LM-type tests for slope homogeneity in panel data models

Size: px
Start display at page:

Download "LM-type tests for slope homogeneity in panel data models"

Transcription

1 LM-type tests for slope homogenety n panel data models Jörg Bretung Unversty of Cologne Chrstoph Rolng Deutsche Bundesbank azar Salsh BGSE, Unversty of Bonn. July 4, 206 Abstract Ths paper employs the Lagrange Multpler LM prncple to test parameter homogenety across cross-secton unts n panel data models. The test can be seen as a generalzaton of the Breusch-Pagan test aganst random ndvdual effects to all regresson coeffcents. Whle the orgnal test procedure assumes a lkelhood framework under normalty, several useful varants of the LM test are presented to allow for non-normalty, heteroskedastcty and serally correlated errors. Moreover, the tests can be convenently computed va smple artfcal regressons. We derve the lmtng dstrbuton of the LM test and show that f the errors are not normally dstrbuted, the orgnal LM test s asymptotcally vald f the number of tme perods tends to nfnty. A smple modfcaton of the score statstc yelds an LM test that s robust to non-normalty f the number of tme perods s fxed. Further adjustments provde versons of the LM test that are robust to heteroskedastcty and seral correlaton. We compare the local power of our tests and the statstc proposed by Pesaran and Yamagata The results of the Monte Carlo experments suggest that the LM-type test can be substantally more powerful, n partcular, when the number of tme perods s small. JEL classfcaton: C2; C33 Keywords: panel data model; random coeffcents; LM test; heterogenous coeffcents We are grateful to the edtor Mchael Jansson and anonymous referee for helpful and constructve comments on an earler verson of ths paper. Correspondng author: Unversty of Cologne, Insttute of Econometrcs, Cologne, Germany. Emal: bretung@statstk.un-koeln.de, chrstoph.rolng@bundesbank.de, salsh@un-bonn.de. The vews expressed n ths paper do not necessarly reflect the vews of Deutsche Bundesbank or ts staff.

2 Introducton In classcal panel data analyss t s assumed that unobserved heterogenety s captured by ndvdual-specfc constants, whether they are assumed to be fxed or random. In many applcatons, however, t cannot be ruled out that slope coeffcents are also ndvdualspecfc. For nstance, heterogenous preferences among ndvduals may result n ndvdualspecfc prce or ncome elastctes. Ignorng ths form of heterogenety may result n based estmaton and nference. Therefore, t s mportant to test the assumpton of slope homogenety before applyng standard panel data technques such as the least-squares dummy-varable LSDV estmator for the fxed effect panel data model. If there s evdence for ndvdual-specfc slope parameters, economsts are nterested n estmatng a populaton average lke the mean of the ndvdual-specfc coeffcents. Pesaran and Smth 995 advocate mean group estmaton, where n a frst step the model s estmated separately for each cross-secton unt. In a second step, the untspecfc estmates are averaged to obtan an estmator for the populaton mean of the parameters. Alternatvely, Swamy 970 proposes a generalzed least squares GLS estmator for the random coeffcents model, whch assumes that the ndvdual regresson coeffcents are randomly dstrbuted around a common mean. In ths paper we derve a test for slope homogenety by employng the LM prncple wthn a random coeffcents framework, whch allows us to formulate the null hypothess of slope homogenety n terms of K restrctons on the varance parameters. Hence, the LM approach substantally reduces the number of restrctons to be tested compared to the set of K lnear restrctons on the coeffcents mpled by the test proposed by Pesaran and Yamagata 2008, henceforth referred to as PY. Ths does not mean, however, that our test s confned to detect random devaton n the coeffcents. In fact our test s optmal aganst the alternatve of random coeffcents but t s also powerful aganst any systematc varatons of the regresson coeffcents. Our approach s related but not dentcal to the condtonal LM test recently suggested by Juhl and Lugovskyy 204 whch s referred to as the JL test. The man dfference s that the latter test s derved for a more restrctve alternatve, where t s assumed that the ndvdual-specfc slope coeffcents attached to the K regressors have dentcal varances. In contrast, our test focuses on the alternatve that the coeffcents have dfferent varances whch allows us to test for heterogenety n a subset of the regresson coeffcents. Furthermore, the dervaton of our test follows the orgnal LM prncple nvolvng the nformaton matrx, whereas the JL test employs the outer product of the scores as an estmator of the nformaton matrx. Our smulaton study suggest that both non-standard features of the latter test may result n sze dstortons n small samples and a szable loss n power. An mportant advantage of the JL test s however that t s robust aganst non-gaussan and heteroskedastc errors. We therefore propose varants of the

3 orgnal LM test that share the robustness aganst non-gaussan and heteroskedastc errors. Furthermore, we also suggest a modfed LM test that s robust to serally correlated errors. Another contrbuton of the paper s the analyss of the local power of the test that allows us to compare the power propertes of the LM and PY tests. Specfcally, we fnd that the locaton parameter of the LM test depends on the cross-secton dsperson of the regresson varances, whereas the locaton parameter of the PY test only depends on the mean of the regressor varances. Thus, f the regressor varances dffer across the panel groups, the gan n power from usng the LM test may be substantal. The outlne of the paper s as follows. In Secton 2 we compare two tests for slope heterogenety recently proposed n the lterature. We ntroduce the random coeffcents model n 3 and lay out the standard assumptons for analyzng the large-sample propertes. In Secton 4 we derve the LM statstc and establsh ts asymptotc dstrbuton. Secton 5 dscusses several varants of the proposed test. Frst, we relax the normalty assumpton and extend the result of the prevous secton to ths more general settng. Second, we propose a regresson-based verson of the LM test. Secton 6 nvestgates the local asymptotc power of the LM test. Secton 7 descrbes the desgn of our Monte Carlo experments and dscusses the results. Secton 8 concludes. 2 Exstng tests To prepare the theoretcal dscusson n the followng sectons, we brefly revew the random coeffcents model and exstng tests. Followng Swamy 970, consder a lnear panel data model y t = x tβ + ɛ t, for =, 2,...,, and t =, 2,..., T, where y t s the dependent varable for unt at tme perod t, x t s a K vector of explanatory varables and ɛ t s an dosyncratc error wth zero mean and varance E ɛ 2 t = σ 2. For the slope coeffcent β we assume β = β + v, where β s a fxed K vector and v s a..d. random vector wth zero mean and K K covarance matrx Σ v. For more detals and extensons of the basc random coeffcent model see Hsao and Pesaran As ponted out by a referee, ths specfcaton may be replaced by some systematc varaton of the coeffcents that depends on observed varables. For example, we may specfy the devatons as β β = Γz +η, where z s some vector of observed varables possbly correlated wth x t. The correspondng varant of the LM test whch s dfferent from our LM test based assumng that v and x t are ndependent wll be optmal aganst ths partcular form of systematc varaton. In general, our test assumng ndependent varaton wth Γ = 0 wll also have power aganst systematc varatons but admttedly our test s not 2

