 997, 998), ad Lee 2003), amog others. The istrumetal variables IVs) are geerated from exogeous regressors ad the spatial weights matrix of the model. However, there are some issues o the 2SLS approach. The mai issue is that the proposed 2SLSE is iefficiet relative to the MLE. For the estimatio of a covetioal liear simultaeous equatio model, the 2SLSE is asymptotically efficiet as the limited iformatio MLE see, e.g., Amemiya 985). This is ot so for the estimatio of the MRSAR model. The 2SLSE has bee show to be cosistet ad asymptotically ormally distributed Kelejia ad Prucha 998). Lee 2003) discusses the best oe B2SLSE) withi the class of IV estimators. By comparig the limitig variace matrices, the 2SLSE ad B2SLSE are less efficiet relative to the MLE whe the disturbaces are ormally distributed). A subtle issue is that the 2SLS method would ot be cosistet whe all the exogeous regressors i the MRSAR model are really irrelevat. Furthermore, it is ot possible to test the joit sigificace of all the exogeous regressors based o those IV estimators Kelejia ad Prucha 998). O the cotrary, the MLE of a MRSAR is cosistet ad its limitig distributio ca be used for testig the joit sigificace of regressio coefficiets. These show that the 2SLS approach is less satisfactory tha the ML approach i some of its statistical properties. However, the 2SLS approach is computatioally simpler tha the ML approach ad is distributio free. For pure spatial autoregressive SAR) processes, a method of momets MOM) has bee itroduced i Kelejia ad Prucha 200). The MOM method is computatioally simpler tha the ML method. Their MOM estimator is cosistet but is ulikely to be efficiet relative to the MLE. Recetly, Lee 200) exteds the MOM estimatio ito a more geeral GMM estimatio framework. Withi that GMM framework, a The asymptotic distributio of their MOM estimator has ot bee established i their article. They have provided Mote Carlo results to demostrate that their MOM estimator is oly slightly iefficiet relative to the QML estimator uder various distributios.

3 best GMM estimator BGMME) ca have the same limitig distributio as the MLE or QML estimator. I this paper, we cosider the possible geeralizatio of the MOM method for the MRSAR model. We suggest a combiatio of the momets i the 2SLS framework with momet fuctios origiated for the estimatio of pure SAR processes. We show that the resultig GMME ca be asymptotically efficiet relative to the 2SLSE ad B2SLSE. The best GMME ca be made available ad it ca be efficiet as the MLE. With this GMM framework, oe may also test the joit sigificace of all the possible exogeous regressors. This paper is orgaized as follows. I Sectio 2, we cosider the estimatio of a MRSAR model. We discuss the momet fuctios that ca be used i additio to the momets based o the orthogoality of exogeous regressors with the model disturbace. Cosistecy ad asymptotic distributio of the GMME will be derived i Sectio 3. I Sectio 4, the GMME ad the 2SLSE are compared. The best selectio of momet fuctios ad IVs will be discussed ad its possible efficiecy property is derived. All the proofs of the results are collected i the appedices. Sectio 5 provides some Mote Carlo results for the compariso of fiite sample properties of estimators. Coclusios are draw i Sectio GMM Estimatio ad Idetificatio of the MRSAR Model The MRSAR model differs from a pure SAR process i the presece of exogeous regressors X as explaatory variables i the model: Y = λw Y + X β + ɛ, 2.) where X is a k dimesioal matrix of ostochastic exogeous variables, W is a spatial weights matrix of kow costats with a zero diagoal, ad the disturbaces ɛ i, i =,,, of the -dimesioal vector ɛ are i.i.d. 0,σ 2 ). Specifically, we assume that Assumptio. The ɛ i are i.i.d. with zero mea, variace σ 2 ad that a momet of order higher tha the fourth exists. Assumptio 2. The elemets of X are uiformly bouded costats, X has the full rak k, ad lim X X exists ad is osigular. Because statistics ivolvig quadratic forms of ɛ will be preset i the estimatio, the existece of the fourth order momet of ɛ i s will guaratee fiite variaces for the quadratic forms. The higher tha the fourth momet coditio i Assumptio is eeded i order to apply a cetral limit theorem due to Kelejia ad Prucha 200). The ostochastic X ad its uiform boudedess coditios i Assumptio 2 are for 2

4 coveiece. If the elemets of X are stochastic ad have ubouded rages, coditios i Assumptio 2 ca be replaced by some fiite momet coditios. The W Y i 2.) is called a spatial lag ad its coefficiet is supposed to represet the spatial effect due to the ifluece of eighborig uits o a sigle spatial uit. The mai iterest i estimatio of the model is, i geeral, the parameters λ ad β. I order to distiguish the true parameters from other possible values i the parameter space, we deote λ 0,β 0, ad σ0 2 as the true parameters which geerate a observed sample. Let θ =λ, β ) ad θ 0 =λ 0,β 0). This model is supposed to be a equilibrium model. The structural equatio 2.) implies the reduced form equatio that Y =I λ 0 W ) X β 0 +I λ 0 W ) ɛ. 2.2) It follows that W Y = W I λ 0 W ) X β 0 + W I λ 0 W ) ɛ ad W Y is correlated with ɛ because, i geeral, EW I λ 0 W ) ɛ ) ɛ )=σ 2 0trW I λ 0 W ) ) 0. There are some regularity coditios o W ad I λ 0 W ) which will be eeded i order that the spatial correlatios betwee uits ca be maageable. The followig assumptio 3 is origiated i the works of Kelejia ad Prucha, e.g., Kelejia ad Prucha 998). A sequece of square matrices {A }, where A = [a,ij ], is said to be uiformly bouded i row sums colum sums) i absolute value if the sequece of row sum matrix orm A = max i=,, j= a,ij colum sum matrix orm A = max j=,, i= a,ij ) are bouded. 2 Assumptio 3. The spatial weights matrices {W } ad {I λw ) } at λ = λ 0 are uiformly bouded i both row ad colum sums i absolute value. Note that we have imposed the uiform boudedess coditio o {I λ 0 W ) } oly at λ = λ 0. The stroger assumptio that {I λw ) } is uiformly bouded i both row ad colum sums i absolute value, uiformly i λ i a compact parameter space of λ) is ot imposed. 3 With the ormal distributio for ɛ, the ukow parameters λ, β ad σ 2 ca be estimated by the ML or QML) method Ord 975). The ML method ivolves the computatio of the determiat I λw ) of I λw ), at each possible value of λ durig a optimizatio search. For the case that W is rowormalized ad the correspodig spatial matrix before row ormalizatio is symmetric, the eigevalues of 2 Properties of those matrix orms ca be foud i Hor ad Johso 985), pp The latter stroger assumptio is eeded for the maximum likelihood approach. For the GMM method that we propose, because the GMM fuctio is a polyomial fuctio of θ, which is relatively simpler fuctio, our aalysis does ot require the stroger uiform boudedess assumptio. 3

5 W are all real. As I λw ) is solely a fuctio of λ ad eigevalues of W, Ord 975) poits out that computatio of I λw ) ca be easily updated as the eigevalues of W eed to be computed oly oce. 4 With a geeral spatial weights matrix which does ot have special properties, such as sparseess ad the symmetry property i Ord 975), the MLE would be difficult to be computed for sample with a large size. Recetly, Smirov ad Aseli 999) discuss the attractability of usig a characteristic polyomial approach. For a computatioal poit of view, a 2SLS method remais the simplest. Kelejia ad Prucha 998) suggest the use of W X, W 2 X, etc., together with X as IVs i a 2SLS for estimatig θ. Lee 2003) shows that the B2SLS correspods to use W I λ 0 W ) X ad X as IV matrices. These 2SLSE ad B2SLSE are computatioally simple ad have closed form expressios. However, the 2SLS ad B2SL methods have the limitatio i that at least oe ocostat regressors i X must have sigificat coefficiets i order that valid istrumetal variables ca be geerated from them. As the 2SLSE ad B2SLSE are based o the existece of relevat ocostat regressors, it is impossible to test the joit sigificace of X with those estimators. 5 Eve if valid istrumetal variables do exist, the 2SLSE may be iefficiet relative to the MLE. By comparig the limitig variace matrices of the MLE ad 2SLSE e.g., i Aseli 988; Aseli ad Bera 998), either the 2SLSE or the B2SLE have the same limitig distributio of the MLE. I this paper, we suggest to icorporate some other momet coditios i additio to those based o X i order to improve upo the efficiecy of the 2SLSE. Let Q be a k x matrix of IVs costructed as fuctios of X ad W i a 2SLS approach. Deote ɛ θ) =I λw )Y X β for ay possible value θ. The momet fuctios correspodig to the orthogoality coditios of Q ad ɛ are Q ɛ θ). Let P be the class of costat matrices which have a zero trace. A subclass P 2 of P cosistig of matrices with a zero diagoal is also iterestig. By selectig matrices P,...,P m from P, we suggest the use of P j ɛ θ)) ɛ θ) i additio to Q ɛ θ) to form a set of momet fuctios. For aalytical tractability, the matrices i P are assumed to have the uiformly boudedess properties as W. 4 However, Kelejia ad Prucha 999) have poited out that Ord s method may suffer from umerically imprecise problems for large sample. 5 Whe β 0 = 0, the model would be a pure spatial autoregressive process. For a spatial autoregressive model with oly a ozero itercept term but o other spatially varyig regressors, W l will ot be a useful istrumet for W Y whe W is row ormalized. Whe W is row-ormalized, W l = l where l is the vector of oes. 4

6 Assumptio 4. The matrices P j s from P are uiformly bouded i both row ad colum sums i absolute value, ad elemets of Q are uiformly bouded. With the selected matrices P j s ad IV matrices Q, the set of momet fuctios forms a vector g θ) =P ɛ θ),,p m ɛ θ),q ) ɛ θ) =ɛ θ)p ɛ θ),,ɛ θ)p mɛ θ),ɛ θ)q ) 2.3) for the GMM estimatio. At θ 0, g θ 0 )=ɛ P ɛ,,ɛ P m ɛ,ɛ Q ), which has a zero mea because EQ ɛ )=Q Eɛ ) = 0 ad Eɛ P j ɛ )=σ 2 0trP j ) = 0 for j =,,m. 6 The ituitio is as follows. The IV variables i Q, ca be used as IV variables for W Y ad X i 2.). The P j ɛ is ucorrelated with ɛ.asw Y = W I λ 0 W ) X β 0 + ɛ ), P j ɛ ca be used as a IV for W Y as log as P j ɛ ad W I λ 0 W ) ɛ are correlated. For ay possible value θ, d θ)p d θ)+σ0tri 2 λ 0 W ) I λw )P I λw )I λ 0 W ) ) Eg θ)) =. d θ)p md θ)+σ0 2trI λ 0 W ) I λw )P mi λw )I λ 0 W ) ), 2.4) Q d θ) where d θ) =X β 0 β) +λ 0 λ)w I λ 0 W ) X β 0. For these momet fuctios to be useful, they have to idetify the true parameter θ 0 of the model. I the GMM framework, the idetificatio coditio requires the uique solutio of the limitig equatios, lim Eg θ)) = 0 at θ 0 Hase 982). The momet equatios correspodig to Q are lim Q d θ) = lim Q X,Q W I λ 0 W ) X β 0 )β 0 β),λ 0 λ) = 0. They will have a uique solutio at θ 0 if Q X,Q W I λ 0 W ) X β 0 ) has a full colum rak, i.e., rak k + ), for large eough. This sufficiet rak coditio implies the ecessary rak coditio that X,W I λ 0 W ) X β 0 ) has a full colum rak k+) ad that Q has a rak at least k + ), for large eough. The sufficiet rak coditio requires Q to be correlated with W Y i the limit as goes to ifiity. This is so because EQ W Y )=Q W I λ 0 W ) X β 0. Uder the sufficiet rak coditio, θ 0 ca thus be idetified via lim Q d θ) =0. The ecessary full rak coditio of X,W I λ 0 W ) X β 0 ) for large is possible oly if the set cosistig of W I λ 0 W ) X β 0 ad X is ot asymptotically liearly depedet. This rak coditio would ot hold, i particular, if β 0 were zero. There are other cases whe this depedece ca occur see, e.g., Kelejia ad Prucha 998). As X has rak k, ifx,w I λ 0 W ) X β 0 ) does ot have a full rak 6 The selectio of the umber of m is ot a importat issue for our GMM approach, because there exists a best P as discussed i a subsequet sectio. Furthermore, from our Mote Carlo study, the selectio of W ad W 2 trw 2 ) I ) provides accurate approximatios for the best oe. 5

