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

Size: px
Start display at page:

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

Transcription

1 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 Ohio State Uiversity By Fei Ji, M.A. Graduate Program i Ecoomics The Ohio State Uiversity 203 Dissertatio Committee Lug-fei Lee, Advisor Stephe Cosslett Robert de Jog

2 Copyright by Fei Ji 203

3 Abstract This dissertatio cosists of three chapters coverig the followig topics i spatial ecoometrics: estimatio, specificatio ad the bootstrap. I Chapter, we first geeralize a approximate measure of spatial depedece, the AP LE statistic (Li et al., 2007), to a spatial Durbi (SD) model. This geeralized AP LE takes ito accout exogeous variables directly ad ca be used to detect spatial depedece origiatig from either a spatial autoregressive (SAR), spatial error (SE) or SD process. However, that measure is ot cosistet. Secodly, by examiig carefully the first order coditio of the cocetrated log likelihood of the SD (or SAR) model, whose first order approximatio geerates the AP LE, we costruct a momet equatio quadratic i the autoregressive parameter that geeralizes a origial estimatio approach i Ord (975) ad yields a closed-form cosistet root estimator of the autoregressive parameter. With a specific momet equatio costructed from a iitial cosistet estimator, the root estimator ca be as efficiet as the MLE uder ormality. Furthermore, whe there is ukow heteroskedasticity i the disturbaces, we derive a modified AP LE ad a root estimator which ca be robust to ukow heteroskedasticity. The root estimators are computatioally much simpler tha the quasi-maximum likelihood estimators. I Chapter 2, we cosider the Cox-type tests of o-ested hypotheses for spatial autoregressive (SAR) models with SAR disturbaces. We formally derive the asymptotic distributios of the test statistics. I cotrast to regressio models, we show that the Coxtype ad J-type tests for o-ested hypotheses i the framework of SAR models are ot asymptotically equivalet uder the ull hypothesis. The Cox test i o-spatial settig has bee foud ofte to have large size distortio, which ca be removed by the bootstrap. Cox-type tests for SAR models with SAR disturbaces may also have large size distortio. We show that the bootstrap is cosistet for Cox-type tests i our framework. Performaces ii

4 of the Cox-type ad J-type tests as well as their bootstrapped versios i fiite samples are compared via a Mote Carlo study. These tests are of particular iterest whe there are competig models with differet spatial weights matrices. Usig bootstrapped p-values, the Cox tests have relatively high power i all experimets ad ca outperform J-type ad several other related tests i some cases. Chapter 3 is cocered with the use of the bootstrap for spatial ecoometric models. We show that the bootstrap for spatial ecoometric models ca be studied based o liear-quadratic (LQ) forms of disturbaces. By provig the uiform covergece of the cumulative distributio fuctio for LQ forms to that of a ormal distributio, we show that the bootstrap is geerally cosistet for test statistics that ca be approximated by LQ forms, which iclude Mora s I, Cox-type ad spatial J-type test statistics. Possible asymptotic refiemets of the bootstrap for spatial ecoometric models may be studied based o some asymptotic expasios for LQ forms. We discuss two cases: whe the disturbaces are ormal, we directly show the existece of Edgeworth expasios for LQ forms ad apply the result to show that the bootstrap for Mora s I ca provide asymptotic refiemets; whe the disturbaces are ot ormal, we show the existece of a oe-term asymptotic expasio of LQ forms based o martigales, which sheds light o the secod-order correctess of the bootstrap for LQ forms. iii

5 Dedicatio Dedicated to my wife, Yuqi Wag, ad my parets, Chagrog Ji ad Guifag Ya iv

6 Ackowledgemets I would like to express my deepest gratitude to my advisor, Lug-fei Lee, for his ivaluable guidace, ecouragemet ad geerous help i all aspects of my life. He is always willig to sped his precious time with me o research issues ad givig suggestios o other matters. His great ethusiasm ad rigorous attitudes towards research are life-log wealth for me. I am grateful to my dissertatio committee members, Professor Robert de Jog ad Professor Stephe Cosslett, for their valuable commets ad ecouragemet. I would also like to thak Professor Jaso Blevis for servig o my third year paper committee ad givig valuable suggestios. I thak Professor Lucia Du for her solicitude ad helpful advice ad Professor Daeho Kim for his woderful suggestios o my job market paper presetatio. May fellow classmates ad frieds have bee helpig ad supportig me, which I wish to thak here. My special thaks go to Feg Guo, Ya Bao ad Feg Zhou. Last but ot the least, I would like to thak my wife, Yuqi Wag, ad my parets, Chagrog Ji ad Guifag Ya, for their love, support ad uderstadig. v

7 Vita Jue B.S., Ecoomics, Huazhog Uiversity of Sciece ad Techology December M.A., Ecoomics, The Ohio State Uiversity 2008 Preset Graduate Teachig Associate, The Ohio State Uiversity Fields of Study Major Field: Ecoomics vi

8 Table of Cotets Abstract ii Dedicatio iv Ackowledgemets v Vita vi List of Tables x List of Figures xi Chapter : Approximated Likelihood ad Root Estimators for Spatial Iteractio i Spatial Autoregressive Models Itroductio A Root Estimator A Root Estimator: Homoskedastic Case A Root Estimator: Heteroskedastic Case Mote Carlo Study Coclusio Chapter 2: Cox-type Tests for Competig Spatial Autoregressive Models with Spatial Autoregressive Disturbaces Itroductio Cox-type Tests Relatioship ad Compariso betwee the Cox-type ad J-type Tests Mote Carlo Study Empirical Illustratio Coclusio Appedix 2. Notatios ad Expressios vii

9 Appedix 2.2 The Exteded Wald ad Exteded Score Tests Appedix 2.2. The Exteded Wald Test Appedix The Exteded Score Test Chapter 3: O the Bootstrap for Spatial Ecoometric Models Itroductio Statistics i Spatial Ecoometrics ad LQ Forms Cosistecy of the Bootstrap Mora s I Spatial J-type Tests Cox-type Tests Asymptotic Refiemets Normal Disturbaces No-ormal Disturbaces Coclusio Appedix 3. Assumptios Appedix 3.. Assumptios for Mora s I Appedix 3..2 Assumptios for the Spatial J Tests: J Appedix 3..3 Assumptios for the Spatial J Tests: J Refereces Appedix A: Lemmas Appedix A.: Geeral Lemmas Appedix A.2: Lemmas for Chapter Appedix A.3: Lemmas for Chapter Appedix A.3.: Lemmas for the SARAR Model Appedix A.3.2 Lemmas for the Spatial J Tests: J Appedix A.3.3 Lemmas for the Spatial J Tests: J Appedix A.3.4 Lemmas for Cox-type Tests Appedix B: Proofs Appedix B.: Proofs for Chapter Appedix B.2: Proofs for Chapter viii

10 Appedix B.3: Proofs for Chapter ix

11 List of Tables Table. Compariso of the computig time, bias, STD ad RMSE of the QMLE, RE sar ad AP LE sar whe the sample size is large Table 2. Sets of Experimets Table 3. Empirical size ad power for experimet set I with ormal disturbaces ad = Table 4. Empirical size ad power for experimet set I with chi-square disturbaces ad = Table 5. Empirical size ad power for experimet set II with ormal disturbaces ad = Table 6. Empirical size ad power for experimet set II with chi-square disturbaces ad = Table 7. Empirical size ad power for experimet set III with ormal disturbaces ad = Table 8. Empirical size ad power for experimet set III with chi-square disturbaces ad = Table 9. Empirical size ad power computed usig asymptotic p-values for experimet set I with = Table 0. Empirical size ad power computed usig asymptotic p-values for experimet set II with = Table. Empirical size ad power computed usig asymptotic p-values for experimet set III with = Table 2. Testig results with a housig data set (Whether H 0 is rejected or ot) x

12 List of Figures Figure. Compariso of the bias, STD ad RMSE of the QMLE, RE sar, AP LE sar, ACME sar, RE sd, AP LE sd ad ACME sd whe the DGP is the SAR model uder homoskedasticity Figure 2. Compariso of the bias, STD ad RMSE of the QMLE, RE sd, AP LE se, AP LE sd ad ACME sd whe the DGP is the SE model uder homoskedasticity Figure 3. Compariso of the bias, STD ad RMSE of the QMLE, RE sar, AP LE sar, ACME sar, RE sd, AP LE sd ad ACME sd whe the DGP is the SAR model uder heteroskedasticity (HD-) Figure 4. Compariso of the bias, STD ad RMSE of the QMLE, RE sar, AP LE sar, ACME sar, RE sd, AP LE sd ad ACME sd whe the DGP is the SAR model uder heteroskedasticity (HD-2) Figure 5. Compariso of the bias, STD ad RMSE of the the QMLE, RE sd, AP LE se, AP LE sd ad ACME sd whe the DGP is the SE model uder heteroskedasticity (HD-) 2 Figure 6. Compariso of the bias, STD ad RMSE of the QMLE, RE sd, AP LE se, AP LE sd ad ACME sd whe the DGP is the SE model uder heteroskedasticity (HD-2) xi