4 The null hypothess of slope homogenety s β = β 2 = = β = β, 2 whch s equvalent to testng Σ v = 0. To test hypothess 2, Swamy suggests the statstc Ŝ = = β β X X WLS β s β WLS, 2 wth X = x,..., x T and β = X X X y s the ordnary least squares OLS estmator of for panel unt, and t =,..., T. The common slope parameter β s estmated by the weghted least-squares estmator β WLS = = X X s 2 where s 2 denotes the standard OLS estmator of σ 2. = X y s 2 Intutvely, f the regresson coeffcents are dentcal, the dfferences between the ndvdual estmators and the pooled estmator should be small. Therefore, Swamy s test rejects the null hypothess of homogenous slopes for large values of ths statstc, whch possesses a lmtng χ 2 dstrbuton wth K degrees of freedom as s fxed and T. Pesaran and Yamagata 2008 emphasze that n many emprcal applcatons s large relatve to T and the approxmaton by a χ 2 dstrbuton s unrelable. PY adapt the test to a settng n whch and T jontly tend to nfnty. In partcular, they assume ndvdual-specfc ntercepts and derve a test for the hypothess β = = β = β n, y t = α + x tβ + ɛ t. 3 The analogue of the pooled weghted least squares estmator above elmnates the unobserved fxed effects, β WFE = = X M ι X σ 2 = X M ι y σ 2, where M ι = I T ι T ι T /T, and ι T s a T vector of ones. A natural estmator for σ 2 y X β Mι y X β s σ 2 = T K, optmal aganst alternatve wth systematcally varyng coeffcents. 3

5 where β = X M ι X X M ι y and the test statstc becomes Ŝ = = β β X M ι X WFE β σ β WFE. 2 Employng a jont lmt theory for and T, PY obtan the lmtng dstrbuton as = Ŝ K 2K d 0,, 4 provded that, T and /T 0. Thus, by approprately centerng and standardzng the test statstc, nference can be carred out by resortng to the standard normal dstrbuton, provded the tme dmenson s suffcently large relatve to the crosssecton dmenson. PY propose several modfed versons of ths test, whch for brevty we shall refer to as the tests or statstcs. In partcular, to mprove the small sample propertes of the test, PY suggest the adjusted statstc under normally dstrbuted errors see Remark 2 n PY, adj = S K T +, 5 2K T K where S s computed as Ŝ but replacng σ2 σ 2 = by the varance estmator y X βfe Mι y X βfe T, 6 where β FE = X M ι X X M ι y s the standard fxed effects wthn-group = = estmator. ote that ths asymptotc framework does not seem to be well suted for typcal panel data applcatons where s large relatve to T. Therefore, t wll be of nterest to derve a test statstc that s vald when T s small say T = 0 and s very large say = 000, whch, for nstance, s encountered n mcroeconomc panels. scores The test statstc proposed by Juhl and Lugovskyy 204 s based on the ndvdual S = û M ι X X M ι û σ 2 trx M ι X, where û = y X βfe and tr A denotes the trace of the matrx A. The condtonal LM statstc results as CLM = S = S S S. 7 = = 4

6 It s nterestng to compare ths test statstc to the PY test whch s based on the sum Ŝ = = Ŝ wth Ŝ = β β X M ι X WFE β σ β WFE 2 = σ 2 u M ι X X M ι X X M ι u + o p f and T tend to nfnty. ote that lm EŜ = K. The man dfference between the JL and the PY statstcs s that the statstc S neglects the addtonal nverse σ 2 X M ι X n the statstc Ŝ. Thus, although these two test statstcs are derved from dfferent statstcal prncples, the fnal test statstcs are essentally testng the ndependence of u and M ι X or Eu M ι X W X M ι u = σ 2 Etr [M ι X W X M ι ] wth W = I K for the JL test and W = σ 2 X M ι X for the PY test. 3 Model and Assumptons Consder a lnear panel data model wth random coeffcents, y = X β + ɛ, 8 β = β + v, 9 for =, 2,...,, where y s a s a T vector of observatons on the dependent varable for cross-secton unt, and X s a T K matrx of possbly stochastc regressors. To smplfy the exposton we assume a balanced panel wth the same number of observaton n each panel unt see also Remark of Lemma. The vector of random coeffcents s decomposed nto a common non-stochastc vector β and a vector of ndvdual-specfc dsturbances v. Let X = [X, X 2,..., X ]. In order to construct the LM test statstc for slope homogenety we start wth model 8-9 under stylzed assumptons. However, n Secton 5 these assumptons wll be relaxed to accommodate more general and emprcally relevant setups. assumptons are mposed on the errors and the regressor matrx: The followng Assumpton The error vectors are dstrbuted as ɛ X d 0, σ 2 I T and v X d 0, Σ v, where Σ v = dag σ 2 v,,..., σ 2 v,k. The errors ɛ and v j are ndependent from each other for all and j. Assumpton 2 For the regressors we assume E x t,k 4+δ < C < for some δ > 0, for all =, 2...,, t =, 2,..., T and k =, 2..., K. The lmtng matrx lm E X X exsts and s postve defnte for all and T. 5

7 In Assumpton, the random components of the slope parameters are allowed to have dfferent varances but we assume that there s no correlaton among the elements of v. ote that ths framework s more general than the one consdered by Juhl and Lugovskyy 204 who assume Ev v = σ 2 vi K. The latter assumpton seems less appealng f there are szable dfferences n the magntudes of the coeffcents. Furthermore, the power of the test depends on the scalng of regressors, whereas the local power of our test s nvarant to a rescalng of the regressors see Theorem 5. The alternatve hypothess can be further generalzed by allowng for a correlaton among the elements of the error vector v. However, ths would ncrease the dmenson of the null hypothess to KK+/2 restrctons and t s therefore not clear whether accountng for the covarances helps to ncrease the power of the test. Obvously, f all varances are zero, then the covarances are zero as well. 2 Let u = X v + ɛ. Stackng observatons wth respect to yelds y = Xβ + u, 0 where y = y,..., y and u = u,..., u. The T T covarance matrx of u s gven by Ω E [uu X] = X Σ v X + σ 2 I T X Σ v X + σ2 I T The hypothess of fxed homogeneous slope coeffcents, β = β for all, corresponds to testng aganst the alternatve H 0 : σ 2 v,k = 0, for k =,..., K, H :. K σv,k 2 > 0, k= that s, under the alternatve at least one of the varance parameters s larger than zero. 4 The LM Test for Slope Homogenety Let θ = σ 2 v,,..., σ 2 v,k, σ2. Under Assumpton the correspondng log-lkelhood functon results as l β, θ = T 2 log2π 2 log Ω θ 2 y Xβ Ω θ y Xβ. 2 2 We also conducted Monte Carlo smulatons allowng for non-zero dagonal elements n the matrx Σ v. We found that the results are qute smlar to the settng where Σ v s dagonal. 6

8 The restrcted ML estmator of β under the null hypothess concdes wth the pooled OLS estmator β = X X X y and the correspondng resdual vector and estmated resdual varance are denoted by ũ = y X β and σ 2. The followng lemma presents the score and the nformaton matrx derved from the log-lkelhood functon n 2. Lemma The score vector evaluated under the null hypothess s gven by S l θ = H0 2 σ 4 = = ũ X X ũ σ 2 X X ũ X K X K where X k s the k-th column of X for k =, 2,..., K.. ũ σ 2 X K The nformaton matrx evaluated under the null hypothess s 0 X K, 3 [ ] I σ 2 2 l E θ θ H0 = 2 σ 4 = = X X X 2 X 2 2 = = X X X K X = = X K X K X K X K 2 X X 2 X 2 X 2 2 X K X K X X X K X K T., 4 where X k denotes the k-th column of the T K matrx X, k =, 2,..., K and =,...,. Remark It s straghtforward to extend Lemma to unbalanced panel data, where observatons are assumed to be mssng at random. Let X be a T K matrx and ũ be a conformable T vector. The score vector s gven by S = 2 σ 4 = = ũ X X ũ σ 2 X X ũ X K X K. ũ σ 2 X K 0 X K, 7