7 k + ), its rak will be k, ad there will exist a costat vector c such that W I λ 0 W ) X β 0 = X c. The, d θ) =X β 0 β +λ 0 λ)c) ad Q d θ) =Q X β 0 β +λ 0 λ)c). The correspodig momet equatios Q d θ) = 0 will have may solutios but the solutios are all described by the relatio that β = β 0 +λ 0 λ)c as log as Q X has a full rak k. Uder this sceario, β 0 ca be idetified oly if λ 0 is idetifiable. The idetificatio of λ 0 will rely o the remaiig quadratic momet equatios lim tri λ 0 W ) I λw )P j I λw )I λ 0 W ) ) = 0 for j =,,m. I this case, Y = λ 0 W I λ 0 W ) X β 0 )+X β 0 +I λ 0 W ) ɛ = X β 0 + λ 0 c)+u, where u =I λ 0 W ) ɛ. The relatioship u = λ 0 W u + ɛ is a SAR process. The idetificatio of λ 0 thus comes from the SAR process u. Oe ca see from Lee 200) that the set of the limitig quadratic momet equatios has a uique solutio at λ 0 if lim [trp s W I λ 0 W ) ),,trp s m W I λ 0 W ) )] is liearly idepedet of lim [tri λ 0 W ) W P W I λ 0 W ) ),,tri λ 0 W ) W P m W I λ 0 W ) )], where A s = A+A for ay square matrix A. The followig assumptio summarizes some sufficiet coditios for the idetificatio of θ 0 from the momet equatios lim Eg θ)) = 0. Assumptio 5. Either i) lim Q W I λ 0 W ) X β 0,X ) has the full rak k +), or ii) lim Q X has the full rak k, lim trp s j W I λ 0 W ) ) 0 for some j, ad lim [trp s W I λ 0 W ) ),,trp s m W I λ 0 W ) )] is liearly idepedet of lim [tri λ 0 W ) W P W I λ 0 W ) ),,tri λ 0 W ) W P m W I λ 0 W ) )]. I terms of computatio of the GMM estimator, because the momet fuctios i g θ) are quadratic fuctios i λ ad β, the GMM objective fuctio will be of polyomial of order four. The derivatio of the GMM estimator will ivolve the miimizatio of a polyomial fuctio i θ. The computatio of polyomial coefficiets, which do ot ivolve the ukow θ, eed to be doe oce. The evaluatio of the correspodig objective fuctio will thus ivolve the multiplicatio of these polyomial coefficiets with powers of θ s at differet values of θ. The computatio is more complicated tha that of the 2SLS but shall be simpler tha that of the ML approach. The variace matrix of these momets fuctios ivolves variaces ad covariaces of liear ad quadratic forms of ɛ. For ay square matrix A, let vec D A) =a,,a ) deote the colum vector formed with the diagoal elemets of A. The, EQ ɛ ɛ P ɛ )=Q i= j= p,ijeɛ ɛ i ɛ j )= Q vec D P )µ 3 ad Eɛ P j ɛ ɛ P l ɛ )=µ 4 3σ0)vec 4 D P j)vec D P l )+σ0trp 4 j Pl s ) by Lemma A.2, 6

8 where µ 3 = Eɛ 3 i) ad µ 4 = Eɛ 4 i). It follows that varg θ 0 )) = Ω where Ω = ) µ4 3σ0)ω 4 mω m µ 3 ω mq µ 3 Q + V ω m 0, 2.5) with ω m =[vec D P ),,vec D P m )] ad trp P) s trp P s m) 0 V = σ0 4.. ). trp m P s ) trp mpm s ) 0 = m 0 σ4 0 0 Q, 2.6) σ Q Q σ0 2 Q 0 where m =[vecp ),,vecp m )] [vecp s ),,vecp s m )], by usig trab) =veca ) vecb) for ay coformable matrices A ad B. Whe ɛ is ormally distributed, Ω is simplified to V because µ 3 = 0 ad µ 4 =3σ 4 0. If P j s are from P 2,Ω = V also because ω m = 0. I geeral, from 2.3), Ω is osigular if ad oly if both vecp ),,vecp m )) ad Q have full colum raks. This is so, because Ω would be sigular if ad oly if the momets i g θ 0 ) are fuctioally depedet, equivaletly, if ad oly if m j= a jp j = 0 ad Q b = 0 for some costat vector a,,a m,b ) 0. As elemets of P j s ad Q are uiformly bouded by Assumptio 4, it is apparet that ω mq, ω mω m ad Q Q are of order O). Furthermore, because P j P s l is bouded i row or colum sums i absolute value, trp jp s l )=O). Cosequetly, Ω = O). It is thus meaigful to impose the followig covetioal regularity coditio o the limit of Ω. Assumptio 6. The limit of Ω exists ad is a osigular matrix. 7 The variace matrix Ω is eeded to formulate the optimum GMME with g θ). 3. Cosistecy ad Asymptotic Distributios The followig propositio provides the asymptotic distributio of the GMME with a liear trasformatio of the momet equatios, a g θ), where a is a matrix with a full row rak greater tha or equal to k + ). The a is assumed to coverge to a costat full rak matrix a 0. This correspods to the Hase s GMM settig, which illustrates the optimal weightig issue. Assumptio 7. θ 0 is i the iterior of the parameter space Θ, which is a compact subset of R k I this paper, for simplicity, we do ot cosider the large group iteractios sceario as i Case 99), where lim Ω might be sigular. It is possible to exted our aalysis to cover that case but the aalysis would become much algebraically complicated. For a aalysis o the MLE, see Lee 2004). 8 For a oliear extremum estimator, the parameter space would usually be assumed to be a compact set Amemiya 985). The extremum estimate always exist whe the objective fuctio is cotiuous o a compact set. Furthermore, for the proof of cosistecy of the extremum estimator, the uiform covergece argumet will usually require a compact parameter space see the uiform covergece theorem i Amemiya). 7

9 Propositio. Uder Assumptios -5, suppose that P j for j =,,m, are from P ad Q is a k x IV matrix so that a 0 lim Eg θ)) = 0 has a uique root at θ 0 i Θ. The, the GMME ˆθ derived from mi θ Θ g θ)a a g θ) is a cosistet estimator of θ 0, ad ˆθ θ 0 ) D N0, Σ), where ad [ Σ = lim D )a a ] D ) D )a a Ω )a a [ D ) D )a a )] D, 3.) D ) σ 2 = 0 trpw s I λ 0 W ) ) σ0trp 2 mw s I λ 0 W ) ) W I λ 0 W ) X β 0 ) Q 0 0 X Q, 3.2) uder the assumptio that lim a D exists ad has the full rak k +). The rak coditios i Assumptio 5 imply that lim g θ 0 ) = 0 has a uique root at θ 0 ad, hece, its correspodig gradiet matrix lim D of 2.8) has rak k + ). I the presece of Assumptio 5, the extra coditios that lim a Eg θ)) = 0 has a uique root at θ 0 ad the limit of a D has full rak k + ), are simply to elimiate the bad choice of a sequece {a } for the liear combiatio a g θ), which may result i the loss of idetificatio from the origial g. If Assumptio 5 were ot satisfied, the coditio that a 0 lim g θ) = 0 has a uique root at θ 0 would ot be satisfied. 9 The idetificatio coditio is specified for the limitig fuctio. The weaker requiremet that a Eg θ)) = 0 has a uique root at θ 0 for large eough is ot sufficiet for θ 0 to be idetifiably uique, because the objective fuctio might flatte out i the limit. From Propositio, with the momet fuctios g θ) i 2.3), the optimal choice of a weightig matrix a a is Ω ) by the geeralized Schwartz iequality. If P j s are selected from the subclass P 2 or ɛ i s are ormally distributed, Ω i 2.5) will be reduced to the simpler matrix V i 2.6). These variace matrices ca be used to form the optimal GMM objective fuctio with g θ). The σ 2, µ 3, ad µ 4 ca be cosistetly estimated by usig estimated residuals of ɛ from a iitial cosistet estimate of θ The Ω For our GMM approach, the momet fuctios are quadratic fuctios of θ. Because of its simple oliear structure, the uiform covergece argumet of the sample objective fuctio after proper ormalizatio ca be easily established as log as the parameter space is bouded. So the compact parameter space assumptio may be relaxed to a bouded set as log as the miimum of the objective fuctio exists i such a parameter space. 9 Whe the set of W I λ 0 W ) X β 0 ad X were liearly depedet, Propositio would ot cover the large group iteractio case of Case 99) because, i this situatio, trp s j W I λ 0 W ) ) would vaish i the limit ad Assumptios 5 ad 6 eeded be stregtheed. The MLE ˆλ of λ 0 i this situatio is kow to have a slower rate of covergece tha that of the MLE ˆβ of β 0 see, Lee 2004)). 0 The detailed proof is straightforward but tedious ad is omitted here 8

10 ca the be cosistetly estimated as ˆΩ. The followig propositio shows that the feasible optimum GMME OGMME) with a cosistetly estimated ˆΩ has the same limitig distributio as that of the OGMME based o Ω. With the optimum GMM objective fuctio, a overidetificatio test is available. Propositio 2. OGMME ˆθ o, Uder Assumptios -6, suppose that ˆΩ ) Ω ) = o P ), the the feasible derived from mi θ Θ g θ)ˆω g θ) based o g θ) i 2.3) with P j s from P has the asymptotic distributio ˆθo, θ 0 ) D N0, lim D Ω D ) ). 3.3) Furthermore, g ˆθ )ˆΩ g ˆθ ) D χ 2 m + k x ) k + )). 3.4) The 2SLSE ˆβ 2sl, would be icosistet whe the set of W I λ 0 W ) X β 0 ad X is liearly depedet. A obvious example is β 0 = 0. Aother example is a model where the oly relevat variable i X is the itercept term l ad W is row-ormalized Kelejia ad Prucha 998). I that case, W I λ 0 W ) X β 0 = β0 λ 0 l because I λ 0 W ) l = λ 0 l ad W l = l. However, eve whe the set of W I λ 0 W ) X β 0 ad X is liearly depedet, the GMM approach may still work because of the additioal momet fuctios with P j s. The asymptotic distributio of the GMME i 2.9) ca be used to formulate a Wald statistic for testig the overall sigificace of all exogeous variables while that of the 2SLSE ca ot. 4. Efficiecy ad the BGMME θ 0 is The optimal GMME ˆθ o, ca be compared with the 2SLSE. With Q as the IV matrix, the 2SLSE of ˆθ 2sl, =[Z Q Q Q ) Q Z ] Z Q Q Q ) Q Y, 4.) where Z =W Y,X ). The asymptotic distributio of ˆθ 2sl, is ˆθ2sl, θ 0 ) D N0,σ0 2 lim { W I λ 0 W ) X β 0,X ) Q Q Q ) Q W I λ 0 W ) X β 0,X )} ), 4.2) uder the assumptios that lim Q Q is osigular ad lim Q W I λ 0 W ) X β 0,X ) has the full colum rak k + ) Kelejia ad Prucha 998). Because the 2SLSE ca be derived from mi θ ɛ θ)q Q Q ) Q ɛ θ), the 2SLS approach is a special case of the GMM estimatio i Propositio 9

11 with a =0, Q Q ) /2 ) ad a g θ) = Q Q ) /2 Q ɛ θ). It follows from Propositio 2, ˆθ o, shall be efficiet relative to ˆθ 2sl,. Withi the 2SLS framework, by the geeralized Schwartz iequality applied to the asymptotic variace of ˆθ 2ls, i 4.2), the best IV matrix Q will be W I λ 0 W ) X β 0,X ). Usig the best IV matrix for Q i the GMM framework, the resultig GMME shall be efficiet relative to the B2SLE. There is a related questio o whether W I λ 0 W ) X β 0,X ) ca be the best IV matrix i the class of matrices Q with give P j s. The aswer ca be affirmative for cases where the momets Q ɛ do ot iteract with the momets ɛ P jɛ via their correlatios. The covariace of Q ɛ ad ɛ P jɛ, j =,,m,isµ 3 Q ω m, which ca be zero whe µ 3 =0orω m =0. The remaiig issue is o the best selectio of P j s. Whe the disturbace ɛ is ormally distributed or P j s are from P 2, D Ω Cm D = m C m ) + σ0 2 W I λ 0 W ) X β 0,X ) W I λ 0 W ) X β 0,X ), 4.3) where C m =[trp s W I λ 0 W ) ),,trp s mw I λ 0 W ) )]. Note that, because trp j P s l )= 2 trp j s P l s ), m ca be rewritte as m = 2. i) Whe P j s are from P 2, trp s P s ) trp s P m s ). trpm s P s ) trp m s P m s ) = 2 [vecp s ) vecp s m)] [vecp s ) vecp s m)]. trp s jw I λ 0 W ) )= 2 tr P s j[w I λ 0 W ) DiagW I λ 0 W ) )] s) = 2 vec [W I λ 0 W ) DiagW I λ 0 W ) )] s) vecp s j ) for j =,,m,ic m, where DiagA) deotes the diagoal matrix formed by the diagoal elemets of a square matrix A. Therefore, the geeralized Schwartz iequality implies that C m m C m 2 vec [W I λ 0 W ) DiagW I λ 0 W ) )] s) vec [W I λ 0 W ) DiagW I λ 0 W ) )] s) = tr [W I λ 0 W ) DiagW I λ 0 W ) )] s W I λ 0 W ) ). Thus, i the subclass P 2,[W I λ 0 W ) DiagW I λ 0 W ) )] ad together with [W I λ 0 W ) X β 0,X ] provide the set of best IV fuctios. Note that the best selected matrix for the quadratic momet is a sigle matrix which is best relative to ay fiite umber of P j. So ay additioal P j i additio to the best oe will ot play a role i asymptotically) efficiet estimatio. This is so also for the best IV matrix Q for liear momets. 0