13 Chapter : Approximated Likelihood ad Root Estimators for Spatial Iteractio i Spatial Autoregressive Models. Itroductio Li et al. (2007) propose a closed-form measure of spatial depedece, a approximate profilelikelihood estimator (AP LE), based o a pure spatial autoregressive (SAR) model. Their Mote Carlo experimets for spatial weights matrices defied accordig to a secod-order eighborhood structure o toroidal lattices show that the AP LE provides a better assessmet of the stregth of spatial depedece for data geerated by the pure SAR model tha alterative measures such as Mora s I (Mora, 950). Thus, the AP LE provides a better measure of spatial depedece tha Mora s I for exploratory aalyses. It has bee show i Martellosio (200) that Mora s I has zero power to detect spatial correlatio i a SAR model whe the autoregressive coefficiet is large ad close to oe. Li et al. (202) geeralize the AP LE statistic to the spatial error (SE) model to accout for exogeous variables. As both the SAR ad SE models are costraied forms of the more geeral spatial Durbi (SD) model, a approximate measure for spatial depedece of iterest should accout for exogeous variables directly ad provide a good approximatio to the autoregressive parameter i the SD model. This approximate measure ca be used to detect spatial depedece origiatig from either the SAR, SE or SD model. I this chapter, we exted the AP LE to the SD model, which, similarly to Li et al. (2007), is based o a first order approximatio to the first order coditio of the cocetrated log likelihood of the SD model. The origial AP LE as well as the exteded AP LE from the first order approximatios are ot cosistet for the autoregressive parameter due to a systematic bias. Higher order approximatios of the first order coditio may geerate more accurate measures of the autoregressive parameter, but they ivolve multiple roots ad geerally do ot yield closed-form solutios.

14 Treatig the first order coditio differetly, we obtai a momet equatio quadratic i the autoregressive parameter that geerates a closed-form root estimator. Our proposed root estimator geeralizes a estimator origiated i Ord (975) i a geeral settig. For the quadratic momet equatio, coditios uder which oe of the roots is cosistet will be specified. With a iitial cosistet estimator, a momet equatio ca be desiged to geerate a secod step root estimator which is asymptotically as efficiet as the maximum likelihood estimator (MLE) uder ormality. Oce a estimate of the autoregressive parameter is available, other parameters i a SD model may be estimated by least squares (LS) after applyig a spatial filter to the data o the depedet variable. The modified AP LE ad the root estimator ca be used as measures of spatial depedece or simple estimators for the autoregressive parameter i a SD or SAR model, as the SD model ests the SAR model. The proposed estimators ca further be exteded to possess some robust properties. The origial AP LE i Li et al. (2007) has ot accouted for possible heteroskedasticity i the disturbaces. Li et al. (202) argue that a valid trasformatio ca be applied to the SE model, so the exteded AP LE may be calculated with the trasformed data. This is so whe the heteroskedastic variace has a kow fuctioal form. However, if we do ot kow the form of heteroskedasticity, the data could ot be properly trasformed. A misspecified trasformatio ca lead to errors i iferece. With ukow heteroskedasticity, we may adjust the first order coditio to derive a modified AP LE statistic, which we call a approximate cocetrated momet estimator (ACME), ad we ca also adjust the momet equatio to derive a root estimator that is robust to ukow heteroskedasticity. Existig estimatio methods for SAR (SD) models do ot have a closed form ad are usually computatioally ivolved. 2 The MLE or quasi-maximum likelihood estimator (QM- LE) does ot have a closed form (Aseli, 988; Lee, 2004a). The computatio ivolves the evaluatio of the log-determiat of a square matrix with dimesio equal to the sample size at differet parameter values, so it might be computatioally demadig whe the I this situatio, we ca also easily modify our root estimates to accommodate the kow heteroskedasticity because after proper trasformatio, it results i a SAR model with homogeous disturbaces. 2 Because of the correlatio of the spatially lagged depedet variable with disturbaces, the LS estimator is oly cosistet for a subclass of models (Lee, 2002). 2

15 sample size is large. 3 Some empirical applicatios may create large matrices, for example, the US Cesus Bureau collects data at over 250,000 cesus block group locatios ad the Home Mortgage Disclosure Act data have over 00 millio observatios. Because of the computatioal burde of the MLE, eve with sample sizes that might ot be too large, researchers may tur to less efficiet estimatio methods such as the two stage least squares (2SLS) proposed by Kelejia ad Prucha (998). 4 For example, Helms (202) uses the 2SLS estimatio whe the sample size is 6,638. Lee (2007a) cosiders the geeralized method of momets (GMM) estimatio, which combies the quadratic momets that capture the correlatio across the spatial uits with the liear momets used i the 2SLS approach. Compared to the QMLE, the GMM estimator is computatioally simpler ad it ca be as efficiet as the MLE uder ormality. 5 Lee (2007b) proposes a computatioally simpler GMM for the estimatio of SAR models. The method reduces the GMM estimatio of a vector of parameters ito oliear estimatio of oly the autoregressive parameter. It ca reduce the computatioal burde substatially ad it may be as efficiet as the joit GMM estimator uder certai coditios. But it still does ot geerate a closed-form solutio ad searchig over a parameter space is ecessary. Eve though the GMM avoids computig log-determiats of matrices, searchig over a parameter space with large matrices ivolved 3 Various techiques ad simplificatios have bee proposed to tackle this problem, see, for example, Marti (993), Griffith ad Soe (995), Pace (997), Pace ad Barry (997a,b), Barry ad Pace (999), Griffith (2000), Smirov ad Aseli (200), Pace ad LeSage (2004), Pace ad LeSage (2009) ad Smirov ad Aseli (2009). Eve with these techiques ad simplificatios, the computatio ca be still timecosumig. Alterative simplificatios ofte lead to less accurate estimates. We ote that Pace ad LeSage (2009) propose a samplig approach to estimate the log determiat. Their Mote Carlo study shows that the approach ca be very fast i estimatig the log determiat. Give that the log determiat eeds to be evaluated may times at differet parameter values, the actual time of computig a MLE or QMLE may be much loger. 4 Their model is more geeral oe with both a spatial lag of the depedet variable ad a SAR process i the disturbaces. While the autoregressive parameter for the spatial lag of the depedet variable is estimated by 2SLS, the autoregressive parameter i the disturbace process is estimated by GMM with three momet equatios. 5 Liu et al. (200) ad Lee ad Liu (200) cosider the efficiet GMM estimatio of the regular ad high order SAR models with properly modified momet equatios. Their estimator is as efficiet as the MLE uder ormality ad is more efficiet tha the QMLE otherwise. 3

16 could still be computatioally itesive. Our root estimator is asymptotically as efficiet as the MLE uder ormality sice the desiged secod step momet equatio automatically combies the liear ad quadratic momet coditios i a efficiet way. 6 For SAR models with ukow heteroskedasticity, Li ad Lee (200) study the GMM estimatio where liear ad quadratic momet equatios ivolvig both the autoregressive parameter ad parameters for other exogeous variables are used. 7 Our (robust) root estimator is obtaied with a properly modified ad combied momet equatio quadratic i the autoregressive parameter. Thus, for the closed-form root estimator (see Eq. (.24)), o searchig over a parameter space is eeded. Because of the closed form, the root estimator requires little programmig effort. Our Mote Carlo study shows that the root estimator has similar fiite sample performace as the QMLE uder ormality ad the robust root estimator performs well uder ukow heteroskedasticity. Computig the root estimates oly takes slightly loger time tha computig the AP LE, which is much faster tha computig the QMLE. As the computatioal burde of both the modified AP LE ad the root estimate is miimal, they ca be applied to SAR, SE or SD models for huge data sets. The rest of the chapter is orgaized as follows. Sectio.2 itroduces related models ad develops the AP LE ad ACME; Sectio.3 establishes the cosistecy ad asymptotic distributio of our root estimators i both the homoskedastic ad heteroskedastic cases; Sectio.4 presets some Mote Carlo results; Sectio.5 cocludes. Some lemmas ad proofs are collected i the appedices. 6 Both the modified GMM ad our root estimator reduce the estimatio to that of oly the autoregressive parameter, which might lead to better fiite sample performace whe a bias correctio might be costructed ad applied to this sigle estimate, compared to the case whe a complete vector of parameters are estimated joitly ad the the bias correctio is applied to this vector of estimates. 7 Kelejia ad Prucha (200) also cosider the specificatio ad estimatio of the SAR model with SAR disturbaces that has heteroskedastic iovatios. As i Kelejia ad Prucha (998), the autoregressive parameter for the spatial lagged depedet variable is estimated by 2SLS ad the autoregressive parameter i the disturbace process is estimated by GMM with multiple momet equatios. 4