9 where σ 2 = T = = The nformaton matrx s computed accordngly. ũ ũ. Remark 2 If ndvdual-specfc constants α are ncluded n the regresson, then a condtonal verson of the test s avalable cf. Juhl and Lugovskyy 204. The ndvdual effects can be condtoned out by consderng the transformed regresson M ι y = M ι X β + M ι u, 5 wth M ι as defned n Secton 2. The typcal elements of the correspondng score vector result as ũ σ M 4 ι X j X j M ι û σ 2 X j M ι X j, j =,..., K, where ũ = M ι y M ι X β and β s the pooled OLS estmator of the transformed model 5, and σ 2 s the correspondng estmated resdual varance. It follows that we just have to replace the vector X j by the mean-adjusted vector M ι X j n Theorem. Remark 3 It s easy to see that under the more restrctve alternatve Ev v = σ 2 vi K of Juhl and Lugovskyy 204, where σ 2 v, = = σ 2 v,k = σ2 v, the score s smply the sum of all elements of S. Remark 4 otce also that the LM-type statstcs do not requre the restrcton K < T, whch s mportant for the PY approach. Ths s of course not an ssue for the asymptotc framework, where T, however, t can be a substantve restrcton n many emprcal applcatons when T s small. In the followng theorem t s shown that when T s fxed, the LM statstc possesses a χ 2 lmtng null dstrbuton wth K degrees of freedom as. Theorem Under Assumptons, 2 and the null hypothess LM = S I σ 2 S = s Ṽ s d χ 2 K, 6 as and T s fxed, where s s defned as the K vector wth typcal element s k = 2 σ 4 T 2 ũ t x t,k 2 σ 2 = t= = T x 2 t,k, 7 t= 8

10 and the k, l element of the matrx Ṽ s gven by Remark 5 Ṽ k,l = 2 σ 4 T = t= x t,k x t,l 2 T = T t= x 2 t,k = T t,l x 2. 8 If T s fxed, normalty of the regresson dsturbances s requred. If we relax the normalty assumpton, an addtonal term enters the varance of the score vector and the nformaton matrx becomes an nconsstent estmator. Theorem 2 dscusses ths ssue n more detals and derves the asymptotc dstrbuton of the LM test f the errors are not normally dstrbuted. Remark 6 It may be of nterest to restrct attenton to a subset of coeffcents. For example, n the classcal panel data model t s assumed that the constants are ndvdualspecfc and, therefore, the respectve parameters are not ncluded n the null hypothess. Another possblty s that a subset of coeffcents s assumed to be constant across all panel unts. To account for such specfcatons the model s parttoned as y t = β X a t + β 2X b t + β 3X c t + u t. t= The K vector X a t ncludes all regressors that are assumed to have ndvdual-specfc coeffcents stacked n the vector β. The K 2 vector X b t comprses all regressors that are supposed to have homogenous coeffcents. The null hypothess s that the coeffcent vector β 3 attached to the K 3 vector of regressors X c t s dentcal for all panel unts, that s, β 3 = β 3 for all, where β 3 = β 3 + v 3. The null hypothess mples Σ v3 = 0. Let X a 0 0 X b X c 0 X Z = 2 a 0 X2 b X c , 0 0 X a Xb Xc where X a = [X a,..., X a T ] and the matrces X b and X c are defned accordngly. The resduals are obtaned as ũ = I ZZ Z Z y and the columns of the matrx X c are used to compute the LM statstc. Some cauton s requred f a set of ndvdualspecfc coeffcents are ncluded n the panel regresson snce n ths case the ML estmator σ 2 = T = T t= ũ2 t s nconsstent for fxed T and. Ths mples that the expectaton of the score vector 3 s dfferent from zero. Accordngly, the unbased estmator σ 2 = T K K 2 K 3 T ũ 2 t 9 = t= 9

11 must be employed. As a specal case, assume that the constant s ncluded n X c, whereas all other regressors are ncluded n the matrx X b, and X a s dropped. Ths case s equvalent to the test for random ndvdual effects as suggested by Breusch and Pagan 980. The LM statstc then reduces to LM = T [ ũ I ι T ι T ũ ] 2, 2 T ũ ũ where ι T s a T vector of ones, whch s dentcal to the famlar LM statstc for random ndvdual effects. 5 Varants of the LM Test In ths secton we generalzed the LM test statstc by allowng for non-normally dstrbuted, heteroskedastc and serally dependent errors. Frst we show n Secton 5 that the proposed LM test s robust aganst non-normally dstrbuted errors once we assume, T jontly and specfc restrctons on the exstence of hgher-order moments. Moreover, the varants of the test wth non-normally dstrbuted errors are proposed for the settngs when and T s fxed. Second, n Secton 5.2 we propose a varant of the LM test that s robust to heteroskedastc errors. Fnally, Secton 5.3 dscusses how to robustfy the LM test, when the errors are serally correlated. 5. The LM statstc under non-normalty In ths secton we consder useful varants of the orgnal LM statstc under the assumpton that the errors are not normally dstrbuted. Therefore, we replace Assumptons and 2 by: Assumpton ɛ t s ndependently and dentcally dstrbuted wth Eɛ t X = 0, Eɛ 2 t X = σ 2 and E ɛ t 6 X < C < for all and t. Furthermore, ɛ t and ɛ js are ndependently dstrbuted for j and t s. Assumpton 2 For the regressors we assume E x t,k 6 < C < for some δ > 0, for all T =, 2...,, t =, 2,..., T and k =, 2..., K. Further, lm E [x t x t] tend to a postve defnte matrx Q and the lmtng matrx Q := exsts and s postve defnte. T T t= lm T,T = t= T E [x t x t] Assumpton 3 The error vector v s ndependently and dentcally dstrbuted wth Ev X = 0, Ev v X = Σ v, where Σ v = dag σv,, 2..., σv,k 2 and E vk 2+δ X < C < for some δ > 0, for all and k =,..., K. Further, v and ɛ j are ndependent from each other for all and j. 0