12 ii) For the case where ɛ is N0,σ 2 0I ), because, for ay P j P, trpj s W I λ 0 W ) )= 2 vec [W I λ 0 W ) trw I λ 0 W ) ) I ] s) vecpj s ), for j =,, m, the geeralized Schwartz iequality implies that C m mc m tr [W I λ 0 W ) trw I λ 0 W ) ) I ] s W I λ 0 W ) ). Hece, i the broader class P,[W I λ 0 W ) trwi λ0w) ) I ]ad[w I λ 0 W ) X β 0,X ] provide the best set of IV fuctios. For all those cases, i the evet that the set of W I λ 0 W ) X β 0 ad X is liearly depedet, W I λ 0 W ) X β 0 is redudat ad the best IV matrix shall simply be X. 2 I practice, with iitial cosistet estimates ˆλ, ˆβ of λ 0 ad β 0, W I λ 0 W ) ca be estimated as W I ˆλ W ), ad W I λ 0 W ) X β 0 by ŴI ˆλ W ) X ˆβ. The correspodig variace matrix V of these best momet fuctios ca be estimated as ˆV. The followig propositio summarizes the results ad shows that the feasible BGMME has the same limitig distributio as the BGMME. Propositio 3. Uder Assumptios -3, suppose that ˆλ is a -cosistet estimate of λ 0, ˆβ is a cosistet estimate of β 0, ad ˆσ 2 is a cosistet estimate of σ 2 0. Withi the class of GMME s derived with P 2, the BGMME ˆθ 2b, has the limitig distributio that ˆθ2b, θ 0 ) D N0, Σ 2b ) where Σ 2b = lim tr[g DiagG )) s G ]+ G σ0 2 X β 0 ) G X β 0 ) with G = W I λ 0 W ), which is assumed to exist. σ 2 0 σ 2 0 X G X β 0 ) σ 2 0 X X ) G X β 0 ) X, 4.4) I the evet that ɛ N0,σ 2 0 I ), withi the broader class of GMME s derived with P, the BGMME ˆθ b, has the limitig distributio that ˆθ b, θ 0 ) D N0, Σ b ) where ) tr[g trg) I ) s G ]+ G σ Σ b = lim 0 2 X β 0 ) G X β 0 ) G σ0 2 X β 0 ) X X σ G 0 2 X β 0 ) X, 4.5) σ X 0 2 which is assumed to exist. 2 The best IV vector W I λ 0 W ) X β 0 together with X will satisfy the idetificatio Assumptio 5i) as log as they are ot liearly depedet i the limit. I the evet that they are, the model idetificatio will follow from the best matrix W I λ 0 W ) trw I λ 0 W ) )I or W I λ 0 W ) DiagW I λ 0 W ) ) for the quadratic momet Lee 200). With those best IV vectors ad matrices, its variace matrix will be osigular ad Assumptio 6 will also be satisfied.

13 Whe ɛ is N0,σ 2 0I ), the model 2.) ca be estimated by the ML method. The log likelihood fuctio of the MRSAR model via its reduced form equatio i 2.2) is l L = 2 l2π) 2 l σ2 +l I λw ) 2σ 2 [Y I λw ) X β] I λw )I λw )[Y I λw ) X β]. 4.6) The asymptotic variace of the MLE ˆθ ml,, ˆσ 2 ml, )is trg 2 )+trg G )+ G σ 2 X β 0 ) G X β 0 ) AsyVarˆθ ml,, ˆσ ml,) 2 0 = σ0 2 X σ G 0 2 X β 0 ) trg ) σ0 2 X G X β 0 σ 2 0 X X 0 trg ) σ σ 4 0 see, e.g., Aseli ad Bera 998), p.256). From the iverse of a partitioed matrix, the asymptotic variace of the MLE ˆθ ml, is trg 2 )+trg G )+ G σ0 AsyV arˆθ ml, )= 2 X β 0 ) G X β 0 ) 2 tr2 G ) σ 2 0 σ 2 0 ) X G X β 0 ). X G X β 0 σ 2 0 X X As trg 2 )+trg G ) 2 tr2 G ) = trg trg) I ) s G ), the GMME ˆθ b, has the same limitig distributio as the MLE of θ 0 from Propositio 3. There is a ituitio o the best GMM approach compared with the maximum likelihood oe. The derivatives of the log likelihood i 4.6) are ad l L σ 2 l L β = σ 2 X ɛ θ), = 2σ 2 + 2σ 4 ɛ θ)ɛ θ), l L λ = trw I λw ) )+ σ 2 [W I λw ) X β] ɛ θ)+ σ 2 ɛ θ)[w I λw ) ] ɛ θ). The equatio l L σ 2 = 0 implies that the MLE is ˆσ 2 θ) = ɛ θ)ɛ θ) for a give value θ. By substitutig ˆσ 2 θ) ito the remaiig likelihood equatios, the MLE ˆθ ml, will be characterized by the momet equatios: X ɛ θ) =0, ad [W I λw ) X β] ɛ θ)+ɛ [W θ) I λw ) ] trw I λw ) ) ɛ θ) =0. The similarity of the best GMM momets ad the above likelihood equatios is revealig. The best GMM approach has the liear ad quadratic momets of ɛ θ) i its formatio ad uses cosistetly estimated matrices i its liear ad quadratic forms. 2

14 5. Some Mote Carlo Results The model i the Mote Carlo study is specified as Y = λw Y + X β + X 2 β 2 + X 3 β 3 + ɛ, where x i, x i2 ad x i3 are three idepedetly geerated stadard ormal variables ad are i.i.d. for all i, ad ɛ i s are i.i.d. N0,σ 2 ). Whe the sample size is = 49, the spatial weights matrix W correspods to the weights matrix for the study of crimes across 49 districts i Columbus, Ohio i Aseli 988). For large sample sizes of = 245 ad 490, the correspodig spatial weights matrices are block diagoal matrices with the precedig matrix as their diagoal blocks. These correspod to the poolig, respectively, of five ad te separate districts with similar eighborig structures i each district. The estimatio methods cosidered are the.) 2SLS the two stage least squares method with IV s X,W X, ad W 2 X ; 2.) GMM a simple uweighted GMM approach usig Q =X,W X,W 2 X ) for liear momets ad W ad W 2 trw 2 ) I for quadratic momets with a idetity matrix as the distace matrix); 3.) OGMM the optimum GMM approach usig Q =X,W X,W 2 X ) for liear momets ad W ad W 2 trw 2 ) I for quadratic momets with the iverse of their estimated) variace matrix as the distace matrix); 3 4.) BGMM the best optimum GMM approach by usig X ad I ˆλ W ) X ˆβ for the liear momets, ad W I ˆλ W ) tr[w I ˆλ W ) ]I for the quadratic momet, where ˆλ, ˆβ ) is a iitial cosistet estimate; 5.) ML the maximum likelihood approach. The umber of repetitios is,000 for each case i this Mote Carlo experimet. The regressors are radomly redraw for each repetitio. I each case, we report the mea Mea ad stadard deviatio SD of the empirical distributios of the estimates. To facilitate the compariso of various estimators, their root mea square errors RMSE are also reported. I all the cases of this study, the true λ 0 is set to 0.6. The smallest sample size is = 49, ad the moderate sample sizes are 245 ad 490. The variace of the equatio errors σ 2 0 is 2. The β coefficiets are varied i the experimets. 3 The ormality of the disturbaces is assumed. The σ 2 i the variace matrix of the momets is estimated with the estimated residuals of the model equatio with the simple GMM estimates as its coefficiets. Note that the miimizatio of a objective fuctio i various GMM approaches is performed globally without imposig a restricted parameter space, such as λ lies i, ), i our study. 3

15 Table reports the results of the case where β 0 =.0, β 20 = 0 ad β 30 =.0. I this case, the correspodig variace ratio of xβ 0 with the sum of variaces of xβ 0 ad ɛ is 0.5. If oe igores the iteractio term, this ratio would represet R 2 =0.5 i a regressio equatio. The results idicate that the mai differeces of the various estimatio approaches are o the estimatio of the spatial effect λ. For the small sample size N = 49, the 2SLS is biased upward by 2.7% ad it has also the largest SD compared with the various GMME s ad the MLE. The MLE is biased dowward by 4% ad the OGMME is biased upward by 6.8%. The GMME ad BGMME are essetially ubiased. The BGMME is ot better tha the GMME ad OGMME i terms of SD ad RMSE with this small sample. Amog the various GMM estimates, the OGMME is better i terms of SD ad RMSE. The MLE has the smallest SD ad RMSE amog all these estimates. The estimates of β s of the various methods do ot have much differeces. The estimates of β s have small biases. Whe icreases to 245 or 490, the upward bias of the 2SLS estimate of λ is reduced. All the other estimates are ubiased. Eve for the moderate sample sizes, the 2SLS estimates of λ have apparetly larger SD ad RMSE s tha the correspodig various GMMEs ad MLEs. The BGMME is slightly better tha the OGMME, ad, i tur, the OGMME is slightly efficiet relative to the GMME. The BGMM is efficiet as the MLE whe N = 490. I Table 2, the true parameters are β = 0.2, β 2 = 0, ad β 3 =0.2. The variace of xβ 0 is much smaller tha the variace of ɛ. If oe igores the iteractio term, the implied R 2 is about 0.04 i a regressio equatio. I this case, the λ may be relatively more difficult to be estimated by the 2SLS. The upward bias of the 2SLSE of λ ca be large. The SD ad RMSE of the 2SLS estimates are larger tha those of the MLE ad various GMME s. The OGMME ca be the best amog the various GMM estimates whe N = 49. The BGMME ca be better tha the other GMMEs with larger N. The BGMME estimates ca be as efficiet as the MLE with large N = 490. For the estimates of the β s, there are ot much differeces amog the various estimates. I summary, the 2SLSE of λ has larger biases ad SD s tha those of the various GMME s. 4 The performace of the 2SLSE becomes worse whe the value of R 2 becomes small. The various GMME s ca substatially improve upo the 2SLSE. The OGMME ad BGMME ca be efficiet as the MLE with large sample sizes. 5 The differeces of the various estimators occur oly for the estimatio of λ but ot for 4 This may be so because our cases have oly moderate or very small R 2 values due to the explaatory variables. 5 However, i some repetitios, the BGMME may be sesitive to iitial cosistet estimates i the costructio of G. The results i Table ad Table 2 are based, respectively, o 2SLSE ad GMME 4

16 estimatio of the β s. 6. Coclusio I this paper, we cosider the estimatio of the MRSAR model. The 2SLS method has bee suggested i the literature for the estimatio of the MRSAR model. The 2SLS method ca be applicable oly if some of the spatially varyig exogeous variables are really relevat. It is impossible i the 2SLS framework to test the overall sigificace of all the exogeous variables i the MRSAR model. It is kow that the 2SLSE does ot attai the same limitig distributio of the MLE uder ormal disturbaces) of the MRSAR model. This paper improves upo the 2SLS approach by itroducig additioal momet fuctios i the GMM framework. The resulted GMME ca be efficiet relative to the 2SLSE. It is possible to derive the best GMME s withi certai classes of GMME s. Oe of the BGMME s ca attai the same limitig distributio of the MLE uder ormal disturbaces). Withi the GMM estimatio framework, it is possible to test the overall sigificace of all the exogeous variables i the model. The GMM approach may, i priciple, be geeralized for the estimatio of MRSAR models with higher order spatial lags ad models with both spatial lags ad/or SAR disturbaces. However, may issues for the models with higher order momets have ot bee well uderstood, for example, the proper parameter space of the lag coefficiets ad its idetificatio problem. These will be studied i aother occasio. as iitial estimates. 5