17 .2 The Models, AP LE ad ACME I this sectio, we itroduce the related models, ad the derive the AP LE for the SD model whe ɛ i s are i.i.d., ad the ACME whe ɛ i s may be oly idepedet but with differet ad ukow variaces. A SAR model is specified as y = ρw y + X β + ɛ, (.) where is the sample size, y is a -dimesioal vector of observatios, W is a spatial weights matrix with a zero diagoal, X is a k matrix of exogeous variables, ɛ = (ɛ,..., ɛ ) with ɛ i s beig idepedet with mea zero, ad ρ is a autoregressive parameter. If the spatial depedece is i the disturbaces istead, we have a SE model which is y = X β + u, u = ρw u + ɛ. (.2) Let I deote the -dimesioal idetity matrix. Pre-multiplyig both sides of Eq. (.2) by (I ρw ) yields y = ρw y + X β + W X ( ρβ) + ɛ, (.3) which is a costraied form of the SD model 8 y = ρw y + X β + W X γ + ɛ. (.4) That is, γ i the SD model (.4) is required to be equal to mius ρ times β for the SE process. A regressio model with the SAR process is just the SD model (.4) with γ = 0, so it is also a costraied form of the SD model. Without the costraits, the SD model may also have a iterest of its ow. The W X as regressors may capture exterality arisig form eighbors characteristics (see, e.g., LeSage ad Pace 2009, p. 30). If W is row-ormalized ad X cotais a itercept term, i.e., X = [l, X ], where l is a -dimesioal colum vector of oes ad X is a (k ) matrix, the W X will geerate a colum vector of oes as W X = [l, W X ]. Coefficiets o these two colum vectors of oes should be collected together. If W is ot row-ormalized, the colums of 8 We use this termiology followig LeSage ad Pace (2009). 5

18 X ad W X are i geeral liearly idepedet. I this case, for l i X, W l is the vector of row sums which is a extra regressor. 9 To make later arrative easier, we write the SD model as y = ρw y + Z θ + ɛ, (.5) where Z = [X, W X ] or Z = [X, W X ], depedig o whether both X ad W X cotai a colum vector of oes or ot, ad θ is the correspodig vector of coefficiets. The Z is d with d = 2k or d = 2k. The AP LE ad ACME are derived for the SD model (.5). Our root estimators are also stated with the settig of Eq. (.5). Whe a SAR model rather tha a more geeral SD model is cosidered, just take Z to be X. Let the true parameters of ρ ad θ be ρ 0 ad θ 0. Whe ɛ i s are i.i.d. (0, σ2 ), the true parameter for σ 2 is σ 2 0 ; whe there is ukow heteroskedasticity, E(ɛ ɛ ) = Diag(σ 2,..., σ2 ) = Σ, where Diag(a ) deote a diagoal matrix with the diagoal elemets beig those of the vector a. Let S (ρ) = I ρw ad G (ρ) = W S (ρ). Deote S = S (ρ 0 ) ad G = G (ρ 0 ) for short. Whe ɛ i s are i.i.d. with variace σ 2, the log likelihood fuctio for the model (.5) is L (ρ, θ, σ 2 ) = 2 l(2πσ2 ) + l S (ρ) 2σ 2 [S (ρ)y Z θ] [S (ρ)y Z θ]. Maximizig the fuctio with a fixed ρ, we obtai the QMLEs for θ ad σ 2 as: ˆθ = (Z Z ) Z S (ρ)y, (.6) ˆσ 2 = y S (ρ)m Z S (ρ)y, (.7) where M Z = I Z (Z Z ) Z. Eqs. (.6) ad (.7) are just like the LS estimators after the spatial filter S (ρ) has bee applied to y. Substitutig these expressios ito the log likelihood fuctio, we have the cocetrated log (or profile) likelihood fuctio of ρ: L (ρ) = 2 [l(2π/) + ] + l S (ρ) 2 l[y S (ρ)m Z S (ρ)y ]. The first order coditio for the maximizatio of the cocetrated log likelihood fuctio is: y S (ρ)m Z W y y S (ρ)m Z S (ρ)y tr[g (ρ)] = 0, (.8) 9 If elemets i W are 0 or as i a etwork, a row sum refers to a outdegree. So i such a case, the outdegrees of idividuals form a explaatory variable. 6

19 where tr(a ) deotes the trace of a square matrix A. Multiplyig both sides by y S (ρ)m Z S (ρ)y yields y S (ρ)m Z W y y S (ρ)m Z S (ρ)y tr[g (ρ)] = 0. (.9) Similar to the derivatio of the AP LE i Li et al. (2007), a approximate measure of ρ ca be obtaied from a first order approximatio of the left-had side of Eq. (.9). Note that tr[g (ρ)] = tr[w (I +ρw +... )] ρ tr(w 2 ) as W has a zero diagoal, the approximatio yields y APLE sd = M Z W y y W M Z W y + y M tr(w Z y 2), (.0) For the coveiece of later referece, we also write dow the AP LE for the SAR model as y APLE sar = M X W y y W M X W y + y M tr(w X y 2), (.) where M X = I X (X X ) X. Furthermore, for a pure SAR process, M Z = I. As y W y = y [(W +W )/2]y ad tr(w 2 ) = λ λ, where λ is the vector of W s eigevalues, Eq. (.0) would reduce to which is that give i Li et al. (2007). y [(W + W )/2]y, y W W y + y y λ λ Usig the same approach, Li et al. (202) derive the AP LE for the SE model as APLE se = y M X [(W + W )/2]M X y A, (.2) where A = y M X W W M X y y M X (W + W )(I M X )(W + W )M X y + y M tr(w X y ) 2. The AP LE for the SAR model i Eq. (.) ad that for the SE model i Eq. (.2) have differet forms. Alteratively, we could use the AP LE based the SD model i Eq. (.0) as a approximate measure of spatial depedece origiatig from either the SAR, SE or SD model. Eq. (.9) ca be rewritte as [ y S (ρ) G (ρ) tr[g (ρ)] I ]M Z S (ρ)y = 0. (.3) Whe there is ukow heteroskedasticity, the expectatio of the left-had side of the above equatio over at the true parameters ρ 0, θ 0, σ 2,..., σ2 does ot coverge to zero 7

20 i geeral, sice { [ E y S = { E G tr(g ) (Z θ 0 + ɛ ) [ G tr(g ) I ] M Z S y } = {[ tr G tr(g ] } ) I M Z Σ = {[ tr G tr(g ] } ) I Σ + o() = [ G tr(g ) I ]ii σ2 i + o(), i= } I ]M Z (Z θ 0 + ɛ ) (.4) by Lemma i Appedix C. Uder ukow heteroskedasticity, we may modify Eq. (.3) ito the followig equatio y S (ρ) [ G (ρ) Diag[G (ρ)] ] M Z S (ρ)y = 0, (.5) which is a valid momet equatio because the zero diagoal of G (ρ) Diag[G (ρ)] implies that the expectatio of the left-had side of the equatio over at ρ 0 coverges to zero. Takig a first order Taylor expasio of the left-had side of Eq. (.5) with ρ ad settig it to zero yield a modified AP LE statistic, which we call ACME: ACME sd = For the SAR model, the ACME is ACME sar = For a pure SAR process, Eq. (.6) simplifies to y M Z W y y W M Z W y + y Diag(W 2 )M Z y. (.6) y M X W y y W M X W y + y Diag(W 2 )M X y. (.7) y W y y W W y + y Diag(W 2 )y. (.8) Eqs. (.6) (.8) ca be used as approximate measures of ρ whe ukow heteroskedasticity exists. Eqs. (.0) ad (.6) (or Eqs. (.) ad (.7)) oly differ i the secod terms of their deomiators..3 A Root Estimator.3. A Root Estimator: Homoskedastic Case Eq. (.3) also motivates a exteded GMM root estimator for ρ of the SD model whe ɛ i s are i.i.d.. The matrix G (ρ) I tr[g (ρ)]/ i Eq. (.3) has a zero trace. Not 8