12 otce that, as n Secton 3 under the null hypothess Σ v or v = 0 for all. Hence, Assumpton 3 s not requred for the dervaton of the asymptotc null dstrbuton. To study the behavour of the LM test statstc under local alternatves, Assumpton 3 wll be used n Secton 6. Wth these modfcatons of the prevous setup, the lmtng dstrbuton of the LM statstc s gven n Theorem 2 Under Assumptons, 2 and the null hypothess, LM d χ 2 K, 20 as, T jontly. Generalzng the model to allow for non-normally dstrbuted errors ntroduces a new term nto the varance of the score: the k, l element of the covarance matrx now becomes see equaton 49 n appendx A.2 V k,l + µ 4 u 3σ 4 2σ 4 2 = T x 2 t,k T t= = T t= x 2 t,k x 2 t,l T = T t= x 2 t,l, 2 where µ 4 u denotes the fourth moment of the error dstrbuton, and V k,l s as n 8 wth σ 4 replaced by σ 4. The addtonal term depends on the excess kurtoss µ 4 u 3σ 4. Clearly, for normally dstrbuted errors, ths term dsappears, but t devates from zero n the more general setup. Under Assumptons and 2, the frst term V k,l s of order T 2, whle the new component s of order T, such that, when the approprate scalng underlyng the LM statstc s adopted, t vanshes as T. Therefore, the LM statstc as presented n the prevous secton contnues to be χ 2 K dstrbuted asymptotcally. By ncorporatng a sutable estmator of the second term n 2, however, a test statstc becomes avalable that s vald n a framework wth non-normally dstrbuted errors as, whether T s fxed or T. Therefore, denote the adjusted LM statstc by LM adj = s Ṽ adj s, where Ṽadj s as n 2 wth V k,l, σ 4 and µ 4 u defned n 8, σ 4 and µ 4 u replaced by the consstent estmators Ṽk,l = T T = t= ũ4 t for k, l =,..., K. As a consequence of Theorem 2 and the precedng dscusson, we obtan the followng result. Corollary Under Assumptons, 2 and the null hypothess LM adj d χ 2 K,

13 as and T s fxed. Furthermore, LM adj LM p 0, as, T jontly. As mentoned above, once the regresson dsturbances are no longer normally dstrbuted, the fourth moments of the error dstrbuton enter the varance of the score. It s nsghtful to dentfy exactly whch terms gve rse to ths new form of the covarance matrx. Accordng to Lemma, the contrbuton of the -th panel unt to the k-th element of the score vector s ũ X k X k ũ σ 2 X k X k = T x 2 ũ2 t,k t σ 2 + t= T ũ t ũ s x t,k x s,k. 22 t= s t The varance of the frst term on the rght hand sde depends on the fourth moments of the errors. Snce the contrbuton of ths term vanshes f T gets large, t can be dropped wthout any severe effect on the power whenever T s suffcently large. Hence, we consder a modfed score vector as presented n the followng theorem. Theorem 3 Under Assumptons, 2 and the null hypothess, the modfed LM statstc LM = s Ṽ s d χ 2 K, as and T fxed, where s s K vector wth contrbutons for panel unt s,k = σ 4 T t ũ t ũ s x t,k x s,k, 23 t=2 s= for =,...,, k =,..., K, and the k, l element of Ṽ s gven by for k, l =,..., K. Ṽ k,l = σ 4 = T t x t,k x t,l x s,k x s,l, 24 t=2 s= Remark 7 It s mportant to note that ths verson of the LM test s nvald f the panel regresson allows for ndvdual-specfc coeffcents cf. Remark 3. Consder for example the regresson y t = α + x tβ + u t 25 2

14 where α are fxed ndvdual effects and we are nterested n testng H 0 : varβ = 0. The resduals are obtaned as ũ t = y t y x t x β = ut u x t x β β. It follows that n ths case Eũ t ũ s x t,k x s,k 0 and, therefore, the modfed scores 23 result n a based test. To sdestep ths dffculty, orthogonal devatons e.g. Arellano and Bover 995 can be employed to elmnate the ndvdual-specfc constants yeldng yt = β x t + u t t = 2, 3,..., T, [ t wth yt = y t t ] y s, t t where x t and u t are defned analogously. It s well known that f u t s..d. so s u t. It follows that the modfed LM statstc can be constructed by usng the OLS resduals ũ t nstead of ũ t. Ths approach can be generalzed to arbtrary ndvdual-specfc regressors x a t. Let X a = [x a,..., x a T ] denote the ndvdual-specfc T K regressor matrx n the regresson s= y = X a β + X b β 2 + X c β 3 + u, 26 see Remark 3. Furthermore, let M a = I T X a X a X a X a, and let M a denote the T K T matrx that results from elmnatng the last K rows from M a such that M a M a s of full rank. The model 26 s transformed as y = X b β 2 + X c β 3 + u, 27 where y = Ξ a y and Ξ a = M a a M /2 a M. It s not dffcult to see that Eu u = σ 2 I T K and, thus, the modfed scores 23 can be constructed by usng the resduals of 27, where the tme seres dmenson reduces to T K. ote that orthogonal devatons result from lettng X a be a vector of ones. To revew the results of ths secton, the mportant new feature n the model wthout assumng normalty s that the fourth moments of the errors enter the varance of the score. The nformaton matrx of the orgnal LM test derved under normalty does not ncorporate hgher order moments, but the test remans applcable as T. To apply the LM test n the orgnal framework when T s fxed and errors are no longer normal we can proceed n two ways. A drect adjustment of the nformaton matrx to account for 3

15 hgher order moments yelds a vald test. Alternatvely, we can adjust the score tself and restrct attenton to that part of the score that does not ntroduce hgher order moments nto the varance. In the next secton, we further pursue the second route of dealng wth non-normalty and thereby robustfy the test aganst heteroskedastcty and seral correlaton. 5.2 The regresson-based LM statstc In ths secton we offer a convenent way to compute the proposed LM statstc va a smple artfcal regresson. Moreover, the regresson-based form of the LM test s shown to be robust aganst heteroskedastc errors. Followng the decomposton of the score contrbuton n 22 and the dscusson thereafter, we construct the Outer Product of Gradents OPG varant of the LM test based on the second term n 22. Rewrtng the correspondng elements of the score contrbutons of panel unt as s,k = T t ũ t ũ s x t,k x s,k, 28 t=2 s= for k =,..., K. ote that we dropped the factor / σ 4 as ths factor cancels out n the fnal test statstc. Ths gves the usual LM-OPG varant LM opg = = s = s s = s, 29 where s = [ s,,..., s,k]. An asymptotcally equvalent form of the LM-OPG statstc can be formulated as a Wald-type test for the null hypothess ϕ = 0 n the auxlary regresson where ũ t = K z t,k ϕ k + e t, for =,...,, t =,..., T 30 k= t z t,k = x t,k ũ s x s,k for k =,..., K. Therefore, wth the Ecker-Whte heteroskedastcty-consstent varance s= estmator, the regresson based test statstc results as LM reg = = T ũ t z t t=2 = T t=2 ũ 2 t z t z t = T ũ t z t, 3 It follows from the arguments smlar as n Theorem 3 that M reg test statstc s asymptotcally χ 2 dstrbuted but t turns out to be robust aganst heteroskedastcty: t=2 4

16 Corollary 2 Under Assumpton but allowng for heteroscedastc errors such that E[ɛ 2 t X] = σ 2 t < C <, Assumpton 2 and the null hypothess LM reg d χ 2 K, 32 as and T s fxed. It s mportant to note that the LM-OPG varant cannot be appled to resduals from a fxed effect regresson, see Remark 7. Furthermore, the replacement of the resduals by orthogonal forward devaton wll not fx ths problem snce orthogonal forward devatons are no longer serally uncorrelated f the errors are heteroskedastc. Therefore, a verson of the test s requred that s robust aganst autocorrelated errors. 5.3 The LM statstc under serally dependent errors In ths secton we propose a varant of the LM test statstc that accommodates serally correlated errors, that s, we relax Assumptons as follows: Assumpton The T error vector ɛ s ndependently and dentcally dstrbuted wth Eɛ X = 0, Eɛ ɛ X = Eɛ ɛ = Σ and E [ ɛ t 4+δ X ] < C < for some δ > 0 and all and t. The T T matrx Σ s postve defnte wth typcal element σ ts for t, s =,..., T. ote that Assumpton allows for heteroscedastcty and seral dependence across tme, however, t restrcts the error vector ɛ to be d across ndvduals. Under ths assumpton the expectaton of the score vector 23 s under the null hypothess E[u t u s x t,k, x s,k ] = σ ts E[x t,k x s,k ]. We therefore suggest a modfcaton for autocorrelated errors based on the adjusted K score vector s wth typcal element s k s,k = = = s,k for k =,..., K and T t ũ t ũ s σ ts x t,k x s,k, 33 t=2 s= where σ ts = = ũtũ s. The asymptotc propertes of the LM statstc based on the modfed score vector are presented n Theorem 4 Let LM ac = s Ṽ s, 5