17 Appedix A: Some Useful Lemmas I this appedix, we list some lemmas which are useful for the proofs of the results i the text. Lemma A. Suppose that the sequeces of -dimesioal colum vectors {z } ad {z 2 } are uiformly bouded. If {A } are uiformly bouded i either row or colum sums i absolute value, the z A z 2 = O). Proof: This is trivial. Lemma A.2 Suppose that ɛ,,ɛ are i.i.d. radom variables with zero mea, fiite variace σ 2 ad fiite fourth momet µ 4. The, for ay two square matrices A ad B, Eɛ Aɛ ɛ Bɛ )=µ 4 3σ 4 0)vec DA)vec D B)+σ 4 0[trA)trB)+trAB s )], where B s = B + B. Proof: This is a Lemma i Lee 200). Lemma A.3 Suppose that {A } are uiformly bouded i both row ad colum sums i absolute value. The ɛ,...,ɛ are i.i.d. with zero mea ad its fourth momet exists. The, Eɛ A ɛ )=O), varɛ A ɛ )=O), ɛ A ɛ = O P ), ad ɛ A ɛ Eɛ A ɛ )=o P ). Proof: Lee 200). Lemma A.4 Suppose that A is a matrix with its colum sums beig uiformly bouded i absolute value, elemets of the k matrix C are uiformly bouded, ad ɛ,,ɛ are i.i.d. with zero mea ad fiite variace σ 2. The, C A ɛ = O P ) ad C A ɛ = o p ). Furthermore, if the limit of C A A C exists ad is positive defiite, the C A ɛ D N0,σ 2 lim C A A C ). Proof: See Lee 2004). Lemma A.5 Suppose that {A } is a sequece of symmetric matrices with row ad colum sums uiformly bouded i absolute value ad b = b,,b ) is a -dimesioal vector such that sup i= b i 2+η < for some η > 0. The ɛ,,ɛ are i.i.d. radom variables with zero mea ad fiite variace σ 2, ad its momet E ɛ 4+2δ ) for some δ>0 exists. Let σ 2 Q be the variace of Q where Q = ɛ A ɛ + b ɛ σ 2 tra ). Assume that the variace σ 2 Q is bouded away from zero at the rate. The, Q σ Q D N0, ). Proof: See Kelejia ad Prucha 200). Lemma A.6 Suppose that Q θ) Q θ)) coverges i probability to zero uiformly i θ Θ which is a covex set, ad { Q θ)} satisfies the idetificatio uiqueess coditio at θ 0. Let ˆθ ad ˆθ be, 6

18 respectively, the miimizers of Q θ) ad Q θ) i Θ. If Q θ) Q θ)) = o P ) uiformly i θ Θ, the both ˆθ ad ˆθ coverge i probability to θ 0. I additio, suppose that 2 Q θ) coverges i probability to a well defied limitig matrix, uiformly i θ Θ, which is osigular at θ 0, ad Q θ 0) = O P ). If 2 Q θ) i θ Θ ad Q θ0) distributio. 2 Q θ) )=o P ) uiformly Qθ0) )=o P ), the ˆθ θ 0 ) ad ˆθ θ 0 ) have the same limitig Proof: The covergece of ˆθ to θ 0 follows from the uiform covergece of Q θ) Q θ)) to zero i probability ad the uiqueess idetificatio coditio of { Q θ)} White 994). As Q θ) Q θ)) = Q θ) Q θ))+ Q θ) Q θ)) = o P ) uiformly i θ Θ, the covergece of ˆθ to θ 0 i probability follows. For the limitig distributio, the Taylor expasio of Q θ) ˆθ θ 0 )= = = 2 Q ) θ ) Q θ 0 ) ) 2 Q θ ) + o P ) 2 Q θ ) at θ 0 implies that Q θ 0 ) ) Q θ 0 ) + o P ). + o P )) Thus, ˆθ θ 0 ) ad ˆθ θ 0 ) have the same limitig distributio. Q.E.D. Lemma A.7 Suppose that the elemets of the k matrices X are uiformly bouded for all ; ad lim X X exists ad this limitig matrix is osigular, the the projectors, X X X ) X ad I X X X ) X, are uiformly bouded i both row ad colum sums i absolute value. Proof: See Lee 2004). Lemma A.8 Suppose that { W } ad { S λ 0 ) }, where is a matrix orm, are bouded. The { S λ) }, where S λ) =I λw, are uiformly bouded i a eighborhood of λ 0. Proof: See Lee 2004). I the followig Lemmas ad Appedix B, some simplified otatios shall be used to miimize the presetatio of mathematical terms. For ay square matrix A, we shall deote the adjusted matrix A tra) I )ora DiagA)) by A d. Furthermore, with the spatial weights matrix W, deote S λ) = I λw, S = S λ 0 ), G λ) =W I λw ), ad G = G λ 0 ). Lemma A.9 Suppose that z ad z 2 are -dimesioal colum vectors of costats which elemets are uiformly bouded, the costat matrices A are uiformly bouded i the maximum colum sum orm, ad ɛ i s i ɛ =ɛ,,ɛ ), are i.i.d. with zero mea ad a fiite variace σ 2. Let ˆλ be a 7

19 -cosistet estimate of λ0. The, uder Assumptio 3, ) z G ˆλ ) G ) z 2 = o P ), z G ˆλ ) G ) d z 2 = o P ); ad 2) z G ˆλ ) G ) A ɛ = o P ), z G ˆλ ) G ) d A ɛ = o P ). Proof: As S S ˆλ ) = ˆλ λ 0 )W, it follows that G ˆλ ) G = W [S ˆλ ) S ] = W S ˆλ )[S S ˆλ )]S =ˆλ λ 0 )G ˆλ )G, ad G ˆλ ) G ) d =ˆλ λ 0 )G ˆλ )G ) d. A further expasio implies that G ˆλ ) G =ˆλ λ 0 )G 2 +ˆλ λ 0 ) 2 G ˆλ )G 2 ad G ˆλ ) G ) d = ˆλ λ 0 )G 2d +ˆλ λ 0 ) 2 G ˆλ )G 2 )d. We ote that, uder Assumptio 3, because S is uiformly bouded i both row ad colum sums i absolute value, S λ) ad, hece, G λ) must be uiformly bouded i both row ad colum sums i absolute value uiformly i λ i a small eighborhood of λ 0 by Lemma A.8. As ˆλ is cosistet, it follow that G ˆλ ) is uiformly bouded i both row ad colum sums i absolute value with probability oe. Therefore, Lemma A. implies that z G G ˆλ )z 2 = O P ). Hece, z G ˆλ ) G ) z 2 =ˆλ λ 0 ) z G G ˆλ )z 2 = o P ) as ˆλ λ 0 = o P ). Similarly, z G ˆλ ) G ) d z 2 = o P ). This proves ). For 2), by the further expasio of G ˆλ ) aroud G, z G ˆλ ) G ) A ɛ = ˆλ λ 0 ) z G 2 A ɛ +ˆλ λ 0 ) 2 z G 2 G ˆλ )A ɛ. Lemma A.4 implies that z G 2 A ɛ = o P ). Therefore, with the -cosistet ˆλ, the first term o the right had side is o P ). The remaider term is also o P ). This is so as follows. Let be the maximum colum sum orm. Because the product of matrices uiformly bouded i colum sums i absolute value is uiformly bouded i colum sums i absolute value, G G ˆλ )A c for some costat c for all. As elemets of z are uiformly bouded, there exists a costat c 2 such that z c 2. It follows that ˆλ λ 0 ) 2 z 2 G G ˆλ )A ɛ [ ˆλ λ 0 )] 2 ɛ 3/2 z G 2 G ˆλ )A c c 2 [ ˆλ λ 0 )] 2 ɛ i. 2 The result 2) follows because λ λ 0 )=o P ) ad i= ɛ i = O P ) by the strog law of large i= umbers. Similarly argumets are applicable to z G ˆλ ) G ) d A ɛ. Q.E.D. Lemma A.0 Let A ad B be matrices, uiformly bouded i both row ad colum sums i absolute value. The ɛ i s i ɛ =ɛ,,ɛ ) are i.i.d. with zero mea ad its fourth momet exists. Suppose that ˆλ is a -cosistet estimator of λ 0. The, uder Assumptio 3, i) ɛ A G ˆλ ) G ) d B ɛ = o P ), ad 8

20 ii) ɛ G ˆλ ) G ) d ɛ = o P ). Proof: This is a case of Lemma A.9 i Lee 200). Q.E.D. Appedix B: Proofs Proof of Propositio. For cosistecy, we first show that a g θ) a Eg θ)) will coverge i probability uiformly i θ Θ to zero. Let a =a,,a m,a x ) where a x is a row) subvector. The a g θ) =ɛ θ) m j= a jp j )ɛ θ)+a x Q ɛ θ). By expasio, ɛ θ) =d θ)+ɛ +λ 0 λ)g ɛ where d θ) =λ 0 λ)g X β 0 + X β 0 β). It follows that m m ɛ θ) a j P j )ɛ θ) =d θ) a j P j )d θ)+l θ)+q θ), j= j= where l θ) =d θ) m j= a jp s j)ɛ +λ 0 λ)g ɛ ) ad q θ) =ɛ +λ 0 λ)ɛ G ) m j= a jp j )ɛ + λ 0 λ)g ɛ ). The term l θ) is liear i ɛ. By expasio, l θ) =λ 0 λ) X β 0 ) G m a j Pj)ɛ s +β 0 β) m X a j Pj)ɛ s j= +λ 0 λ) 2 m X β 0 ) G a j Pj)G s ɛ +λ 0 λ)β 0 β) m X a j Pj)G s ɛ = o P ), j= by Lemma A.4, uiformly i θ Θ. The uiform covergece i probability follows because l θ) is simply a quadratic fuctio of λ ad β ad Θ is a bouded set. Similarly, q θ) = ɛ m a j P j )ɛ +λ 0 λ) m ɛ G a j Pj)ɛ s +λ 0 λ) 2 m ɛ G a j P j )G ɛ j= =λ 0 λ) σ2 0 m j= j= a j trg P s j)+λ 0 λ) 2 σ2 0 j= j= j= m a j trg P j G )+o P ), uiformly i θ Θ, by Lemma A.3 ad Eɛ P j ɛ )=σ 2 0trP j ) = 0 for all j =,,m. Cosequetly, ɛ θ) m a j P j )ɛ θ) = m d θ) j= j= +λ 0 λ) 2 σ2 0 j= a j P j )d θ)+λ 0 λ) σ2 0 m a j trpjg s ) j= m a j trg P jg )+o P ), uiformly i θ Θ. As g θ) is a quadratic fuctio of θ ad Θ is bouded, j= a Eg θ)) is uiformly equicotiuous o Θ. The idetificatio coditio ad the uiform equicotiuity of a Eg θ)) imply that the idetificatio uiqueess coditio for 2 Eg θ)a a Eg θ)) must be satisfied. The cosistecy of the GMME ˆθ follows from the uiform covergece ad the idetificatio uiqueess coditio White 994). 9

21 For the asymptotic distributio of ˆθ, by the Taylor expasio of g ˆθ ) a a g ˆθ )=0atθ 0, As ɛθ) ˆθ θ 0 )= [ g ˆθ ) = W Y,X ), it follows that gθ) a a g θ ) ] g ˆθ ) a a g θ 0 ). = P s ɛ θ),,p s mɛ θ),q ) W Y,X ). Explicitly, ɛ θ)p j s W Y = ɛ θ)p j s G X β 0 + ɛ θ)p j s G ɛ. By Lemmas A.4 ad A.3, ɛ θ)pjg s X β 0 = d θ)pjg s X β 0 + ɛ PjG s X β 0 +λ 0 λ) ɛ G PjG s X β 0 ad uiformly i θ Θ. Hece, = d θ)p s jg X β 0 + o P ), ɛ θ)p s jg ɛ = d θ)p s jg ɛ + ɛ P s jg ɛ + λ 0 λ)ɛ G P s jg ɛ = σ2 0 trp s jg )+λ 0 λ) σ2 0 trg P s jg )+o P ), ɛ θ)p s jw Y = d θ)p s jg X β 0 + σs 0 trp s jg )+λ 0 λ) σ2 0 trg P s jg )+o P ), uiformly i θ Θ. At θ 0, d θ 0 ) = 0 ad, hece, ɛ θ 0)Pj s W Y = σ2 0 trpj s G )+o P ). At θ 0, ɛ θ 0 )Pj s X = o P ). Fially, Q W Y = Q G X β 0 + Q G ɛ = Q G X β 0 + o P ). I coclusio, g θ ) = D + o P ) with D i 3.2). O the other had, Lemma A.5 implies that a g θ 0 )= [ɛ m j= a j P j )ɛ + a x Q ɛ ] D N0, lim a Ω a ). The asymptotic distributio of ˆλ λ 0 ) follows. Q.E.D. Proof of Propositio 2. The geeralized Schwartz iequality implies that the optimal weightig matrix for a a i Propositio is Ω ). For cosistecy, cosider g θ)ˆω g θ) = g θ)ω g θ)+ g θ)ˆω Ω )g θ). With a = Ω ) /2 i Propositio, Assumptio 6 implies that a 0 = lim Ω ) /2 exits. Because a 0 is osigular, the idetificatio coditio of θ 0 correspods to the uique root of lim E g θ)) = 0 at θ 0, which is satisfied by Assumptio 5. Hece, the uiform covergece i probability of g θ)ω g θ) to a well defied limit uiformly i θ Θ follows by a similar argumet i the proof of Propositio. So it remais to show that g θ)ˆω orm for vectors ad matrices. The, Ω )g θ) =o P ) uiformly i θ Θ. Let be the Euclidea g θ)ˆω Ω )g θ) g θ) ) 2 ˆΩ ) Ω ). 20