21 accoutig for the ρ s i the matrix, Eq. (.3) is quadratic i ρ. Replacig the matrix with ay costat matrix P satisfyig tr(p M Z ) = 0 (or tr(p ) = 0) 0, a cosistet GMM root estimator ca be derived by solvig the equatio g (ρ) = y S (ρ)p M Z S (ρ)y = 0, (.9) because the expectatio of g (ρ 0 ) is zero: E[g (ρ 0 )] = E[(Z θ 0 + ɛ ) P M Z (Z θ 0 + ɛ )] = σ0 2 tr(p M Z ) = 0. (.20) The P = G I tr(g )/ or P = G I tr(g M Z )/ is expected to geerate a root estimator that is asymptotically as efficiet as the MLE uder ormality sice Eq. (.9) with P = G I tr(g )/ is essetially the first order coditio of the cocetrated log likelihood fuctio Eq. (.3), eve though there is a sigle momet equatio. The form of the momet equatio automatically combies the liear ad quadratic momets i a way such that the root estimator ca be efficiet uder ormality, ulike Lee (2007a) or Lee (2007b), where liear momets are used together with the quadratic momets as a system with optimum weightig by the iverse of their variace-covariace matrix. This is ot surprisig because the sigle momet equatio is motivated from the first order coditio of the cocetrated log likelihood fuctio. Oce a cosistet estimator of ρ is available, Eqs. (.6) ad (.7) ca be used to calculate estimates for β ad σ 2, respectively. To establish the cosistecy of the root estimator, the followig regularity coditios are assumed. Assumptio.. ɛ i s i ɛ = (ɛ,..., ɛ ) are i.i.d. (0, σ0 2 ) ad the momet E( ɛ4+η i ) exists for some η > 0. 0 We still have a cosistet estimator if tr(p ) = 0 istead of tr(p M Z ) = 0. This is so because for the expectatio of the left-had side of Eq. (.9) at ρ 0, the additioal term divided by is σ2 0 tr[p Z (Z Z ) Z ] = σ2 0 tr[z P Z (Z Z ) ] = O( d ), which is ot exactly zero but coverges to zero as goes to ifiity. However, usig a matrix P such that tr(p M Z ) = 0 might have better small sample properties. Give ay matrix A, such a P matrix ca be costructed as P = A tr(am Z ) d I. For a pure SAR process, Eq. (.9) will be reduced to y S (ρ)p S (ρ)y = 0. I Ord (975), based o the motivatio of a modified LS estimatio, he cosidered the quadratic momet y S (ρ)w S (ρ)y = 0, but dismissed it i favor of the MLE i terms of efficiecy. The Eq. (.9) with a class of P provides a geeral framework icludig the Ord s momet equatio. Oe ca overcome the relative iefficiecy of the Ord s momet estimator by the selectio of a efficiet P as above. 9

22 Assumptio.2. Matrices {W } ad {S } are bouded i both row ad colum sum orms (Hor ad Johso, 985). The diagoal elemets of W are zero. Assumptio.3. Elemets of X are uiformly bouded costats, Z has full colum rak ad lim Z Z exists ad is osigular. Assumptio.4. Costat -dimesioal square matrices {P = [p,ij ]} which satisfy tr(p M Z ) = 0 are bouded i both row ad colum sum orms. The existece of a momet higher tha the fourth order of the disturbaces i Assumptio 2. is eeded for the applicatio of the cetral limit theorem for liear ad quadratic forms (Kelejia ad Prucha, 200). The boudedess i row ad colum sum orms of a sequece of matrices i Assumptio.2 origiated i Kelejia ad Prucha (998, 999, 200). Assumptio.3 is required for coveiece, as i Lee (2004a). As P is ofte geerated from W, it is reasoable to assume that {P } are bouded i both row ad colum sum orms. The quadratic momet equatio Eq. (.9) has two roots i geeral. Uder certai coditios, oe of the roots is cosistet. Let B s = B + B for ay -dimesioal square matrix B. Propositio.. Uder Assumptios 2..4, if (Z θ 0 ) P M Z G (Z θ 0 ) + 2 σ2 0 tr(p s G s ) were o-egative, the cosistet root for ρ 0 of Eq. (.9) would be ˆρ = b b 2 4a c 2a, (.2) where a = y W P M Z W y, b = y (P M Z ) s W y ad c = y P M Z y ; but if (Z θ 0 ) P M Z G (Z θ 0 ) + 2 σ2 0 tr(p s G s ) were egative, the cosistet root would be ˆρ 2 = b + b 2 4a c 2a, (.22) whe lim [(Z θ 0 ) G P M Z G (Z θ 0 ) + σ 2 0 tr(g P G )] 0. I the case that lim [(Z θ 0 ) G P M Z G (Z θ 0 ) + σ 2 0 tr(g P G )] = 0, ˆρ 3 = c /b is the uique cosistet root if lim [(Z θ 0 ) P M Z G (Z θ 0 ) + 2 σ2 0 tr(p s G s )] 0. The coditios that lim [(Z θ 0 ) G P M Z G (Z θ 0 ) + σ 2 0 tr(g P G )] 0 ad lim [(Z θ 0 ) P M Z G (Z θ 0 ) + 2 σ2 0 tr(p s G s )] 0 guaratee that a / ad b / do 0

23 ot coverge to zero i probability, respectively. Let H (ρ) = G (ρ) tr(g (ρ)m Z ) d M Z ad f(p ) = (Z θ 0 ) P M Z G (Z θ 0 )+ 2 σ2 0 tr(p s G s ) = (Z θ 0 ) P M Z H (Z θ 0 )+ 2 σ2 0 tr(p s H s ) with H = H (ρ 0 ). 2 The sig of f(p ) depeds o the correlatio betwee P ad H. If P = H, the f(h ) 0 ad Eq. (.2) is the cosistet root whe a / o P (). By cotiuity, f ( H (ρ) ) is o-egative whe ρ is close to ρ 0. I empirical applicatios, ρ 0 is ofte positive, the P = H (0.5) or P = H (0) = W tr(w M Z ) d M Z could geerate a cosistet root estimator of the form Eq. (.2). Give P, the scalars a, b ad c are products of vectors ad matrices, so the computatioal cost of Eq. (.2) or (.22) is miimal. The asymptotic distributio of the cosistet root ˆρ ca be derived from a first order expasio of g (ˆρ ) = 0 at ρ 0. As g (ρ 0 ) is quadratic i the disturbaces, the cetral limit theorem for liear ad quadratic forms is applicable. Propositio.2. The cosistet root ˆρ i Propositio. has the asymptotic distributio that (ˆρ ρ 0 ) D N ( 0, Ω ), where Ω = V ρ Σ 2 ρ with { V ρ = lim σ 2 0 (Z θ 0 ) P M Z P (Z θ 0 ) + 2 E(ɛ 3 i)(z θ 0 ) P M Z Diag(P M Z )l + [E(ɛ 4 i) 3σ0] 4 p 2,ii + 2 σ4 0 tr(pp s ) s } i= ad Σ ρ = lim [(Z θ 0 ) P M Z G (Z θ 0 ) + 2 σ2 0 tr(p s G s )] beig assumed to exist ad be o-zero. The V ρ i the above propositio is the limit of the variace of g (ρ 0 ), so it is geerally positive. Whe E(ɛ 3 i ) = E(ɛ4 i ) 3σ4 0 = 0, e.g., ɛ i s are i.i.d. ormal, the asymptotic variace of ˆρ reduces to Ω = lim σ2 0 (Zθ 0) P M Z P (Zθ 0)+ 2 σ4 0 tr(p sp s). The, by applyig the Cauchy [(Z θ 0 ) P M Z G (Z θ 0 )+ 2 σ2 0 tr(p sgs )]2 iequality, H is the best P matrix such that the asymptotic variace of this cosistet 2 Note that usig P = G tr(g M Z ) M d Z ad P = G tr(g M Z ) I d geerate the same root estimator. We use H for arrative coveiece, but we may use G tr(g M Z ) I d whe calculatig a root estimate.