17 where Ṽ s a K K matrx wth typcal element and Ṽ k,l = = T t=2 T t t δ tsτq x t,k x s,k x τ,l x q,l 34 τ=2 s= q= δtsτq = ũ jt ũ js ũ jτ ũ jq σ ts σ τq. j= Under Assumptons, 2, the null hypothess 2 and as wth T fxed the LM ac statstc has a χ 2 K lmtng dstrbuton. ote that ths verson of the test has a good sze control rrespectve of seral dependence n errors. However, the test nvolves some power loss relatve to the orgnal test statstcs when errors are serally uncorrelated, whch s not surprsng gven a more general setup of ths varant of the test. The respectve asymptotc power results are analyzed n the next secton see Remark 0. Secton 7 elaborates n detal on the sze-power propertes of the LM ac n fnte samples. 6 Local Power The am of ths secton s twofold. Frst, we nvestgate the dstrbutons of the LM-type test under sutable sequences of local alternatves. Two cases are of nterest, wth T fxed and, T jontly, whch are presented n the respectve theorems below. Second, we adopt the results of PY to our model n order to compare the local asymptotc power of the two tests. To formulate an approprate sequence of local alternatves, we specfy the random coeffcents n 9 n a setup n whch T s fxed. The error term v s as n Assumpton wth elements of Σ v gven by σ 2 v,k = c k, 35 where c k > 0 are fxed constants for k =,..., K. The asymptotc dstrbuton of the LM statstc results as follows. Theorem 5 Under Assumptons, 2 and the sequence of local alternatves 35, LM d χ 2 K µ, as and T fxed, wth non-centralty parameter µ = c Ψc, where c = c,..., c K 6

18 and Ψ s a K K matrx wth k, l element Ψ k,l = 2σ plm 4 T 2 x t,k x t,l T = t= = T t= x 2 t,k T t,l x 2. = t= In order to relax the assumpton of normally dstrbuted errors we adopt Assumpton for v, where the sequence of local alternatves s now gven by σ 2 v,k = c k T, 36 for k =,..., K. ote that accordng to Theorem 2 we requre T. Theorem 6 Under Assumptons, 2, 3 and the sequence of alternatves 36, LM d χ 2 K µ, as, T, wth non-centralty parameter µ = c Ψc, where c = c,..., c K and Ψ s a K K matrx wth k, l element Remark 8 Ψ k,l = 2σ 4 plm,t T = 2 T x t,k x t,l. As n Secton 5. above, when the normalty assumpton s relaxed, local power can be studed for LM under Assumptons, 2 and 3 when T s fxed. specfcaton of local alternatves as n Theorem 5 apples. The non-centralty parameter of the lmtng non-central χ 2 dstrbuton results as µ = c Ψ c wth t= The Ψ k,l = σ 4 plm T t x t,k x t,l x s,k x s,l, = t=2 s= for k, l =,..., K. Remark 9 Gven the results for the modfed statstc LM n remark 8, and the fact that s = = s = Z ũ, we expect a smlar result for the regresson-based LM statstc LM reg to hold. Recall that LM uses Ṽ as an estmator of the varance of s see 24, whle LM reg employs T = t=2 ũ2 t z t z t. Under the null hypothess, t s not dffcult to see that these two estmators are asymptotcally equvalent. Under the alternatve, when studyng the k, l element of the varance of LM reg, we obtan see 7

19 appendx A.2 for detals = T ũ 2 t z t,k z t,l = t=2 + T t t ɛ 2 tx t,k x t,l ɛ s x s,k ɛ s x s,l = t=2 = t=2 s= s= T ɛ 2 tv B t X v + o p, 37 wth the K K matrx Bt X = x t t,k s= x t sx s,k xt,l s= x sx s,l. The frst term on the rght-hand sde n 37 has the same probablty lmt as Ṽk,l, the lmtng covarance matrx element Ψ k,l. In contrast to LM, however, the varance estmator of the regresson-based test nvolves addtonal quadratc forms such as v B X t v, contrbutng to the estmator. Snce, n a setup wth fxed T and the local alternatves σv,k 2 = c k, T ɛ 2 tv B t X v = O p /2, = t=2 the varance estmator remans consstent. In small samples, however, the addtonal term results n a bas of the varance estmator and may deterorate the power of the regresson-based test. See the appendx for detals about the above result and the Monte Carlo experments n Secton 7. Remark 0 The arguments of Remark 8 can be used to derve the local power of the LM ac statstc that accounts for seral correlaton n errors. The same specfcaton of local alternatves apples. The non-centralty parameter of the lmtng non-central χ 2 dstrbuton takes the quadratc form µ = c Ψ c wth Ψ k,l = plm T t T t u t u s u τ u q σ ts σ τq x t,k x s,k x τ,l x q,l, = t=2 s= τ=2 q= for k, l =,..., K. In the absence of seral correlaton t can be shown that the LM ac test nvolve a loss of power. To llustrate ths fact assume for smplcty that K = sngle regressor case. Further, the score vector n 33 can be equvalently wrtten as ŝ = T t ũ t ũ s σ ts x t,k x s,k = = t=2 s= T t ũ t ũ s xt,k x s,k C ts, 38 = t=2 s= where C ts = = x tx s. Thus, demeanng of ũ t ũ s s equvalent wth demeanng of x t,k x s,k. In the case of no autocorrelaton and 38 t follows that Ψ Ψ s postve sem-defnte. Therefore, the modfcaton 33 tends to reduce the power of the LM ac test when compared to LM. 8

20 We now proceed to examne the local power of the statstc of PY n model 8 and 9 under the sequence of local alternatves 36. In our homoskedastc setup, the dsperson statstc becomes S = β β X X β β, σ 2 = wth β as the OLS estmator n 0 as above. Usng ths expresson, the statstc s computed as n 4. The next theorem presents the asymptotc dstrbuton of the statstc under the local alternatves as specfed above. Ths result follows drectly from Secton 3.2 n PY. Theorem 7 Under Assumptons, 2, 3 and the sequence of local alternatves 36 d λ,, as, T, provded /T 0, where λ = Λ c/ 2K and Λ s a K vector wth typcal element Λ k = σ 2 plm,t T T = t= x 2 t,k, for k =,..., K. In Theorem 7, the mean of the lmtng dstrbuton of s slghtly dfferent from the result n Secton 3.2 n PY. Here, v s random and ndependently dstrbuted from the regressors and, therefore, the second term of the respectve expresson n PY s zero. Remark Consder for smplcty a scalar regressor x t that s..d. across and t wth unformly bounded fourth moments. Let E [x t ] = 0 and E [x 2 t] = σ 2,x, that s, the regressor s assumed to have a unt-specfc varaton whch s constant over tme for a gven unt. We obtan E T T t= x 2 t 2 = σ 2,x 2 + O T, mplyng µ = c 2 /2σ 4 lm = σ 2,x 2 n Theorem 6. To gan further nsght, we thnk of σ 2,x 2 as beng randomly dstrbuted n the cross-secton such that the noncentralty parameter results as [ µ = c2 σ 2σ E ] 2 2 4,x = c2 V ar [ ] [ ] σ 2 2σ 4,x + E σ 2 2,x. 39 9