22 To show that this is of probability order o P ) uiformly i θ Θ, it is sufficiet to show that g θ) = O P ) uiformly i θ Θ. From the proof of Propositio, [g θ) Eg θ))] = o P ) uiformly i θ Θ. O the other had, as d θ)p j d θ) =λ 0 λ) 2 X β 0 ) G P j G X β 0 +λ 0 λ) X β 0 ) G P s jx β 0 β) +β 0 β) X P j X β 0 β) =O P ), uiformly i θ Θ by Lemma A., it follows that Eɛ θ)p jɛ θ)) = d θ)p jd θ)+λ 0 λ)σ0 2 trp j s G )+λ 0 λ) 2 σ0 2 trg P jg )=O), uiformly i θ Θ. Similarly, EQ ɛ θ)) = Q d θ) =λ 0 λ) Q G X β 0 + Q X β 0 β) =O) uiformly i θ Θ. These imply that Eg θ)) = O) uiformly i θ Θ. Cosequetly, by the Markov iequality, g θ) = O P ) uiformly i θ Θ. Therefore, g θ)ˆω Ω )g θ) = o P ), uiformly i θ Θ. The cosistecy of the feasible optimum GMME ˆθ o, follows. For the limitig distributio, as g ˆθ ) = D + o P ) from the proof of Propositio, ˆθo, θ 0 )= = [ D g ˆθ ) Ω ) ˆΩ ) D ] D g ˆθ ) Ω g ˆθ ) ) g θ 0 )+o P ). ) ˆΩ g θ 0 ) The limitig distributio of ˆθ o θ 0 ) i 3.3) follows from this expasio. For the overidetificatio test, by the Taylor expasio, g ˆθ o, )= g θ 0 ) + where A = I D [ D g ˆθ o )ˆΩ g ˆθ o ) = g θ 0)A Ω = g θ 0) Ω ) /2 Ω D χ 2 m + k x ) k + )), g θ ) ˆθo, θ 0 )= g θ 0 ) ) D ] D Ω ). Therefore, ) A g θ 0 )+o P ) D g θ 0 ) ˆθo, θ 0 )+o P ) = A + o P ), { I Ω D ) /2 [D Ω D } ) ] D Ω ) /2 Ω ) /2 g θ 0 )+o P ) because g θ 0 ) D N0, lim Ω ) as i the proof of Propositio. Q.E.D. 2

23 Proof of Propositio 3. For the feasible best GMM estimatio with P, the vector of momet fuctios is ĝ b, θ) =ɛ θ)ĝ trĝ ) I )ɛ θ),ɛ θ)ĝx ˆβ,ɛ θ)x ), ad the correspodig estimated V is ) ˆV =ˆσ 4 trĝ trĝ ) I ) s Ĝ ) 0. 0 ˆσ ĜX 2 ˆβ,X ) ĜX ˆβ,X ) Whe G X β 0 ad X are liearly depedet, the liear momet of Ĝ X ˆβ would be redudat ad should be dropped. The momet fuctio g b, θ) ad the correspodig proper weightig fuctio ˆV should simply be ĝ b, θ) =ɛ θ)ĝ trĝ ) I )ɛ θ),ɛ θ)x ), ad ˆV =ˆσ 4 trĝ trĝ ) I ) sĝ ) 0 0 ˆσ 2 X X ) The feasible best GMME with P will be derived from mi θ Θ g b, θ) ˆV g b, θ). For the best GMM estimatio with the subclass P 2, its momet fuctios ĝ 2b, ad estimated variace matrix are those with Ĝ DiagĜ ) replacig Ĝ trĝ ) I ) i the precedig expressios. We shall show that the objective fuctios Q θ) =ĝ b, θ) ˆV ĝ b, θ) ad Q θ) =g b, θ)v g b, θ), where g b, is the couter part of ĝ b, with G ad β 0 replacig, respectively, Ĝ ad ˆβ, will satisfy the coditios i Lemma A.6. If so, the GMME from the miimizatio of Q θ) will have the same limitig distributio as that of the miimizatio of Q θ). The differece of Q θ) ad Q θ) ad its derivatives ). ivolve the differece of ĝ b, θ) ad g b, θ) ad their derivatives. Furthermore, oe has to cosider the differece of ˆV ad V. First, cosider ĝ b,θ) g b, θ)). Explicitly, ĝ b,θ) g b, θ)) = ɛ θ)ĝ G ) d ɛ θ), ) Ĝ X ˆβ G X β 0 ) ɛ θ), 0. The ɛ θ) is related to ɛ as ɛ θ) =ɛ +λ 0 λ)g ɛ +d θ) where d θ) =λ 0 λ)g X β 0 +X β 0 β). It follows that X G ɛ θ) = X G ɛ +λ 0 λ) X G G ɛ + X G d θ) = X G d θ) +o P ) by Lemma A.4, uiformly i θ Θ. O the other had, Lemma A. implies that X G d θ) =O P ) uiformly i θ Θ. The uiformity follows because d θ) is liear i λ ad β. Hece X G ɛ θ) =O P ) uiformly i θ Θ. Similarly, Lemma A.9 implies X Ĝ G ) ɛ θ) = X Ĝ G ) ɛ +λ 0 λ) X Ĝ G ) G ɛ + X Ĝ G ) d θ) =o P ) uiformly i θ Θ. It follows that Ĝ X ˆβ G X β 0 ) ɛ θ) = ˆβ X Ĝ G ) ɛ θ) +ˆβ β 0 ) X G ɛ θ) = o P ) because ˆβ β 0 = o P ). Similarly, Lemmas A.9 ad A.0 imply that ɛ θ)ĝ G ) d ɛ θ) =o P ) uiformly i θ Θ. Hece, we coclude that ĝ b,θ) g b, θ)) = o P ) uiformly i θ Θ. 22

24 Cosider the derivatives of ĝ b, θ) ad g b, θ). As the secod derivatives of ɛ θ) with respect to θ are zero because ɛ θ) is liear i θ, it follows that ɛ g b, θ) θ)g ds ɛ θ) = G X β 0 ) ɛ θ) X ɛ θ), 2 g b, θ) = ɛ θ) G ds ɛ θ) 0. 0 The first order derivatives of ɛ θ) is ɛθ) = W Y,X ). Because W Y = G X β 0 + G ɛ, W Y ) Ĝ G ) ds ɛ θ) = G X β 0 ) Ĝ G ) ds d θ)+ G X β 0 ) Ĝ G ) ds ɛ +λ 0 λ)g ɛ ) + d θ)ĝ G ) ds G ɛ + ɛ G Ĝ G ) ds ɛ +λ 0 λ)g ɛ )=o P ), uiformly i θ Θ, ad W Y ) Ĝ G ) ds W Y = X β 0 ) G Ĝ G ) ds G X β X β 0 ) G Ĝ G ) ds G ɛ + ɛ G Ĝ G ) ds G ɛ = o P ), by Lemmas A.9 ad A.0. Similarly, Lemmas A.9 ad A.0 imply that X Ĝ G ) ds ɛ θ) =o P ), X Ĝ G ) ds W Y = o P ), ad X Ĝ G ) ds X = o p ). Hece, we coclude that ĝb,θ) g b,θ) )=o P ) ad 2 ĝ b,θ) 2 g b,θ) )=o P ) uiformly i θ Θ. For the cases uder cosideratio, ˆV ad V are block diagoal matrices. Without loss of geerality, cosider the situatio that G X β 0 ad X are ot liearly depedet for large. Thus, ˆV = trĝ ds Ĝ) ) ) 0 trg ds 0 ˆσ  2 G, V = ) 0 0 σ0a 2, where A =G X β 0,X ) G X β 0,X ) ad  =Ĝ X ˆβ,X ) Ĝ X ˆβ,X ). The differece of Ĝ ad G is Ĝ G )=ˆλ λ 0 )G 2 +ˆλ λ 0 ) 2 Ĝ G 2. This implies that trgds Ĝ G )) = ˆλ λ 0 ) trgds G 2 )+ˆλ λ 0 ) 2 trgds ĜG 2 )=o P ), because trgds G 2 )=O), trgds ĜG 2 )=O P ) ad ˆλ λ 0 )=o P ). Similarly, tr[ĝ G ) ds Ĝ ]=ˆλ λ 0 ) trg2ds Ĝ)+ˆλ λ 0 ) 2 tr[ĝ G 2 )ds Ĝ ]=o P ). Therefore, [trĝds ds Ĝ) trg G )] = tr[ĝds G ds )Ĝ+G ds Ĝ G )] = o P ). Cosider the remaiig block matrix. Because ˆσ 2 is a cosistet estimate of σ 2 0 ad A = O) by Lemma A., ˆσ2  σ0a 2 )=ˆσ 2  A )+ˆσ 2 σ0) 2 A =ˆσ 2  A )+o P ). 23

25 The differece  A )iso p ) because Ĝ X ˆβ ) X G X β 0 ) X = ˆβ X Ĝ G )X + ˆβ β 0 ) X G X = o P ) ad ĜX ˆβ ) ĜX ˆβ ) G X β 0 ) G X β 0 ) = ˆβ X Ĝ G ) Ĝ X ˆβ + ˆβ X G Ĝ G )X ˆβ +ˆβ + β 0 ) X G G X ˆβ β 0 )=o P ), by Lemmas A. ad A.9. I coclusio, ˆV V = o P ). It follows that ˆV ) V ) = o P ) by the cotiuous mappig theorem. Furthermore, because ĝ b,θ) g b, θ)) = o P ), ad [g b,θ) Eg b, θ))] = o P ) uiformly i θ Θ, ad sup θ Θ Eg b,θ)) = O) i the proof of Propositio 2, g b,θ) ad ĝb,θ) are stochastically bouded, uiformly i θ Θ. Similarly, g b,θ), ĝ b,θ), 2 g b,θ) ad 2 ĝ b,θ) are stochastically bouded, uiformly i θ Θ. With the uiform covergece i probability ad uiformly stochastic boudedess properties, the differece of Q θ) ad Q θ) ca be ivestigated. By expasio, Q θ) Q θ)) = ĝ b,θ) ˆV ĝ b, θ) g b, θ)) + g b,θ) ˆV uiformly i θ Θ. Similarly, for each compoet θ l of θ, 2 Q θ) 2 Q θ) l l = 2 [ ĝ b, θ) ˆV ĝ b, θ) +ĝ l b, θ) ˆV 2 ĝ b, θ) l = o P ). Fially, because ĝ b, θ0) ˆV limit theorems i Lemmas A.4 ad A.5, Q θ 0) =2{ ĝ b, θ 0) =2 ĝ b, θ 0) Q θ 0 ) ) ˆV ˆV This differece will be of order o P ) if V )ĝ b, + g b,θ)v ĝ b, θ) g b, θ)) = o P ), g b,θ) l V g b, θ) + g b, θ)v 2 g b, θ) l )] g b, θ0) V )=o P ) as above, ad g b, θ 0 )=O P ) by the cetral ĝ b, θ 0 ) g b, θ 0 )) + ĝ b,θ 0 ) ĝ b, θ 0 ) g b, θ 0 )) + o P ). ˆV ĝ b, θ 0 ) g b, θ 0 )) = o P ). g b, θ 0) V ) g b, θ 0 )} Lemma A.0 implies that the compoet ɛ Ĝ G ) d ɛ = o P ). Lemmas A.4 ad A.9 imply that [ĜX ˆβ ) G X β 0 ) ]ɛ = ˆβ X Ĝ G ) ɛ +ˆβ β 0 ) X G ɛ = o P ), ad ˆβ X ɛ β 0 X ɛ )=ˆβ β 0 ) X ɛ = o P ). Hece ĝ b, θ 0 ) g b, θ 0 )) = o P ). Fially, the results of the propositio follows from Lemma A.6 ad the precedig propositios. Q.E.D. 24

GMM and 2SLS Estimation of Mixed Regressive, Spatial Autoregressive Models. Lung-fei Lee* Department of Economics. Ohio State University, Columbus, OH

GMM and 2SLS Estimation of Mixed Regressive, Spatial Autoregressive Models. Lung-fei Lee* Department of Economics. Ohio State University, Columbus, OH GMM ad 2SLS Estimatio of Mixed Regressive, Spatial Autoregressive Models by Lug-fei Lee* Departmet of Ecoomics Ohio State Uiversity, Columbus, OH October 200 Abstract The GMM method ad the classical 2SLS

More information

11 THE GMM ESTIMATION

11 THE GMM ESTIMATION Cotets THE GMM ESTIMATION 2. Cosistecy ad Asymptotic Normality..................... 3.2 Regularity Coditios ad Idetificatio..................... 4.3 The GMM Iterpretatio of the OLS Estimatio.................