24 root estimator is the smallest. As poited out earlier, with the best P (= H ) matrix, the cosistet root estimator has the form (b b 2 4a c )/(2a ) whe a / o P (). Propositio.3. Whe E(ɛ 3 i ) = E(ɛ4 i ) 3σ4 0 = 0, suppose that lim {(Z θ 0 ) G M Z G (Z θ 0 ) + 2 σ2 0[tr(G s G s ) tr2 (G s )]} exists ad is o-zero, the best root estimator is ˆρ b, = b b 2 4a c 2a, (.23) where a = y W H M Z W y, b = y (H M Z ) s y ad c = y H M Z y, i the sese that (ˆρ b, ρ 0 ) D N(0, Ω b ) with Ω b Ω, where Ω b = σ0{ 2 [ lim (Z θ 0 ) G M Z G (Z θ 0 ) + ( 2 σ2 0 tr(g s G s ) tr2 (G s ) )]}. Whe E(ɛ 3 i ) = E(ɛ4 i ) 3σ4 0 = 0, the asymptotic variace Ω b for the best root estimator i the above propositio is the same as that for the QMLE (Lee, 2004a). Whe the coditio E(ɛ 3 i ) = E(ɛ4 i ) 3σ4 0 = 0 does ot hold, the root estimator ˆρ b, may lose efficiecy. Note that o matter whether the coditio holds or ot, ˆρ b, i the above propositio is the cosistet root estimator whe H is used as the P matrix. As H ivolves the ukow parameter ρ 0, it ca be estimated by usig a iitial cosistet estimator for ρ 0. A estimated H would geerate a root estimator with the same limitig distributio as ˆρ b,. Propositio.4. Suppose that ˆρ is a -cosistet estimator of ρ 0, ad The the root estimator lim [(Z θ 0 ) G H M Z G (Z θ 0 ) + σ0 2 tr(g H G )] 0. ρ b, = ˆb ˆb2 4â ĉ 2â, (.24) where â = y W H (ˆρ )M Z W y, ˆb = y ( H (ˆρ )M Z ) sw y ad ĉ = y H (ˆρ )M Z y, is cosistet ad has the same limitig distributio as ˆρ b,. 2

25 A iitial cosistet estimator ˆρ may be derived by usig H (0) as the P matrix. Based o H (ˆρ ), Eq. (.24) is the best root estimator whe E(ɛ 3 i ) = E(ɛ4 i ) 3σ4 0 = 0.3 We shall use the otatio RE sd for this root estimator based o the SD model. Replacig Z with X everywhere above, we obtai the root estimator RE sar specifically for the SAR model. Note that the expressio for H (ρ) ivolves a matrix iverse (I ρw ), which is computatioally itesive for large sample sizes. 4 If ρw < with a matrix orm, the we have the expasio (I ρw ) = I +ρw +ρ 2 W ad (I ρw ) [I + ρw + + ρ r W] r = ρ r+ W r+ (I ρw ) < ρw r+ /( ρw ). I the secod step of computig a root estimate, we may start from usig a few term approximatio (I + ˆρ W + + ˆρ r W r )W tr[(i+ˆρw + +ˆρ r W r )W M Z ] d M Z of H (ˆρ ) i Eq. (.24). If the chage of the root estimate i absolute value is smaller tha a chose tolerace level, we ca stop ad report the estimate; otherwise, we may use (r + ) term approximatio of H (ρ) ad also use the ewly computed estimate from the last step i computig a ew ˆρ usig Eq. (.24). We could use more ad more terms to approximate H (ρ) util the tolerace criterio is met. This procedure turs out to be very efficiet i our Mote Carlo study..3.2 A Root Estimator: Heteroskedastic Case Whe there is ukow heteroskedasticity i disturbaces, from Eq. (.4), the expectatio of the left-had side of the momet equatio Eq. (.9) over is tr(p M Z Σ ), which geerally does ot coverge to zero eve if tr(p M Z ) = 0. I order to derive a cosistet root estimator from solvig Eq. (.9), we require P M Z to have a zero diagoal, so that 3 Because a iitial cosistet estimator ˆρ, e.g., derived with H (0)[= W tr(w M Z ) M d Z ], has a closed form expressio, the feasible two step root estimator will also have a closed form expressio as the aalytical expressio of the iitial ˆρ ca be substituted ito H (ˆρ ) i its derivatio. Li et al. (2007) have emphasized o closed form statistics for exploratory aalyses. They propose the AP LE but dismiss Ord s quadratic root estimator because of the eed to solve the quadratic momet equatio (as well as its possible iefficiecy). They have overlooked possible aalytical solutios of a quadratic equatio. 4 The best GMM estimator i Lee (2007a) or Lee (2007b) also ivolves this matrix iverse. 3

26 the expectatio of the left-had side of Eq. (.9) at ρ 0 would be zero. 5 Assumptio.5. The costat -dimesioal square matrices {P }, which satisfy that P M Z has a zero diagoal, are bouded i both row ad colum sum orms. We make the followig assumptio about the ukow heteroskedasticity. Assumptio.6. ɛ i s i ɛ = (ɛ,..., ɛ ) are idepedet (0, σi 2 ) ad the momets E ɛ 4+η i for some η > 0 exist ad are uiformly bouded for all ad i. The cosistet root is described i the followig propositio. The regularity coditios are similar to those i Propositio. after takig ito accout the heteroskedastic variace matrix Σ. Propositio.5. Uder Assumptios.2,.3,.5 ad.6, if (Z θ 0 ) P M Z G (Z θ 0 ) + tr(σ P s G ) were o-egative, the cosistet root would be ˆρ = b b 2 4a c 2a, (.25) where a = y W P M Z W y, b = y (P M Z ) s W y ad c = y P M Z y ; if tr(σ P s G ) + (Z θ 0 ) P M Z G (Z θ 0 ) were egative, the cosistet root would be ˆρ 2 = b + b 2 4a c 2a, (.26) whe lim [(Z θ 0 ) G P M Z G (Z θ 0 ) + tr(σ G P G )] 0. I the case that lim [(Z θ 0 ) G P M Z G (Z θ 0 ) + tr(σ G P G )] = 0, ˆρ 3 = c /b is the uique cosistet root if lim [(Z θ 0 ) P M Z G (Z θ 0 ) + tr(σ P s G )] 0. 5 As i the homoskedastic case, if P, istead of P M Z, is required to have a zero diagoal, the we ca still obtai a cosistet GMM estimator, sice the expectatio of the momet equatio over at ρ 0 coverges to zero as goes to ifiity i this case. We require P M Z to have a zero diagoal so that better small sample properties may be obtaied. Let P = [P,..., P ] ad M Z = [M,..., M ], where P i ad M i are -dimesioal vectors, the P M Z has a zero diagoal meas that P satisfies P im i = 0, i =,...,. So we have may choices of the P matrix. I particular, give ay -dimesioal square matrix A, we may let P = A Diag(A M Z )[Diag(M Z )] or P = A Diag(A M Z )[Diag(M Z )] M Z, if every diagoal elemet of M Z is o-zero. I the case that some diagoal elemets of M Z are zero, we may simply let P = A Diag(A ), or adjust A to be A such that the correspodig diagoal elemets of A M Z are zero ad let P = A Diag(A M Z )[Diag(M Z )] or P = A Diag(A M Z )[Diag(M Z )] M Z, where B deotes a geeralized matrix iverse for a matrix B. 4

27 The coditios that lim [(Z θ 0 ) G P M Z G (Z θ 0 ) + tr(σ G P G )] 0 ad lim [(Z θ 0 ) P M Z G (Z θ 0 ) + tr(σ P s G )] 0 are equivalet to oe-zero probability limits of a / ad b /, respectively. Propositio.6. The cosistet root ˆρ i Propositio.5 has the asymptotic distributio that (ˆρ ρ 0 ) D N(0, Ω), where Ω = V ρ Σ 2 ρ with V ρ = lim [(Z θ 0 ) P M Z Σ M Z P (Z θ 0 ) + tr(σ P Σ P s )] ad Σ ρ = lim [(Z θ 0 ) P M Z G (Z θ 0 ) + tr(σ P s G )] beig assumed to exist ad be o-zero. Note that V ρ is the limit of the variace of g (ρ 0 ). As cotrary to the homogeous variace case, the third ad fourth momets of o-ormal disturbaces do ot play a role i the asymptotic variace of the estimator due to the desig of P M Z havig a zero diagoal. Sice the asymptotic variace of the cosistet root i the above propositio ivolves ukow heteroskedasticity terms, the best selectio of the matrix P may be uavailable. If oe of the diagoal elemets of M Z is zero, a possible choice of P i practice might be the cosistetly estimated G Diag(G M Z )[Diag(M Z )]. To get a estimator for G, we may first derive a iitial cosistet estimator ˆρ for ρ 0 based o the momet equatio y S (ρ) { W Diag(W M Z )[Diag(M Z )] } M Z S (ρ)y = 0. The usig G (ˆρ ) Diag[G (ˆρ )M Z ][Diag(M Z )] as the P matrix i the momet equatio, we derive the root estimator ˆρ. We shall call this root estimator RE sd. The root estimator specifically for the SAR model is deoted by RE sar..4 Mote Carlo Study We coduct some Mote Carlo experimets to ivestigate fiite sample performaces ad computig times of the QMLE ad various versios of RE, AP LE ad ACME. The DGP is either the SAR or SE model. For the QMLE, the likelihood fuctio is derived as follows: igore (ukow) heteroskedasticity eve though there might be ad form the likelihood based o the SAR model if the DGP is the SAR process or based o the SE model if the 5