21 Smlarly, under these assumptons, we fnd λ = c σ 2 2 E [ ] σ,x Comparng the mean of the normal dstrbuton of the statstc n 40 wth the noncentralty parameter of the asymptotc χ 2 dstrbuton of the LM statstc n 39, we see that the man dfference between the two tests s that the varance of σ 2,x contrbutes to the power of the LM statstc but not to the power of the test. If V ar [ σ 2,x] = 0 such that σ 2,x = σ 2 x for all, the LM test and the test have the same asymptotc power n ths example. If, however, V ar [ σ 2,x] > 0, so that there s varaton n the varance of the regressor n the cross-secton, the LM test has larger asymptotc power. To llustrate ths pont, we examne the local asymptotc power functons of the LM and the test for two cases, usng the expressons n 39 and 40. Fgure see appendx C shows the local asymptotc power of the LM sold lne and the test dashed lne as a functon of c when σ 2,x has a χ 2 dstrbuton. Fgure 2 repeats ths exercse for σ 2,x drawn from a χ 2 2 dstrbuton. In both cases, the LM test has larger asymptotc power. The power gan s substantal for the frst case, but dmnshes for the second. Ths pattern s expected, as the varance of σ 2,x contrbutes relatvely more to the non-centralty parameter n the frst specfcaton. Ths dscusson exemplfes the dfference between the LM-type tests and the statstc n terms of the local asymptotc power n a smplfed framework. The analyss suggests that the LM-type tests are partcularly powerful n an emprcally relevant settng n whch there s non-neglgble varaton n the varances of the regressors between panel unts. Havng studed the large samples propertes of the LM tests under the null and the alternatve hypothess n our model, we now evaluate the fnte-sample sze and power propertes of the LM-type tests n a Monte Carlo experment. 7 Monte Carlo Experments 7. Desgn After dervng LM-type tests n the random coeffcent model, we now turn to study the small-sample propertes of the proposed test and t varants. The am of ths secton s to evaluate the performance of the tests n terms of ther emprcal sze and power n several dfferent setups, relatng to the theoretcal dscusson of Sectons 4-6. We consder the followng test statstcs: the orgnal LM statstc presented n Theorem, the adjusted LM statstc that adjusts the nformaton matrx to account for fourth moments of the error dstrbuton see Corollary, the score-modfed LM statstcs see Theorem 3 and Theorem 4 and the regresson-based, heteroskedastcty-robust LM statstc see Secton 20

22 5.2. As a benchmark, we consder PY s statstc adj gven n 5. Followng the notes n Table n PY, the test usng adj s carred out as a two-sded test. In addton, the CLM test n 7 s ncluded, whch s also a two-sded test. We consder the followng data-generatng process wth normally dstrbuted errors as the standard desgn: y t = α + x tβ + ɛ t, ɛ t α d 0,, 4 d 0, 0.25, x t,k = α + ϑ x t,k, k =, 2, 3, ϑ x t,k β d 0, σ 2 x,k, d 3 ι 3, Σ v, under the null hypothess: Σ v = under the alternatve: Σ v = , where =, 2,...,, t =, 2,..., T. Hence, to smulate a model under the null the slope vector β s generated as a 3 vector of ones ι 3 for all. As dscussed n Secton 6 the varances of the regressors play an mportant role. In our benchmark specfcaton we generate the varances as σ 2 x,k = η,k η,k d χ 2, 44 The choce of the χ 2 dstrbuton for σx,k 2 s made analogous to the Monte Carlo experment n PY. We then consder varatons of ths specfcaton below. All results are based on 5,000 Monte Carlo replcatons. We choose {0, 20, 30, 50, 00, 200}, T {0, 20, 30}, as we would lke to study the small sample propertes of the test procedures when the tme dmenson s small. In our frst set of Monte Carlo experments the errors are normally dstrbuted; therefore we focus on the standard LM test. We also nclude ther respectve heteroskedastcty-robust regresson varants for ths exercse. 2

23 7.2 ormally dstrbuted errors Panel A of Table see Appendx B shows the rejecton frequences when the null hypothess s true. The adj test has rejecton frequences close to the nomnal sze of 5% for all combnatons of and T, whle the CLM test rejects the null hypothess too often, n partcular for small. Devatons from the nomnal sze for the the standard LM test and the regresson-based test are small and dsappear as ncreases, as expected from Theorem. Panel B of Table shows the correspondng rejectons frequences under the alternatve hypothess. The LM test outperforms the adj and the CLM test n general. Ths observaton holds n partcular for T = 0 where the power gan s consderable. The LM reg varant, although as powerful as the adj test for T = 0, suffers from a power loss relatve to the standard LM test. Ths power loss may be due to the small sample bas of the varance estmator, see Remark 9. Followng Remark 7 the varants of the LM tests are computed as follows. Frst, the ndvdual-specfc fxed effects α are elmnated by transformng the data usng orthogonal forward devatons see Arellano and Bover 995. The LM statstcs are then computed usng the transformed data. The results presented n Panel A of Table 2 ndcate that by employng forward orthogonalzaton all varants of the LM test have sze reasonably close to the nomnal level. By comparng panel B of Table and the rejecton rates under the alternatve n panel B of Table 2 we see that the power s very smlar n both setups confrmng usefulness of the forward orthogonalzaton procedure for the LM tests. 7.3 on-normal errors We now nvestgate the LM test when the errors are no longer normally dstrbuted, thereby buldng on the results of Secton 5.. The errors n 4 are generated from a t-dstrbuton wth 5 degrees of freedom, scaled to have unt varance. All other specfcatons of the standard desgn reman unchanged. In addton to the statstcs already consdered, we now nclude the adjusted LM statstc see corollary and the scoremodfed statstc see Theorem 3. Panel A n Table 3 reports the rejecton frequences under the null hypothess n ths case. We notce that the LM test has substantal sze dstortons when T s fxed and ncreases, whch s expected from Theorem 2. However, the adjusted LM statstc LM adj and the modfed score statstc LM are both successful n controllng the type-i error. Panel B of Table 3 shows rejecton frequences under the alternatve hypothess. The power gan of the LM test relatve to the adj test s notceable when T = 0 or T = 20. We found smlar results when the errors are χ 2 dstrbuted wth two degrees of freedom, centered and standardzed to have mean zero and varance equal to one. Gven the smlarty of the results for t and χ 2 dstrbuted errors, we do not present the latter results. 22