More information

Statistical Inference Based on Extremum Estimators

Statistical Inference Based on Extremum Estimators T. Rotheberg Fall, 2007 Statistical Iferece Based o Extremum Estimators Itroductio Suppose 0, the true value of a p-dimesioal parameter, is kow to lie i some subset S R p : Ofte we choose to estimate 0

More information

Slide Set 13 Linear Model with Endogenous Regressors and the GMM estimator

Slide Set 13 Linear Model with Endogenous Regressors and the GMM estimator Slide Set 13 Liear Model with Edogeous Regressors ad the GMM estimator Pietro Coretto pcoretto@uisa.it Ecoometrics Master i Ecoomics ad Fiace (MEF) Uiversità degli Studi di Napoli Federico II Versio: Friday

More information

Journal of Econometrics

Journal of Econometrics Joural of Ecoometrics 59 () 9 Cotets lists available at ScieceDirect Joural of Ecoometrics joural homepage: wwwelseviercom/locate/jecoom A efficiet GMM estimator of spatial autoregressive models Xiaodog

More information

Economics 241B Relation to Method of Moments and Maximum Likelihood OLSE as a Maximum Likelihood Estimator

Economics 241B Relation to Method of Moments and Maximum Likelihood OLSE as a Maximum Likelihood Estimator Ecoomics 24B Relatio to Method of Momets ad Maximum Likelihood OLSE as a Maximum Likelihood Estimator Uder Assumptio 5 we have speci ed the distributio of the error, so we ca estimate the model parameters

More information

A supplement to Asymptotic Distributions of Quasi-Maximum Likelihood Estimators for Spatial Autoregressive. Appendix A: Some Useful Lemmas

A supplement to Asymptotic Distributions of Quasi-Maximum Likelihood Estimators for Spatial Autoregressive. Appendix A: Some Useful Lemmas A supplemet to Asymptotic Distributios of Quasi-Maximum Likelihood Estimators for Spatial Autoregressive Models (for referece oly; ot for publicatio) Appedix A: Some Useful Lemmas A. Uiform Boudedess of

More information

GMM estimation of spatial autoregressive models with unknown heteroskedasticity

GMM estimation of spatial autoregressive models with unknown heteroskedasticity Accepted Mauscript GMM estimatio of spatial autoregressive models with ukow heteroskedasticity Xu Li, Lug-fei Lee PII: S0304-4076(09)00288-7 DOI: 0.06/j.jecoom.2009.0.035 Referece: ECONOM 3288 To appear

More information

Asymptotic Results for the Linear Regression Model

Asymptotic Results for the Linear Regression Model Asymptotic Results for the Liear Regressio Model C. Fli November 29, 2000 1. Asymptotic Results uder Classical Assumptios The followig results apply to the liear regressio model y = Xβ + ε, where X is

More information

Efficient GMM LECTURE 12 GMM II

Efficient GMM LECTURE 12 GMM II DECEMBER 1 010 LECTURE 1 II Efficiet The estimator depeds o the choice of the weight matrix A. The efficiet estimator is the oe that has the smallest asymptotic variace amog all estimators defied by differet

More information

Lecture 19: Convergence

Lecture 19: Convergence Lecture 19: Covergece Asymptotic approach I statistical aalysis or iferece, a key to the success of fidig a good procedure is beig able to fid some momets ad/or distributios of various statistics. I may

More information

Notes On Median and Quantile Regression. James L. Powell Department of Economics University of California, Berkeley

Notes On Median and Quantile Regression. James L. Powell Department of Economics University of California, Berkeley Notes O Media ad Quatile Regressio James L. Powell Departmet of Ecoomics Uiversity of Califoria, Berkeley Coditioal Media Restrictios ad Least Absolute Deviatios It is well-kow that the expected value

More information

Regression with an Evaporating Logarithmic Trend

Regression with an Evaporating Logarithmic Trend Regressio with a Evaporatig Logarithmic Tred Peter C. B. Phillips Cowles Foudatio, Yale Uiversity, Uiversity of Aucklad & Uiversity of York ad Yixiao Su Departmet of Ecoomics Yale Uiversity October 5,

More information

Properties and Hypothesis Testing

Properties and Hypothesis Testing Chapter 3 Properties ad Hypothesis Testig 3.1 Types of data The regressio techiques developed i previous chapters ca be applied to three differet kids of data. 1. Cross-sectioal data. 2. Time series data.

More information

Random Variables, Sampling and Estimation

Random Variables, Sampling and Estimation Chapter 1 Radom Variables, Samplig ad Estimatio 1.1 Itroductio This chapter will cover the most importat basic statistical theory you eed i order to uderstad the ecoometric material that will be comig

More information

Resampling Methods. X (1/2), i.e., Pr (X i m) = 1/2. We order the data: X (1) X (2) X (n). Define the sample median: ( n.

Resampling Methods. X (1/2), i.e., Pr (X i m) = 1/2. We order the data: X (1) X (2) X (n). Define the sample median: ( n. Jauary 1, 2019 Resamplig Methods Motivatio We have so may estimators with the property θ θ d N 0, σ 2 We ca also write θ a N θ, σ 2 /, where a meas approximately distributed as Oce we have a cosistet estimator

More information

GMM Estimation of Spatial Autoregressive Models with Autoregressive and Heteroskedastic Disturbances

GMM Estimation of Spatial Autoregressive Models with Autoregressive and Heteroskedastic Disturbances City Uiversity of New York (CUNY) CUNY Academic Works Ecoomics Workig Papers CUNY Academic Works 203 GMM Estimatio of Spatial Autoregressive Models with Autoregressive ad Heteroskedastic Disturbaces Osma

More information

1 Covariance Estimation

1 Covariance Estimation Eco 75 Lecture 5 Covariace Estimatio ad Optimal Weightig Matrices I this lecture, we cosider estimatio of the asymptotic covariace matrix B B of the extremum estimator b : Covariace Estimatio Lemma 4.

More information

Essays in Spatial Econometrics: Estimation, Specification Test and the Bootstrap. Dissertation

Essays in Spatial Econometrics: Estimation, Specification Test and the Bootstrap. Dissertation Essays i Spatial Ecoometrics: Estimatio, Specificatio Test ad the Bootstrap Dissertatio Preseted i Partial Fulfillmet of the Requiremets for the Degree Doctor of Philosophy i the Graduate School of The

More information

ECONOMETRIC THEORY. MODULE XIII Lecture - 34 Asymptotic Theory and Stochastic Regressors

ECONOMETRIC THEORY. MODULE XIII Lecture - 34 Asymptotic Theory and Stochastic Regressors ECONOMETRIC THEORY MODULE XIII Lecture - 34 Asymptotic Theory ad Stochastic Regressors Dr. Shalabh Departmet of Mathematics ad Statistics Idia Istitute of Techology Kapur Asymptotic theory The asymptotic

More information

Summary and Discussion on Simultaneous Analysis of Lasso and Dantzig Selector

Summary and Discussion on Simultaneous Analysis of Lasso and Dantzig Selector Summary ad Discussio o Simultaeous Aalysis of Lasso ad Datzig Selector STAT732, Sprig 28 Duzhe Wag May 4, 28 Abstract This is a discussio o the work i Bickel, Ritov ad Tsybakov (29). We begi with a short

More information

Chapter 3. Strong convergence. 3.1 Definition of almost sure convergence

Chapter 3. Strong convergence. 3.1 Definition of almost sure convergence Chapter 3 Strog covergece As poited out i the Chapter 2, there are multiple ways to defie the otio of covergece of a sequece of radom variables. That chapter defied covergece i probability, covergece i

More information

An Introduction to Asymptotic Theory

An Introduction to Asymptotic Theory A Itroductio to Asymptotic Theory Pig Yu School of Ecoomics ad Fiace The Uiversity of Hog Kog Pig Yu (HKU) Asymptotic Theory 1 / 20 Five Weapos i Asymptotic Theory Five Weapos i Asymptotic Theory Pig Yu

More information

Cox-type Tests for Competing Spatial Autoregressive Models with Spatial Autoregressive Disturbances

Cox-type Tests for Competing Spatial Autoregressive Models with Spatial Autoregressive Disturbances Cox-type Tests for Competig Spatial Autoregressive Models with Spatial Autoregressive Disturbaces Fei Ji a,, Lug-fei Lee a a Departmet of Ecoomics, The Ohio State Uiversity, Columbus, OH 430 USA Abstract

More information

MA Advanced Econometrics: Properties of Least Squares Estimators

MA Advanced Econometrics: Properties of Least Squares Estimators MA Advaced Ecoometrics: Properties of Least Squares Estimators Karl Whela School of Ecoomics, UCD February 5, 20 Karl Whela UCD Least Squares Estimators February 5, 20 / 5 Part I Least Squares: Some Fiite-Sample

More information

Linear regression. Daniel Hsu (COMS 4771) (y i x T i β)2 2πσ. 2 2σ 2. 1 n. (x T i β y i ) 2. 1 ˆβ arg min. β R n d

Linear regression. Daniel Hsu (COMS 4771) (y i x T i β)2 2πσ. 2 2σ 2. 1 n. (x T i β y i ) 2. 1 ˆβ arg min. β R n d Liear regressio Daiel Hsu (COMS 477) Maximum likelihood estimatio Oe of the simplest liear regressio models is the followig: (X, Y ),..., (X, Y ), (X, Y ) are iid radom pairs takig values i R d R, ad Y

More information

7.1 Convergence of sequences of random variables

7.1 Convergence of sequences of random variables Chapter 7 Limit Theorems Throughout this sectio we will assume a probability space (, F, P), i which is defied a ifiite sequece of radom variables (X ) ad a radom variable X. The fact that for every ifiite

More information

LINEAR REGRESSION ANALYSIS. MODULE IX Lecture Multicollinearity

LINEAR REGRESSION ANALYSIS. MODULE IX Lecture Multicollinearity LINEAR REGRESSION ANALYSIS MODULE IX Lecture - 9 Multicolliearity Dr Shalabh Departmet of Mathematics ad Statistics Idia Istitute of Techology Kapur Multicolliearity diagostics A importat questio that

More information

1 General linear Model Continued..

1 General linear Model Continued.. Geeral liear Model Cotiued.. We have We kow y = X + u X o radom u v N(0; I ) b = (X 0 X) X 0 y E( b ) = V ar( b ) = (X 0 X) We saw that b = (X 0 X) X 0 u so b is a liear fuctio of a ormally distributed

More information

Convergence of random variables. (telegram style notes) P.J.C. Spreij

Convergence of random variables. (telegram style notes) P.J.C. Spreij Covergece of radom variables (telegram style otes).j.c. Spreij this versio: September 6, 2005 Itroductio As we kow, radom variables are by defiitio measurable fuctios o some uderlyig measurable space

More information

Rates of Convergence by Moduli of Continuity

Rates of Convergence by Moduli of Continuity Rates of Covergece by Moduli of Cotiuity Joh Duchi: Notes for Statistics 300b March, 017 1 Itroductio I this ote, we give a presetatio showig the importace, ad relatioship betwee, the modulis of cotiuity

More information

Solution to Chapter 2 Analytical Exercises

Solution to Chapter 2 Analytical Exercises Nov. 25, 23, Revised Dec. 27, 23 Hayashi Ecoometrics Solutio to Chapter 2 Aalytical Exercises. For ay ε >, So, plim z =. O the other had, which meas that lim E(z =. 2. As show i the hit, Prob( z > ε =

More information

Linear Regression Demystified

Linear Regression Demystified Liear Regressio Demystified Liear regressio is a importat subject i statistics. I elemetary statistics courses, formulae related to liear regressio are ofte stated without derivatio. This ote iteds to

More information

Econ 325 Notes on Point Estimator and Confidence Interval 1 By Hiro Kasahara

Econ 325 Notes on Point Estimator and Confidence Interval 1 By Hiro Kasahara Poit Estimator Eco 325 Notes o Poit Estimator ad Cofidece Iterval 1 By Hiro Kasahara Parameter, Estimator, ad Estimate The ormal probability desity fuctio is fully characterized by two costats: populatio

More information

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.436J/15.085J Fall 2008 Lecture 19 11/17/2008 LAWS OF LARGE NUMBERS II THE STRONG LAW OF LARGE NUMBERS

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.436J/15.085J Fall 2008 Lecture 19 11/17/2008 LAWS OF LARGE NUMBERS II THE STRONG LAW OF LARGE NUMBERS MASSACHUSTTS INSTITUT OF TCHNOLOGY 6.436J/5.085J Fall 2008 Lecture 9 /7/2008 LAWS OF LARG NUMBRS II Cotets. The strog law of large umbers 2. The Cheroff boud TH STRONG LAW OF LARG NUMBRS While the weak

More information

The Method of Least Squares. To understand least squares fitting of data.