28 DGP is the SE process. For RE sd, i the homoskedastic case, the iitial estimator ˆρ is the root ˆρ of Eq. (.9) with P = W tr(w M Z ) d I, ad RE sd deotes the correspodig root estimator Eq. (.24) i Propositio.4 with P = G (ˆρ ) tr[g (ˆρ)M Z ] d I ; i the heteroskedastic case, the iitial estimator is the root ˆρ of Eq. (.25) with P = W Diag(W M Z )[Diag(M Z )], ad RE sd deotes the correspodig root estimator ˆρ of Eq. (.25) with P = G (ˆρ ) Diag[G (ˆρ )M Z ][Diag(M Z )]. The root estimate RE sar specific for the SAR model is derived similarly. We cosider three differet spatial weights matrices W, W 2 ad W 3. The W is the circular world matrix cosidered i Arraiz et al. (200). Specifically, each of the first /3 ad last /3 rows except the first ad last rows oly has two o-zero elemets, 6 which are i the positios (i, i ) ad (i, i + ) ad are equal to 0.5. For the first row, the o-zero elemets are i the positios (, 2) ad (, ) ad they are equal to 0.5; for the last row, the o-zero elemets are i the positios (, ) ad (, ) ad they are also equal to 0.5. Each of the middle /3 rows has 0 o-zero elemets, which are i the positios (i, i 5),..., (i, i ), (i, i + ),..., (i, i + 5) ad are equal to 0.. The W 2 ad W 3 are geerated accordig to, respectively, the quee ad rook criteria o regular m m grids, leadig to a sample size of = m 2. We use the row-ormalized W 2 ad W 3. The exogeous variable matrix X cosists of a itercept term, a exogeous variable draw from the ormal distributio N(3, ), ad the third oe draw from the uiform distributio U(, 2). The true parameter vector correspodig to these exogeous variables is β 0 = (0.8, 0.2,.5). The desig of exogeous variables ad correspodig parameters has bee used i Li ad Lee (200). For the homoskedastic case, the error terms are radomly draw from the ormal distributio N(0, ). For the heteroskedastic case, two desigs of heteroskedasticity are cosidered: Heteroskedasticity desig (HD-): For W, the stadard deviatio (STD) is equal to a costat times the umber of o-zero elemets i each row; 7 for W 2 ad W 3, the STD is equal to a costat times the absolute value of the secod exogeous 6 Whe /3 is ot a iteger, the smallest iteger larger tha /3 is take. 7 This desig is oe used i Arraiz et al. (200). 6

29 variable. 8 The costats are chose such that the average STD is equal to 0.5. Heteroskedasticity desig 2 (HD-2): For W, the STD is equal to a costat times the iverse of the umber of o-zero elemets i each row; for W 2 ad W 3, the STD is equal to a costat times the iverse of the absolute value of the secod exogeous variable. Agai the costats are chose to make the average STD be equal to 0.5. We calculate various measures of the autoregressive coefficiet by focusig o oegative ρ 0 values, as this is usually the case i empirical applicatios. For each case of ρ 0, the umber of repetitios is Figs. 6 compare the mea, STD ad root mea square error (RMSE) of the QMLE ad differet versios of RE, AP LE ad ACME. For these figures, we have = 400. Figs. ad 2 are the case whe there is o ukow heteroskedasticity i the disturbaces. Whe the DGP is the SAR process, from Fig., the QMLE ad RE sar have similarly small bias (i absolute value) for differet spatial weights matrices ad ρ 0 s, while the biases of AP LE sar, ACME sar, AP LE sd ad ACME sd are oly small whe ρ 0 is close to zero ad geerally icrease as ρ 0 icreases. I terms of bias, AP LE sar has ot show a advatage over AP LE sd, though AP LE sar is based o the DGP. The AP LE sar ad AP LE sd have similar bias for W ad W 2, but AP LE sar has large bias for large ρ 0 s i the case of W 2. The QMLE, RE sar, AP LE sar ad ACME sar have similar STD that is smaller tha those of RE sd, AP LE sd ad ACME sd, which is expected sice the latter oes are based o the more geeral SD model. It is oted that for W, the bias of ACME sd is sigificatly larger tha that of other statistics. The RMSEs of differet statistics show similar patters as their biases. Whe the DGP is the SE process, the bias, STD, RMSE of the QMLE, RE sd, AP LE se, AP LE sd ad ACME sd are plotted i Fig. 2. The biases of statistics other tha AP LE se have similar patters as the correspodig oes i Fig.. The AP LE se have ot show a advatage over AP LE sd i terms of smaller bias, which is obvious for W 2 for which AP LE se usually has larger bias tha AP LE sd. The STDs of all statistics are very similar. The AP LE se, AP LE sd ad ACME sd have larger RMSEs tha 8 For W 2 ad W 3, (m 2) 2 rows would have the same umber of o-zero elemets, which is approximately 00[(m 4)/m]% of the total umber of rows. If the same heteroskedasticity desig as for W is used, there would be little heteroskedasticity. 7

30 the QMLE ad RE sd for W ad W 3, while all statistics have similar RMSEs for W 2. Figs. 3 6 show the results whe there is ukow heteroskedasticity i the disturbaces. Figs. 3 ad 4 correspod to the DGP beig the SAR process but with differet desigs of heteroskedasticity, ad Figs. 5 ad 6 correspod to the DGP beig the SE process with differet variaces. I geeral, RE sar ad RE sd have the smallest bias i Figs. 3 ad 4 ad RE sd has the smallest bias i Figs. 5 ad 6. Sice the QMLE has igored the heteroskedasticity, it may geerate large bias i some cases, e.g., its bias is close to 0.2 whe ρ 0 = 0.6 for W i Fig. 5. I most figures, however, the QMLE has relatively small bias. The statistics derived from homoskedastic models AP LE sar, AP LE se ad AP LE sd geerally have relatively small bias whe ρ 0 is small ad relatively large bias whe ρ 0 is large. We ote that for W i Fig. 5, both AP LE se ad AP LE sd have very large bias for positive ρ 0 s. I Figs. 3 ad 4, like AP LE sar ad AP LE sd, ACME sar ad ACME sd have large bias for large ρ 0 s; i Figs. 5 ad 6, ACME sd have large bias for large ρ 0 s except for the case with W i Fig. 5, where the bias of ACME sd is smaller tha those of the QMLE, AP LE se ad AP LE se. The STDs ad RMSEs i Figs. 3 ad 4 are similar to the correspodig oes i Fig., ad the STDs ad RMSEs i Figs. 5 ad 6 are similar to the correspodig oes i Fig. 2. Table compares the computig times ad fiite sample properties of differet statistics whe the sample size is large. The DGP is the SAR model. We focus o the QMLE, RE sar ad AP LE sar, as computig the other statistics above are expected to take similar time. To compute RE sar, we use the procedure described i the last paragraph of subsubsectio.3. which starts from usig a two term approximatio (I + ˆρ W + ˆρ 2 W 2 )W tr[(i +ˆρ W +ˆρ2 W 2)W M Z ] d M Z of H (ˆρ ) i Eq. (.24). The tolerace criteria for RE sar ad the QMLE are both set to be The reported results are from Matlab o a desktop computer with Itel Core i processor ad 8 gigabyte memory. For the same sample size ad spatial weights matrix i the DGP, while computig a AP LE sar takes about the same time for differet ρ 0 s, computig the QMLE ad RE sar take more time whe ρ 0 becomes larger. For moderate values of ρ 0, computig RE sar oly takes slightly loger time tha computig the AP LE sar ad is at least 8 times faster tha computig the QMLE. The bias, STD ad RMSE have the same patter as we have see i Fig.. 8

31 Bias Bias with W ρ 0 STD STD with W ρ 0 RMSE QMLE RE sar APLE sar ACME sar RMSE with W RE sd APLE sd ACME sd ρ 0 0 Bias with W 2 0. STD with W 2 0. RMSE with W 2 Bias 0.05 STD 0.05 RMSE ρ 0 ρ 0 ρ 0 Bias Bias with W 3 STD STD with W 3 RMSE RMSE with W ρ ρ ρ 0 Figure : Compariso of the bias, STD ad RMSE of the QMLE, RE sar, AP LE sar, ACME sar, RE sd, AP LE sd ad ACME sd whe the DGP is the SAR model uder homoskedasticity. Bias Bias with W ρ 0 STD STD with W ρ 0 RMSE QMLE RE sd APLE se RMSE with W APLE sd ACME sd ρ Bias with W 2 0. STD with W 2 0. RMSE with W 2 Bias STD 0.05 RMSE ρ ρ ρ Bias with W STD with W 3 RMSE with W 3 Bias ρ 0 STD ρ 0 RMSE ρ 0 Figure 2: Compariso of the bias, STD ad RMSE of the QMLE, RE sd, AP LE se, AP LE sd ad ACME sd whe the DGP is the SE model uder homoskedasticity. 9

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

It should be unbiased, or approximately unbiased. Variance of the variance estimator should be small. That is, the variance estimator is stable.