24 7.4 Serally correlated errors To study the mpact of serally correlated errors on the test statstcs we adjust the DGP as follows: y t = x tβ + ɛ t, ɛ t = ρɛ t + ρ 2 /2 et, d for =, 2,...,, t =, 2,..., T, where e t 0,. Under the null hypothess β = for all whle under the alternatve β s generated as n 43. The regressors, x t,k, k =, 2, 3 are generated as x t,k = φ,k x t,k + φ 2,k /2 ϑ x t,k, φ,k ϑ x t,k d U[0.05, 0.95], d 0, σ 2 x,k, where σ 2 x,k = η,k wth η,k d χ 2. Parameters φ,k and σ x,k are fxed across replcatons. Results of ths smulaton experment are reported n Table 4. Panel A and B show the rejecton frequences under the null hypothess n case of small seral dependence.e., ρ = 0.2, Panel A and moderate dependence.e., ρ = 0.5, Panel B. For all LM based test statstcs, except the LM ac test, we observe substantal sze devatons from the nomnal level. However, the LM ac test s successful n controllng the type-i error. Further, sze propertes of PY test are also sgnfcantly affected by autocorrelated errors. ote that ths fact s already documented and studed n Blomqust and Westerlund 203. Panel C of Table 4 reports power propertes of the test under no seral correlaton.e., ρ = 0, buldng on the dscusson n Remark 0. We observe that the LM ac test nvolve a 5 0% power loss compared to the LM test. Ths relatve power loss des out f T ncreases. 8 Concludng remarks In ths paper we examne the problem of testng slope homogenety n a panel data model. We develop testng procedures usng the LM prncple. Several varants are consdered that robustfy the orgnal LM test wth respect to non-normalty, heteroscedastcty and serally correlated errors. By studyng the local power we dentfy cases where the LMtype tests are partcularly powerful relatve to exstng tests. In sum, our Monte Carlo experments suggest that the LM test are powerful testng procedures to detect slope 23

25 homogenety n short panels n whch the tme dmenson s small relatve to the crosssecton dmenson. The LM approach suggested n ths paper may be extended n future research by allowng for dynamc specfcatons wth lagged dependent varables and cross sectonally or serally correlated errors. References Arellano, M. and O. Bover 995. Another look at the nstrumental varable estmaton of error-components models. Journal of Econometrcs 68, Baltag, B., Q. Feng, and C. Kao 20. Testng for sphercty n a fxed effects panel data model. The Econometrcs Journal 4, Blomqust, J. and J. Westerlund 203. Testng slope homogenety n large panels wth seral correlaton. Economcs Letters 2 3, Breusch, T. S. and A. R. Pagan 980. The Lagrange multpler test and ts applcatons to model specfcaton n econometrcs. Revew of Economc Studes 47, Harvlle, D. A Maxmum lkelhood approaches to varance component estmaton and to related problems. Journal of the Amercan Statstcal Assocaton 72, Honda, Y Testng the error components model wth non-normal dsturbances. Revew of Economc Studes 52, Hsao, C. and M. H. Pesaran Random coeffcent models. In L. Mátyás and P. Sevestre Eds., The Econometrcs of Panel Data, Chapter 6. Sprnger. Juhl, T. and O. Lugovskyy 204. A test for slope homogenety n fxed effects models. Econometrc Revews 33, Pesaran, M. H. and R. Smth 995. Estmatng long-run relatonshps from dynamc heterogenous panels. Journal of Econometrcs 68, Pesaran, M. H. and T. Yamagata Testng slope homogenety n large panels. Journal of Econometrcs 42, Phllps, P. C. B. and H. R. Moon 999. Lnear regresson lmt theory for nonstatonary panel data. Econometrca 67 5, 057. Swamy, P. A. V. B Effcent nference n a random coeffcent regresson model. Econometrca 38, Ullah, A Fnte-sample econometrcs. Oxford Unversty Press. Wand, M. P Vector dfferental calculus n statstcs. The Amercan Statstcan 56, 8. 24

26 Whte, H Asymptotc theory for econometrcans. Emerald. Wens, D. P On moments of quadratc forms n non-sphercally dstrbuted varables. Statstcs 23 3,

27 A Appendx: Proofs nstead of full expressons and T through- To economze on notaton we use out ths appendx. and t = t= A. Prelmnary results We frst present an mportant result concernng the asymptotc effect of the estmaton error β β on the test statstcs. Defne A k = X k X k X k X k I T. T X A k u for k =,..., K. Fur- Lemma A. Let R k XAX = thermore let R k = σ 4 σ 4 X A k X and R k XAu = k k β β R 2σ 2 XAX β β 2 β β R XAu, for k =,..., K. Under Assumptons, 2 and the null hypothess the followng propertes hold f T s fxed: R k XAX = O p, R k XAu = O p /2, R k = O p, for k =,..., K. Proof. Usng the defnton of A j R k XAX = X X k X k yelds X T t x 2 t,k X X. The frst term s a K K matrx wth typcal l, m element x t,l x t,k x t,m x t,k = O p, t t as a consequence of Assumpton 2, whle t x2 t,k /T = O p and X X = O p. Recall that under the null hypothess, u = ɛ. Thus R k XAu = X X k X k u T t x 2 t,k X u. 26

28 The frst and the second term are O p /2 by a the central lmt theorem CLT for ndependent random varables and Assumpton 2. Combnng and together wth the fact that β β = O p yelds the result. Lemma A.2 Under Assumptons, 2 and the null hypothess the followng propertes hold for and T : R k XAX = O p T 2, R k XAu = O p /2 T 3/2, R k T = O p T, whch s defned as R k for k =,..., K. n Lemma A., Proof. Followng the proof of Lemma A. the element of the frst term of R k XAX s O p T 2, whereas the second term s O p T by Assumpton 2 whch yelds statement. otce n R k XAu has two terms as n Lemma A., where the frst one has zero mean and varance of order T 3. Therefore by Lemma n Baltag, Feng, and Kao 20 we have that X X j X j u = O p T 3/2 and by Lemma 2 n PY that X X j X j u = O p /2 T 3/2 and X u = O p /2 T /2. These results and the fact that T β β = O p mply. A.2 Proofs of the man results Proof of Lemma We use the followng rules for matrx dfferentatons: l = [ ] θ k 2 tr Ω Ω + θ k 2 [ ] l E = 2 [ Ω θ k θ tr Ω l θ k [ u Ω Ω Ω Ω θ l ] Ω u, 45 θ ] k, 46 for k, l =, 2,..., K +, see, e.g., Harvlle 977 and Wand Frst, X Σ v X = k σv,kx 2 k X k, wth X k denotng the k-th column vector of X. Hence X Σ v X 0... = k 0 X Σ v X σ 2 v,ka k, 27

29 wth the T T matrx, A k = X k X k X k Xk, for k =,..., K, and X k denotes the k-th column of the T K matrx X. Thus, Ω = k σ 2 v,ka k + σ 2 I T and { Ω A k, for k =, 2,..., K, = θ k I T, for k = K +. Under the null hypothess we have Ω = σ 2 I T. Usng 45 we obtan { l tr [A 2 σ θ k = 2 k ] + ũ A 2 σ 4 k ũ, for k =, 2,..., K H0 0, for k = K +, where σ 2 = T ũ ũ, ũ = I T X X X X y. The representaton of the score vector follows from tr [A k ] = Xt,k 2 = X k X k, t where X k denotes the k-th column of the T K matrx X. Smlarly, 46 yelds [ ] l tr [A 2σ 4 k A l ], for k, l =, 2,..., K, E = X θ k θ 2σ l H 0 k X k, for k =, 2,..., K, and l = K +, 4 T, for k = l = K +, 2σ 4 Usng the fact that A k and A l are block-dagonal, tr [A k A l ] = [ tr X k X k X l ] X l = X k X l 2, where X k denotes the -th column of X, whch yelds the form of the nformaton matrx presented n the lemma. Proof of Theorem 28