The Method of Least Squares. To understand least squares fitting of data. The Method of Least Squares KEY WORDS Curve fittig, least square GOAL To uderstad least squares fittig of data To uderstad the least squares solutio of icosistet systems of liear equatios 1 Motivatio Curve

More information

Econ 325/327 Notes on Sample Mean, Sample Proportion, Central Limit Theorem, Chi-square Distribution, Student s t distribution 1.

Econ 325/327 Notes on Sample Mean, Sample Proportion, Central Limit Theorem, Chi-square Distribution, Student s t distribution 1. Eco 325/327 Notes o Sample Mea, Sample Proportio, Cetral Limit Theorem, Chi-square Distributio, Studet s t distributio 1 Sample Mea By Hiro Kasahara We cosider a radom sample from a populatio. Defiitio

More information

Basics of Probability Theory (for Theory of Computation courses)

Basics of Probability Theory (for Theory of Computation courses) Basics of Probability Theory (for Theory of Computatio courses) Oded Goldreich Departmet of Computer Sciece Weizma Istitute of Sciece Rehovot, Israel. oded.goldreich@weizma.ac.il November 24, 2008 Preface.

More information

Supplementary Material to A General Method for Third-Order Bias and Variance Corrections on a Nonlinear Estimator

Supplementary Material to A General Method for Third-Order Bias and Variance Corrections on a Nonlinear Estimator Supplemetary Material to A Geeral Method for Third-Order Bias ad Variace Correctios o a Noliear Estimator Zheli Yag School of Ecoomics, Sigapore Maagemet Uiversity, 90 Stamford Road, Sigapore 178903 emails:

More information

ECON 3150/4150, Spring term Lecture 3

ECON 3150/4150, Spring term Lecture 3 Itroductio Fidig the best fit by regressio Residuals ad R-sq Regressio ad causality Summary ad ext step ECON 3150/4150, Sprig term 2014. Lecture 3 Ragar Nymoe Uiversity of Oslo 21 Jauary 2014 1 / 30 Itroductio

More information

An Introduction to Randomized Algorithms

An Introduction to Randomized Algorithms A Itroductio to Radomized Algorithms The focus of this lecture is to study a radomized algorithm for quick sort, aalyze it usig probabilistic recurrece relatios, ad also provide more geeral tools for aalysis

More information

Journal of Multivariate Analysis. Superefficient estimation of the marginals by exploiting knowledge on the copula

Journal of Multivariate Analysis. Superefficient estimation of the marginals by exploiting knowledge on the copula Joural of Multivariate Aalysis 102 (2011) 1315 1319 Cotets lists available at ScieceDirect Joural of Multivariate Aalysis joural homepage: www.elsevier.com/locate/jmva Superefficiet estimatio of the margials

More information

x iu i E(x u) 0. In order to obtain a consistent estimator of β, we find the instrumental variable z which satisfies E(z u) = 0. z iu i E(z u) = 0.

x iu i E(x u) 0. In order to obtain a consistent estimator of β, we find the instrumental variable z which satisfies E(z u) = 0. z iu i E(z u) = 0. 27 However, β MM is icosistet whe E(x u) 0, i.e., β MM = (X X) X y = β + (X X) X u = β + ( X X ) ( X u ) \ β. Note as follows: X u = x iu i E(x u) 0. I order to obtai a cosistet estimator of β, we fid

More information

EECS564 Estimation, Filtering, and Detection Hwk 2 Solns. Winter p θ (z) = (2θz + 1 θ), 0 z 1

EECS564 Estimation, Filtering, and Detection Hwk 2 Solns. Winter p θ (z) = (2θz + 1 θ), 0 z 1 EECS564 Estimatio, Filterig, ad Detectio Hwk 2 Sols. Witer 25 4. Let Z be a sigle observatio havig desity fuctio where. p (z) = (2z + ), z (a) Assumig that is a oradom parameter, fid ad plot the maximum

More information

Geometry of LS. LECTURE 3 GEOMETRY OF LS, PROPERTIES OF σ 2, PARTITIONED REGRESSION, GOODNESS OF FIT

Geometry of LS. LECTURE 3 GEOMETRY OF LS, PROPERTIES OF σ 2, PARTITIONED REGRESSION, GOODNESS OF FIT OCTOBER 7, 2016 LECTURE 3 GEOMETRY OF LS, PROPERTIES OF σ 2, PARTITIONED REGRESSION, GOODNESS OF FIT Geometry of LS We ca thik of y ad the colums of X as members of the -dimesioal Euclidea space R Oe ca

More information

Probability and Statistics

Probability and Statistics ICME Refresher Course: robability ad Statistics Staford Uiversity robability ad Statistics Luyag Che September 20, 2016 1 Basic robability Theory 11 robability Spaces A probability space is a triple (Ω,

More information

The standard deviation of the mean

The standard deviation of the mean Physics 6C Fall 20 The stadard deviatio of the mea These otes provide some clarificatio o the distictio betwee the stadard deviatio ad the stadard deviatio of the mea.. The sample mea ad variace Cosider

More information

Supplemental Material: Proofs

Supplemental Material: Proofs Proof to Theorem Supplemetal Material: Proofs Proof. Let be the miimal umber of traiig items to esure a uique solutio θ. First cosider the case. It happes if ad oly if θ ad Rak(A) d, which is a special

More information

Stochastic Matrices in a Finite Field

Stochastic Matrices in a Finite Field Stochastic Matrices i a Fiite Field Abstract: I this project we will explore the properties of stochastic matrices i both the real ad the fiite fields. We first explore what properties 2 2 stochastic matrices

More information

CEU Department of Economics Econometrics 1, Problem Set 1 - Solutions

CEU Department of Economics Econometrics 1, Problem Set 1 - Solutions CEU Departmet of Ecoomics Ecoometrics, Problem Set - Solutios Part A. Exogeeity - edogeeity The liear coditioal expectatio (CE) model has the followig form: We would like to estimate the effect of some

More information

First, note that the LS residuals are orthogonal to the regressors. X Xb X y = 0 ( normal equations ; (k 1) ) So,

First, note that the LS residuals are orthogonal to the regressors. X Xb X y = 0 ( normal equations ; (k 1) ) So, 0 2. OLS Part II The OLS residuals are orthogoal to the regressors. If the model icludes a itercept, the orthogoality of the residuals ad regressors gives rise to three results, which have limited practical

More information

Definition 4.2. (a) A sequence {x n } in a Banach space X is a basis for X if. unique scalars a n (x) such that x = n. a n (x) x n. (4.

Definition 4.2. (a) A sequence {x n } in a Banach space X is a basis for X if. unique scalars a n (x) such that x = n. a n (x) x n. (4. 4. BASES I BAACH SPACES 39 4. BASES I BAACH SPACES Sice a Baach space X is a vector space, it must possess a Hamel, or vector space, basis, i.e., a subset {x γ } γ Γ whose fiite liear spa is all of X ad

More information

Random Walks on Discrete and Continuous Circles. by Jeffrey S. Rosenthal School of Mathematics, University of Minnesota, Minneapolis, MN, U.S.A.

Random Walks on Discrete and Continuous Circles. by Jeffrey S. Rosenthal School of Mathematics, University of Minnesota, Minneapolis, MN, U.S.A. Radom Walks o Discrete ad Cotiuous Circles by Jeffrey S. Rosethal School of Mathematics, Uiversity of Miesota, Mieapolis, MN, U.S.A. 55455 (Appeared i Joural of Applied Probability 30 (1993), 780 789.)

More information

17. Joint distributions of extreme order statistics Lehmann 5.1; Ferguson 15

17. Joint distributions of extreme order statistics Lehmann 5.1; Ferguson 15 17. Joit distributios of extreme order statistics Lehma 5.1; Ferguso 15 I Example 10., we derived the asymptotic distributio of the maximum from a radom sample from a uiform distributio. We did this usig

More information

GEL estimation and tests of spatial autoregressive models

GEL estimation and tests of spatial autoregressive models GEL estimatio ad tests of spatial autoregressive models Fei Ji a ad Lug-fei Lee b a School of Ecoomics, Shaghai Uiversity of Fiace ad Ecoomics, ad Key Laboratory of Mathematical Ecoomics (SUFE), Miistry

More information

Near Unit Root in the Spatial Autoregressive Model

Near Unit Root in the Spatial Autoregressive Model Near Uit Root i the Spatial Autoregressive Model Lug-fei Lee Departmet of Ecoomics Ohio State Uiversity Columbus, OH 430, USA Email: l ee@eco.ohio-state.edu Jihai Yu Departmet of Ecoomics Uiversity of

More information

Lecture 2: Monte Carlo Simulation

Lecture 2: Monte Carlo Simulation STAT/Q SCI 43: Itroductio to Resamplig ethods Sprig 27 Istructor: Ye-Chi Che Lecture 2: ote Carlo Simulatio 2 ote Carlo Itegratio Assume we wat to evaluate the followig itegratio: e x3 dx What ca we do?

More information

6.3 Testing Series With Positive Terms

6.3 Testing Series With Positive Terms 6.3. TESTING SERIES WITH POSITIVE TERMS 307 6.3 Testig Series With Positive Terms 6.3. Review of what is kow up to ow I theory, testig a series a i for covergece amouts to fidig the i= sequece of partial

More information

10-701/ Machine Learning Mid-term Exam Solution

10-701/ Machine Learning Mid-term Exam Solution 0-70/5-78 Machie Learig Mid-term Exam Solutio Your Name: Your Adrew ID: True or False (Give oe setece explaatio) (20%). (F) For a cotiuous radom variable x ad its probability distributio fuctio p(x), it

More information

Math Solutions to homework 6

Math Solutions to homework 6 Math 175 - Solutios to homework 6 Cédric De Groote November 16, 2017 Problem 1 (8.11 i the book): Let K be a compact Hermitia operator o a Hilbert space H ad let the kerel of K be {0}. Show that there

More information

Cox-type Tests for Competing Spatial Autoregressive Models with Spatial Autoregressive Disturbances

Cox-type Tests for Competing Spatial Autoregressive Models with Spatial Autoregressive Disturbances Cox-type Tests for Competig Spatial Autoregressive Models with Spatial Autoregressive Disturbaces Fei Ji a,, Lug-fei Lee a a Departmet of Ecoomics, Ohio State Uiversity, Columbus, OH 430 USA Abstract I

More information

Rank tests and regression rank scores tests in measurement error models

Rank tests and regression rank scores tests in measurement error models Rak tests ad regressio rak scores tests i measuremet error models J. Jurečková ad A.K.Md.E. Saleh Charles Uiversity i Prague ad Carleto Uiversity i Ottawa Abstract The rak ad regressio rak score tests

More information

7.1 Convergence of sequences of random variables

7.1 Convergence of sequences of random variables Chapter 7 Limit theorems Throughout this sectio we will assume a probability space (Ω, F, P), i which is defied a ifiite sequece of radom variables (X ) ad a radom variable X. The fact that for every ifiite

More information

Determinants of order 2 and 3 were defined in Chapter 2 by the formulae (5.1)

Determinants of order 2 and 3 were defined in Chapter 2 by the formulae (5.1) 5. Determiats 5.. Itroductio 5.2. Motivatio for the Choice of Axioms for a Determiat Fuctios 5.3. A Set of Axioms for a Determiat Fuctio 5.4. The Determiat of a Diagoal Matrix 5.5. The Determiat of a Upper

More information

Notes for Lecture 11

Notes for Lecture 11 U.C. Berkeley CS78: Computatioal Complexity Hadout N Professor Luca Trevisa 3/4/008 Notes for Lecture Eigevalues, Expasio, ad Radom Walks As usual by ow, let G = (V, E) be a udirected d-regular graph with

More information

MATHEMATICAL SCIENCES PAPER-II

MATHEMATICAL SCIENCES PAPER-II MATHEMATICAL SCIENCES PAPER-II. Let {x } ad {y } be two sequeces of real umbers. Prove or disprove each of the statemets :. If {x y } coverges, ad if {y } is coverget, the {x } is coverget.. {x + y } coverges

More information

Mathematical Statistics - MS

Mathematical Statistics - MS Paper Specific Istructios. The examiatio is of hours duratio. There are a total of 60 questios carryig 00 marks. The etire paper is divided ito three sectios, A, B ad C. All sectios are compulsory. Questios

More information

b i u x i U a i j u x i u x j

b i u x i U a i j u x i u x j M ath 5 2 7 Fall 2 0 0 9 L ecture 1 9 N ov. 1 6, 2 0 0 9 ) S ecod- Order Elliptic Equatios: Weak S olutios 1. Defiitios. I this ad the followig two lectures we will study the boudary value problem Here