It should be unbiased, or approximately unbiased. Variance of the variance estimator should be small. That is, the variance estimator is stable. Chapter 10 Variace Estimatio 10.1 Itroductio Variace estimatio is a importat practical problem i survey samplig. Variace estimates are used i two purposes. Oe is the aalytic purpose such as costructig

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

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

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

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

ECE 901 Lecture 12: Complexity Regularization and the Squared Loss

ECE 901 Lecture 12: Complexity Regularization and the Squared Loss ECE 90 Lecture : Complexity Regularizatio ad the Squared Loss R. Nowak 5/7/009 I the previous lectures we made use of the Cheroff/Hoeffdig bouds for our aalysis of classifier errors. Hoeffdig s iequality

More information

1 Inferential Methods for Correlation and Regression Analysis

1 Inferential Methods for Correlation and Regression Analysis 1 Iferetial Methods for Correlatio ad Regressio Aalysis I the chapter o Correlatio ad Regressio Aalysis tools for describig bivariate cotiuous data were itroduced. The sample Pearso Correlatio Coefficiet

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

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

FACULTY OF MATHEMATICAL STUDIES MATHEMATICS FOR PART I ENGINEERING. Lectures

FACULTY OF MATHEMATICAL STUDIES MATHEMATICS FOR PART I ENGINEERING. Lectures FACULTY OF MATHEMATICAL STUDIES MATHEMATICS FOR PART I ENGINEERING Lectures MODULE 5 STATISTICS II. Mea ad stadard error of sample data. Biomial distributio. Normal distributio 4. Samplig 5. Cofidece itervals

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

Essays On Spatial Econometrics: Estimation Methods And Applications

Essays On Spatial Econometrics: Estimation Methods And Applications City Uiversity of New York CUNY CUNY Academic Works Dissertatios, Theses, ad Capstoe Projects Graduate Ceter 2--205 Essays O Spatial Ecoometrics: Estimatio Methods Ad Applicatios Osma Doga Graduate Ceter,

More information

Economics 326 Methods of Empirical Research in Economics. Lecture 18: The asymptotic variance of OLS and heteroskedasticity

Economics 326 Methods of Empirical Research in Economics. Lecture 18: The asymptotic variance of OLS and heteroskedasticity Ecoomics 326 Methods of Empirical Research i Ecoomics Lecture 8: The asymptotic variace of OLS ad heteroskedasticity Hiro Kasahara Uiversity of British Columbia December 24, 204 Asymptotic ormality I I

More information

Since X n /n P p, we know that X n (n. Xn (n X n ) Using the asymptotic result above to obtain an approximation for fixed n, we obtain

Since X n /n P p, we know that X n (n. Xn (n X n ) Using the asymptotic result above to obtain an approximation for fixed n, we obtain Assigmet 9 Exercise 5.5 Let X biomial, p, where p 0, 1 is ukow. Obtai cofidece itervals for p i two differet ways: a Sice X / p d N0, p1 p], the variace of the limitig distributio depeds oly o p. Use the

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

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

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

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

10. Comparative Tests among Spatial Regression Models. Here we revisit the example in Section 8.1 of estimating the mean of a normal random

10. Comparative Tests among Spatial Regression Models. Here we revisit the example in Section 8.1 of estimating the mean of a normal random Part III. Areal Data Aalysis 0. Comparative Tests amog Spatial Regressio Models While the otio of relative likelihood values for differet models is somewhat difficult to iterpret directly (as metioed above),

More information

Stochastic Simulation

Stochastic Simulation Stochastic Simulatio 1 Itroductio Readig Assigmet: Read Chapter 1 of text. We shall itroduce may of the key issues to be discussed i this course via a couple of model problems. Model Problem 1 (Jackso

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

A Note on Box-Cox Quantile Regression Estimation of the Parameters of the Generalized Pareto Distribution

A Note on Box-Cox Quantile Regression Estimation of the Parameters of the Generalized Pareto Distribution A Note o Box-Cox Quatile Regressio Estimatio of the Parameters of the Geeralized Pareto Distributio JM va Zyl Abstract: Makig use of the quatile equatio, Box-Cox regressio ad Laplace distributed disturbaces,

More information

Random Matrices with Blocks of Intermediate Scale Strongly Correlated Band Matrices

Random Matrices with Blocks of Intermediate Scale Strongly Correlated Band Matrices Radom Matrices with Blocks of Itermediate Scale Strogly Correlated Bad Matrices Jiayi Tog Advisor: Dr. Todd Kemp May 30, 07 Departmet of Mathematics Uiversity of Califoria, Sa Diego Cotets Itroductio Notatio

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

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

Lecture 33: Bootstrap

Lecture 33: Bootstrap Lecture 33: ootstrap Motivatio To evaluate ad compare differet estimators, we eed cosistet estimators of variaces or asymptotic variaces of estimators. This is also importat for hypothesis testig ad cofidece

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

Study the bias (due to the nite dimensional approximation) and variance of the estimators

Study the bias (due to the nite dimensional approximation) and variance of the estimators 2 Series Methods 2. Geeral Approach A model has parameters (; ) where is ite-dimesioal ad is oparametric. (Sometimes, there is o :) We will focus o regressio. The fuctio is approximated by a series a ite

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

Kernel density estimator

Kernel density estimator Jauary, 07 NONPARAMETRIC ERNEL DENSITY ESTIMATION I this lecture, we discuss kerel estimatio of probability desity fuctios PDF Noparametric desity estimatio is oe of the cetral problems i statistics I

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

TMA4245 Statistics. Corrected 30 May and 4 June Norwegian University of Science and Technology Department of Mathematical Sciences.

TMA4245 Statistics. Corrected 30 May and 4 June Norwegian University of Science and Technology Department of Mathematical Sciences. Norwegia Uiversity of Sciece ad Techology Departmet of Mathematical Scieces Corrected 3 May ad 4 Jue Solutios TMA445 Statistics Saturday 6 May 9: 3: Problem Sow desity a The probability is.9.5 6x x dx

More information

LECTURE 14 NOTES. A sequence of α-level tests {ϕ n (x)} is consistent if

LECTURE 14 NOTES. A sequence of α-level tests {ϕ n (x)} is consistent if LECTURE 14 NOTES 1. Asymptotic power of tests. Defiitio 1.1. A sequece of -level tests {ϕ x)} is cosistet if β θ) := E θ [ ϕ x) ] 1 as, for ay θ Θ 1. Just like cosistecy of a sequece of estimators, Defiitio

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

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

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

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

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

Complex Analysis Spring 2001 Homework I Solution

Complex Analysis Spring 2001 Homework I Solution Complex Aalysis Sprig 2001 Homework I Solutio 1. Coway, Chapter 1, sectio 3, problem 3. Describe the set of poits satisfyig the equatio z a z + a = 2c, where c > 0 ad a R. To begi, we see from the triagle

More information

Statistical Properties of OLS estimators

Statistical Properties of OLS estimators 1 Statistical Properties of OLS estimators Liear Model: Y i = β 0 + β 1 X i + u i OLS estimators: β 0 = Y β 1X β 1 = Best Liear Ubiased Estimator (BLUE) Liear Estimator: β 0 ad β 1 are liear fuctio of

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

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

Approximate Confidence Interval for the Reciprocal of a Normal Mean with a Known Coefficient of Variation

Approximate Confidence Interval for the Reciprocal of a Normal Mean with a Known Coefficient of Variation Metodološki zvezki, Vol. 13, No., 016, 117-130 Approximate Cofidece Iterval for the Reciprocal of a Normal Mea with a Kow Coefficiet of Variatio Wararit Paichkitkosolkul 1 Abstract A approximate cofidece

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

Because it tests for differences between multiple pairs of means in one test, it is called an omnibus test.

Because it tests for differences between multiple pairs of means in one test, it is called an omnibus test. Math 308 Sprig 018 Classes 19 ad 0: Aalysis of Variace (ANOVA) Page 1 of 6 Itroductio ANOVA is a statistical procedure for determiig whether three or more sample meas were draw from populatios with equal

More information

Element sampling: Part 2

Element sampling: Part 2 Chapter 4 Elemet samplig: Part 2 4.1 Itroductio We ow cosider uequal probability samplig desigs which is very popular i practice. I the uequal probability samplig, we ca improve the efficiecy of the resultig

More information

Sequences A sequence of numbers is a function whose domain is the positive integers. We can see that the sequence

Sequences A sequence of numbers is a function whose domain is the positive integers. We can see that the sequence Sequeces A sequece of umbers is a fuctio whose domai is the positive itegers. We ca see that the sequece 1, 1, 2, 2, 3, 3,... is a fuctio from the positive itegers whe we write the first sequece elemet

More information

4. Partial Sums and the Central Limit Theorem

4. Partial Sums and the Central Limit Theorem 1 of 10 7/16/2009 6:05 AM Virtual Laboratories > 6. Radom Samples > 1 2 3 4 5 6 7 4. Partial Sums ad the Cetral Limit Theorem The cetral limit theorem ad the law of large umbers are the two fudametal theorems

More information

Exponential Families and Bayesian Inference

Exponential Families and Bayesian Inference Computer Visio Expoetial Families ad Bayesia Iferece Lecture Expoetial Families A expoetial family of distributios is a d-parameter family f(x; havig the followig form: f(x; = h(xe g(t T (x B(, (. where

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

Topic 9: Sampling Distributions of Estimators

Topic 9: Sampling Distributions of Estimators Topic 9: Samplig Distributios of Estimators Course 003, 2018 Page 0 Samplig distributios of estimators Sice our estimators are statistics (particular fuctios of radom variables), their distributio ca be

More information

Introductory statistics

Introductory statistics CM9S: Machie Learig for Bioiformatics Lecture - 03/3/06 Itroductory statistics Lecturer: Sriram Sakararama Scribe: Sriram Sakararama We will provide a overview of statistical iferece focussig o the key

More information

Product measures, Tonelli s and Fubini s theorems For use in MAT3400/4400, autumn 2014 Nadia S. Larsen. Version of 13 October 2014.

Product measures, Tonelli s and Fubini s theorems For use in MAT3400/4400, autumn 2014 Nadia S. Larsen. Version of 13 October 2014. Product measures, Toelli s ad Fubii s theorems For use i MAT3400/4400, autum 2014 Nadia S. Larse Versio of 13 October 2014. 1. Costructio of the product measure The purpose of these otes is to preset the

More information

11 Correlation and Regression

11 Correlation and Regression 11 Correlatio Regressio 11.1 Multivariate Data Ofte we look at data where several variables are recorded for the same idividuals or samplig uits. For example, at a coastal weather statio, we might record

More information

Topic 9: Sampling Distributions of Estimators

Topic 9: Sampling Distributions of Estimators Topic 9: Samplig Distributios of Estimators Course 003, 2018 Page 0 Samplig distributios of estimators Sice our estimators are statistics (particular fuctios of radom variables), their distributio ca be

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

Goodness-Of-Fit For The Generalized Exponential Distribution. Abstract

Goodness-Of-Fit For The Generalized Exponential Distribution. Abstract Goodess-Of-Fit For The Geeralized Expoetial Distributio By Amal S. Hassa stitute of Statistical Studies & Research Cairo Uiversity Abstract Recetly a ew distributio called geeralized expoetial or expoetiated

More information

Discrete Mathematics for CS Spring 2008 David Wagner Note 22

Discrete Mathematics for CS Spring 2008 David Wagner Note 22 CS 70 Discrete Mathematics for CS Sprig 2008 David Wager Note 22 I.I.D. Radom Variables Estimatig the bias of a coi Questio: We wat to estimate the proportio p of Democrats i the US populatio, by takig

More information

Estimation for Complete Data

Estimation for Complete Data Estimatio for Complete Data complete data: there is o loss of iformatio durig study. complete idividual complete data= grouped data A complete idividual data is the oe i which the complete iformatio of

More information

POWER COMPARISON OF EMPIRICAL LIKELIHOOD RATIO TESTS: SMALL SAMPLE PROPERTIES THROUGH MONTE CARLO STUDIES*

POWER COMPARISON OF EMPIRICAL LIKELIHOOD RATIO TESTS: SMALL SAMPLE PROPERTIES THROUGH MONTE CARLO STUDIES* Kobe Uiversity Ecoomic Review 50(2004) 3 POWER COMPARISON OF EMPIRICAL LIKELIHOOD RATIO TESTS: SMALL SAMPLE PROPERTIES THROUGH MONTE CARLO STUDIES* By HISASHI TANIZAKI There are various kids of oparametric

More information

Notes 19 : Martingale CLT

Notes 19 : Martingale CLT Notes 9 : Martigale CLT Math 733-734: Theory of Probability Lecturer: Sebastie Roch Refereces: [Bil95, Chapter 35], [Roc, Chapter 3]. Sice we have ot ecoutered weak covergece i some time, we first recall

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

CS284A: Representations and Algorithms in Molecular Biology

CS284A: Representations and Algorithms in Molecular Biology CS284A: Represetatios ad Algorithms i Molecular Biology Scribe Notes o Lectures 3 & 4: Motif Discovery via Eumeratio & Motif Represetatio Usig Positio Weight Matrix Joshua Gervi Based o presetatios by

More information

Essays On Robust Estimators For Non-Identically Distributed Observations In Spatial Econometric And Time Series Models

Essays On Robust Estimators For Non-Identically Distributed Observations In Spatial Econometric And Time Series Models City Uiversity of New York CUNY CUNY Academic Works Dissertatios, Theses, ad Capstoe Projects Graduate Ceter 0-204 Essays O Robust Estimators For No-Idetically Distributed Observatios I Spatial Ecoometric

More information

Chimica Inorganica 3

Chimica Inorganica 3 himica Iorgaica Irreducible Represetatios ad haracter Tables Rather tha usig geometrical operatios, it is ofte much more coveiet to employ a ew set of group elemets which are matrices ad to make the rule

More information

t distribution [34] : used to test a mean against an hypothesized value (H 0 : µ = µ 0 ) or the difference

t distribution [34] : used to test a mean against an hypothesized value (H 0 : µ = µ 0 ) or the difference EXST30 Backgroud material Page From the textbook The Statistical Sleuth Mea [0]: I your text the word mea deotes a populatio mea (µ) while the work average deotes a sample average ( ). Variace [0]: The

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

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

A sequence of numbers is a function whose domain is the positive integers. We can see that the sequence

A sequence of numbers is a function whose domain is the positive integers. We can see that the sequence Sequeces A sequece of umbers is a fuctio whose domai is the positive itegers. We ca see that the sequece,, 2, 2, 3, 3,... is a fuctio from the positive itegers whe we write the first sequece elemet as

More information

Math 61CM - Solutions to homework 3

Math 61CM - Solutions to homework 3 Math 6CM - Solutios to homework 3 Cédric De Groote October 2 th, 208 Problem : Let F be a field, m 0 a fixed oegative iteger ad let V = {a 0 + a x + + a m x m a 0,, a m F} be the vector space cosistig

More information

This is an introductory course in Analysis of Variance and Design of Experiments.

This is an introductory course in Analysis of Variance and Design of Experiments. 1 Notes for M 384E, Wedesday, Jauary 21, 2009 (Please ote: I will ot pass out hard-copy class otes i future classes. If there are writte class otes, they will be posted o the web by the ight before class

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

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

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

ARIMA Models. Dan Saunders. y t = φy t 1 + ɛ t

ARIMA Models. Dan Saunders. y t = φy t 1 + ɛ t ARIMA Models Da Sauders I will discuss models with a depedet variable y t, a potetially edogeous error term ɛ t, ad a exogeous error term η t, each with a subscript t deotig time. With just these three

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

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

SOME THEORY AND PRACTICE OF STATISTICS by Howard G. Tucker

SOME THEORY AND PRACTICE OF STATISTICS by Howard G. Tucker SOME THEORY AND PRACTICE OF STATISTICS by Howard G. Tucker CHAPTER 9. POINT ESTIMATION 9. Covergece i Probability. The bases of poit estimatio have already bee laid out i previous chapters. I chapter 5

More information

Sequences. Notation. Convergence of a Sequence

Sequences. Notation. Convergence of a Sequence Sequeces A sequece is essetially just a list. Defiitio (Sequece of Real Numbers). A sequece of real umbers is a fuctio Z (, ) R for some real umber. Do t let the descriptio of the domai cofuse you; it

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

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

Basis for simulation techniques

Basis for simulation techniques Basis for simulatio techiques M. Veeraraghava, March 7, 004 Estimatio is based o a collectio of experimetal outcomes, x, x,, x, where each experimetal outcome is a value of a radom variable. x i. Defiitios

More information