More information

Algebra of Least Squares

Algebra of Least Squares October 19, 2018 Algebra of Least Squares Geometry of Least Squares Recall that out data is like a table [Y X] where Y collects observatios o the depedet variable Y ad X collects observatios o the k-dimesioal

More information

Spatial Nonstationarity and Spurious Regression: The Case with Row-Normalized Spatial Weights Matrix

Spatial Nonstationarity and Spurious Regression: The Case with Row-Normalized Spatial Weights Matrix Spatial Nostatioarity ad Spurious Regressio: The Case with Row-Normalized Spatial Weights Matrix March 26, 29 Abstract This paper ivestigates the spurious regressio i the spatial settig where the regressat

More information

of the matrix is =-85, so it is not positive definite. Thus, the first

of the matrix is =-85, so it is not positive definite. Thus, the first BOSTON COLLEGE Departmet of Ecoomics EC771: Ecoometrics Sprig 4 Prof. Baum, Ms. Uysal Solutio Key for Problem Set 1 1. Are the followig quadratic forms positive for all values of x? (a) y = x 1 8x 1 x

More information

Distribution of Random Samples & Limit theorems

Distribution of Random Samples & Limit theorems STAT/MATH 395 A - PROBABILITY II UW Witer Quarter 2017 Néhémy Lim Distributio of Radom Samples & Limit theorems 1 Distributio of i.i.d. Samples Motivatig example. Assume that the goal of a study is to

More information

A RANK STATISTIC FOR NON-PARAMETRIC K-SAMPLE AND CHANGE POINT PROBLEMS

A RANK STATISTIC FOR NON-PARAMETRIC K-SAMPLE AND CHANGE POINT PROBLEMS J. Japa Statist. Soc. Vol. 41 No. 1 2011 67 73 A RANK STATISTIC FOR NON-PARAMETRIC K-SAMPLE AND CHANGE POINT PROBLEMS Yoichi Nishiyama* We cosider k-sample ad chage poit problems for idepedet data i a

More information

Lecture 3 The Lebesgue Integral

Lecture 3 The Lebesgue Integral Lecture 3: The Lebesgue Itegral 1 of 14 Course: Theory of Probability I Term: Fall 2013 Istructor: Gorda Zitkovic Lecture 3 The Lebesgue Itegral The costructio of the itegral Uless expressly specified

More information

Self-normalized deviation inequalities with application to t-statistic

Self-normalized deviation inequalities with application to t-statistic Self-ormalized deviatio iequalities with applicatio to t-statistic Xiequa Fa Ceter for Applied Mathematics, Tiaji Uiversity, 30007 Tiaji, Chia Abstract Let ξ i i 1 be a sequece of idepedet ad symmetric

More information

On the Bootstrap for Spatial Econometric Models

On the Bootstrap for Spatial Econometric Models O the Bootstrap for Spatial Ecoometric Models Fei Ji a, Lug-fei Lee a a Departmet of Ecoomics, The Ohio State Uiversity, Columbus, OH 4310 USA Abstract This paper is cocered with the use of the bootstrap

More information

NANYANG TECHNOLOGICAL UNIVERSITY SYLLABUS FOR ENTRANCE EXAMINATION FOR INTERNATIONAL STUDENTS AO-LEVEL MATHEMATICS

NANYANG TECHNOLOGICAL UNIVERSITY SYLLABUS FOR ENTRANCE EXAMINATION FOR INTERNATIONAL STUDENTS AO-LEVEL MATHEMATICS NANYANG TECHNOLOGICAL UNIVERSITY SYLLABUS FOR ENTRANCE EXAMINATION FOR INTERNATIONAL STUDENTS AO-LEVEL MATHEMATICS STRUCTURE OF EXAMINATION PAPER. There will be oe 2-hour paper cosistig of 4 questios.

More information

MATH 320: Probability and Statistics 9. Estimation and Testing of Parameters. Readings: Pruim, Chapter 4

MATH 320: Probability and Statistics 9. Estimation and Testing of Parameters. Readings: Pruim, Chapter 4 MATH 30: Probability ad Statistics 9. Estimatio ad Testig of Parameters Estimatio ad Testig of Parameters We have bee dealig situatios i which we have full kowledge of the distributio of a radom variable.

More information

Investigating the Significance of a Correlation Coefficient using Jackknife Estimates

Investigating the Significance of a Correlation Coefficient using Jackknife Estimates Iteratioal Joural of Scieces: Basic ad Applied Research (IJSBAR) ISSN 2307-4531 (Prit & Olie) http://gssrr.org/idex.php?joural=jouralofbasicadapplied ---------------------------------------------------------------------------------------------------------------------------

More information

Session 5. (1) Principal component analysis and Karhunen-Loève transformation

Session 5. (1) Principal component analysis and Karhunen-Loève transformation 200 Autum semester Patter Iformatio Processig Topic 2 Image compressio by orthogoal trasformatio Sessio 5 () Pricipal compoet aalysis ad Karhue-Loève trasformatio Topic 2 of this course explais the image

More information

Matrix Algebra from a Statistician s Perspective BIOS 524/ Scalar multiple: ka

Matrix Algebra from a Statistician s Perspective BIOS 524/ Scalar multiple: ka Matrix Algebra from a Statisticia s Perspective BIOS 524/546. Matrices... Basic Termiology a a A = ( aij ) deotes a m matrix of values. Whe =, this is a am a m colum vector. Whe m= this is a row vector..2.

More information

On Random Line Segments in the Unit Square

On Random Line Segments in the Unit Square O Radom Lie Segmets i the Uit Square Thomas A. Courtade Departmet of Electrical Egieerig Uiversity of Califoria Los Ageles, Califoria 90095 Email: tacourta@ee.ucla.edu I. INTRODUCTION Let Q = [0, 1] [0,

More information

Machine Learning Theory Tübingen University, WS 2016/2017 Lecture 12

Machine Learning Theory Tübingen University, WS 2016/2017 Lecture 12 Machie Learig Theory Tübige Uiversity, WS 06/07 Lecture Tolstikhi Ilya Abstract I this lecture we derive risk bouds for kerel methods. We will start by showig that Soft Margi kerel SVM correspods to miimizig

More information

Singular Continuous Measures by Michael Pejic 5/14/10

Singular Continuous Measures by Michael Pejic 5/14/10 Sigular Cotiuous Measures by Michael Peic 5/4/0 Prelimiaries Give a set X, a σ-algebra o X is a collectio of subsets of X that cotais X ad ad is closed uder complemetatio ad coutable uios hece, coutable

More information

Bull. Korean Math. Soc. 36 (1999), No. 3, pp. 451{457 THE STRONG CONSISTENCY OF NONLINEAR REGRESSION QUANTILES ESTIMATORS Seung Hoe Choi and Hae Kyung

Bull. Korean Math. Soc. 36 (1999), No. 3, pp. 451{457 THE STRONG CONSISTENCY OF NONLINEAR REGRESSION QUANTILES ESTIMATORS Seung Hoe Choi and Hae Kyung Bull. Korea Math. Soc. 36 (999), No. 3, pp. 45{457 THE STRONG CONSISTENCY OF NONLINEAR REGRESSION QUANTILES ESTIMATORS Abstract. This paper provides suciet coditios which esure the strog cosistecy of regressio

More information

Lecture Note 8 Point Estimators and Point Estimation Methods. MIT Spring 2006 Herman Bennett

Lecture Note 8 Point Estimators and Point Estimation Methods. MIT Spring 2006 Herman Bennett Lecture Note 8 Poit Estimators ad Poit Estimatio Methods MIT 14.30 Sprig 2006 Herma Beett Give a parameter with ukow value, the goal of poit estimatio is to use a sample to compute a umber that represets

More information

Introduction to Optimization Techniques. How to Solve Equations

Introduction to Optimization Techniques. How to Solve Equations Itroductio to Optimizatio Techiques How to Solve Equatios Iterative Methods of Optimizatio Iterative methods of optimizatio Solutio of the oliear equatios resultig form a optimizatio problem is usually

More information

Kolmogorov-Smirnov type Tests for Local Gaussianity in High-Frequency Data

Kolmogorov-Smirnov type Tests for Local Gaussianity in High-Frequency Data Proceedigs 59th ISI World Statistics Cogress, 5-30 August 013, Hog Kog (Sessio STS046) p.09 Kolmogorov-Smirov type Tests for Local Gaussiaity i High-Frequecy Data George Tauche, Duke Uiversity Viktor Todorov,

More information

Statistical Inference (Chapter 10) Statistical inference = learn about a population based on the information provided by a sample.

Statistical Inference (Chapter 10) Statistical inference = learn about a population based on the information provided by a sample. Statistical Iferece (Chapter 10) Statistical iferece = lear about a populatio based o the iformatio provided by a sample. Populatio: The set of all values of a radom variable X of iterest. Characterized

More information

On the Bootstrap for Spatial Econometric Models

On the Bootstrap for Spatial Econometric Models O the Bootstrap for Spatial Ecoometric Models Fei Ji a, Lug-fei Lee a a Departmet of Ecoomics, Ohio State Uiversity, Columbus, OH 430 USA Abstract This paper is cocered about the use of the bootstrap for

More information

Estimating a spatial autoregressive model with an endogenous spatial weight matrix

Estimating a spatial autoregressive model with an endogenous spatial weight matrix Estimatig a spatial autoregressive model with a edogeous spatial weight matrix Xi Qu, Lug-fei Lee The Ohio State Uiversity October 9, Abstract The spatial autoregressive model (SAR) is a stadard tool to

More information

Ω ). Then the following inequality takes place:

Ω ). Then the following inequality takes place: Lecture 8 Lemma 5. Let f : R R be a cotiuously differetiable covex fuctio. Choose a costat δ > ad cosider the subset Ωδ = { R f δ } R. Let Ωδ ad assume that f < δ, i.e., is ot o the boudary of f = δ, i.e.,

More information

Lecture 22: Review for Exam 2. 1 Basic Model Assumptions (without Gaussian Noise)

Lecture 22: Review for Exam 2. 1 Basic Model Assumptions (without Gaussian Noise) Lecture 22: Review for Exam 2 Basic Model Assumptios (without Gaussia Noise) We model oe cotiuous respose variable Y, as a liear fuctio of p umerical predictors, plus oise: Y = β 0 + β X +... β p X p +

More information

[412] A TEST FOR HOMOGENEITY OF THE MARGINAL DISTRIBUTIONS IN A TWO-WAY CLASSIFICATION

[412] A TEST FOR HOMOGENEITY OF THE MARGINAL DISTRIBUTIONS IN A TWO-WAY CLASSIFICATION [412] A TEST FOR HOMOGENEITY OF THE MARGINAL DISTRIBUTIONS IN A TWO-WAY CLASSIFICATION BY ALAN STUART Divisio of Research Techiques, Lodo School of Ecoomics 1. INTRODUCTION There are several circumstaces

More information

CEE 522 Autumn Uncertainty Concepts for Geotechnical Engineering

CEE 522 Autumn Uncertainty Concepts for Geotechnical Engineering CEE 5 Autum 005 Ucertaity Cocepts for Geotechical Egieerig Basic Termiology Set A set is a collectio of (mutually exclusive) objects or evets. The sample space is the (collectively exhaustive) collectio

More information

Topic 9: Sampling Distributions of Estimators

Topic 9: Sampling Distributions of Estimators Topic 9: Samplig Distributios of Estimators Course 003, 2016 Page 0 Samplig distributios of estimators Sice our estimators are statistics (particular fuctios of radom variables), their distributio ca be

More information

G. R. Pasha Department of Statistics Bahauddin Zakariya University Multan, Pakistan

G. R. Pasha Department of Statistics Bahauddin Zakariya University Multan, Pakistan Deviatio of the Variaces of Classical Estimators ad Negative Iteger Momet Estimator from Miimum Variace Boud with Referece to Maxwell Distributio G. R. Pasha Departmet of Statistics Bahauddi Zakariya Uiversity

More information

A statistical method to determine sample size to estimate characteristic value of soil parameters

A statistical method to determine sample size to estimate characteristic value of soil parameters A statistical method to determie sample size to estimate characteristic value of soil parameters Y. Hojo, B. Setiawa 2 ad M. Suzuki 3 Abstract Sample size is a importat factor to be cosidered i determiig

More information

A Hadamard-type lower bound for symmetric diagonally dominant positive matrices

A Hadamard-type lower bound for symmetric diagonally dominant positive matrices A Hadamard-type lower boud for symmetric diagoally domiat positive matrices Christopher J. Hillar, Adre Wibisoo Uiversity of Califoria, Berkeley Jauary 7, 205 Abstract We prove a ew lower-boud form of

More information

Lecture 3. Properties of Summary Statistics: Sampling Distribution

Lecture 3. Properties of Summary Statistics: Sampling Distribution Lecture 3 Properties of Summary Statistics: Samplig Distributio Mai Theme How ca we use math to justify that our umerical summaries from the sample are good summaries of the populatio? Lecture Summary

More information