30 Recall that A k = X k X k T X k X k and rewrte the elements of the scores as 4 σ s k = ũ σ 4 2σ A k 4 ũ, for k =,..., K. Snce ũ = u X β β we have s k = σ 4 u σ 4 2σ A k 4 u + R k, where R k = O p from Lemma A.. Snce [ ] tr A k 0 and, therefore, lm E s = 0. The covarances are obtaned as [ ] Cov u A k u, u A l u X = 2σ 4 tr A k A l 2 = 2σ 4 X k X l X k X k X l T + T X k X k X l X l, T T X l I T, = 0 t follows that Eu A k u = T X l X l and snce u A k u s ndependent of u ja l u j for all j condtonal on X, 2 Cov 2σ 4 = 2σ 4 = V k,l. X k u A k u, 2 X l T u A l u X X k X k X l X l X k The Lapounov condton n the central lmt theorem for ndependent random varables see Whte 200, Theorem 5.0 s satsfed by Assumpton 2 and therefore /2 Ṽ d s 0, I K, where Ṽ replaces σ4 n V by σ 4. By the formula for the parttoned nverse { I σ 2 } :K,:K = Ṽ, X k 29

31 where { } :K,:K denotes the upper-left K K block of the matrx, t follows fnally that S I σ 2 S = s Ṽ s d χ 2 K. Proof of Theorem 2 The proof proceeds n three steps: we derve the covarance matrx of the score vector, we establsh the asymptotc normalty of the score vector and we use these results to establsh the asymptotc dstrbuton of the LM statstc. Defne the K vector s = [s,..., s K ] wth typcal element s k = u 2σ A k 4 u = s 2σ 4,k, 47 where s,k = u A k u and k K. Usng standard results for quadratc forms see e.g., Ullah 2004, appendx A.5, E [ ] [ s,k s,l X = 2σ 4 tr A k E [ ] [ s,k X = σ 2 tr ] [ A l + σ 4 tr where a k denotes the fourth moment of u t. Snce A k ] A k [ tr ] A l ] + µ 4 u 3σ 4 a k a l, s a vector consstng of the man dagonal elements of the matrx A k and µ 4 u E [ ] [ ] [ s,k X E s,l X = σ 4 tr A k ] [ tr A l ], we have Cov s,k, s,l X [ = 2σ 4 tr A k A l ] + µ 4 u 3σ 4 a k a l. 48 Due to the ndependence of u A k Cov s,k, s,l X = 2σ 4 u and u ja l u j for j, t follows that j [ tr A k A l ] + µ 4 u 3σ 4 [ Let V T denote the covarance matrx of s. Insertng the expresson for tr determne the k, l element of V T as V k,l = 2 x 2σ 4 t,k x t,l T t t µ 4 u 3σ 4 + 2σ 4 2 x 2 t,k T = V,k,l + V 2,k,l. t x 2 t,k t t,k x 2 x 2 t,k t x 2 t,l T a k a l. A k t ] A l, we To verfy that a central lmt theorem apples to s, let λ R k, λ = and Z,T = T λ s, where s s a K vector wth elements s,k for k K. Further, E [Z,T ] = 0 30 x 2 t,l 49

Econometrics of Panel Data

Econometrics of Panel Data Econometrcs of Panel Data Jakub Mućk Meetng # 8 Jakub Mućk Econometrcs of Panel Data Meetng # 8 1 / 17 Outlne 1 Heterogenety n the slope coeffcents 2 Seemngly Unrelated Regresson (SUR) 3 Swamy s random

More information

Essays on Heterogeneity and Non-Linearity in Panel Data and Time Series Models

Essays on Heterogeneity and Non-Linearity in Panel Data and Time Series Models Essays on Heterogenety and on-lnearty n Panel Data and Tme Seres Models Inaugural-Dssertaton zur Erlangung des Grades enes Doktors der Wrtschafts- und Gesellschaftswssenschaften durch de Rechts- und Staatswssenschaftlche

More information

Econ107 Applied Econometrics Topic 3: Classical Model (Studenmund, Chapter 4)

Econ107 Applied Econometrics Topic 3: Classical Model (Studenmund, Chapter 4) I. Classcal Assumptons Econ7 Appled Econometrcs Topc 3: Classcal Model (Studenmund, Chapter 4) We have defned OLS and studed some algebrac propertes of OLS. In ths topc we wll study statstcal propertes

More information

A Comparative Study for Estimation Parameters in Panel Data Model

A Comparative Study for Estimation Parameters in Panel Data Model A Comparatve Study for Estmaton Parameters n Panel Data Model Ahmed H. Youssef and Mohamed R. Abonazel hs paper examnes the panel data models when the regresson coeffcents are fxed random and mxed and

More information

Testing for seasonal unit roots in heterogeneous panels

Testing for seasonal unit roots in heterogeneous panels Testng for seasonal unt roots n heterogeneous panels Jesus Otero * Facultad de Economía Unversdad del Rosaro, Colomba Jeremy Smth Department of Economcs Unversty of arwck Monca Gulett Aston Busness School

More information

Econ Statistical Properties of the OLS estimator. Sanjaya DeSilva

Econ Statistical Properties of the OLS estimator. Sanjaya DeSilva Econ 39 - Statstcal Propertes of the OLS estmator Sanjaya DeSlva September, 008 1 Overvew Recall that the true regresson model s Y = β 0 + β 1 X + u (1) Applyng the OLS method to a sample of data, we estmate

More information

The Multiple Classical Linear Regression Model (CLRM): Specification and Assumptions. 1. Introduction

The Multiple Classical Linear Regression Model (CLRM): Specification and Assumptions. 1. Introduction ECONOMICS 5* -- NOTE (Summary) ECON 5* -- NOTE The Multple Classcal Lnear Regresson Model (CLRM): Specfcaton and Assumptons. Introducton CLRM stands for the Classcal Lnear Regresson Model. The CLRM s also

More information

3.1 Expectation of Functions of Several Random Variables. )' be a k-dimensional discrete or continuous random vector, with joint PMF p (, E X E X1 E X

3.1 Expectation of Functions of Several Random Variables. )' be a k-dimensional discrete or continuous random vector, with joint PMF p (, E X E X1 E X Statstcs 1: Probablty Theory II 37 3 EPECTATION OF SEVERAL RANDOM VARIABLES As n Probablty Theory I, the nterest n most stuatons les not on the actual dstrbuton of a random vector, but rather on a number

More information

A Monte Carlo Study for Swamy s Estimate of Random Coefficient Panel Data Model

A Monte Carlo Study for Swamy s Estimate of Random Coefficient Panel Data Model A Monte Carlo Study for Swamy s Estmate of Random Coeffcent Panel Data Model Aman Mousa, Ahmed H. Youssef and Mohamed R. Abonazel Department of Appled Statstcs and Econometrcs, Instute of Statstcal Studes

More information

LINEAR REGRESSION ANALYSIS. MODULE IX Lecture Multicollinearity

LINEAR REGRESSION ANALYSIS. MODULE IX Lecture Multicollinearity LINEAR REGRESSION ANALYSIS MODULE IX Lecture - 30 Multcollnearty Dr. Shalabh Department of Mathematcs and Statstcs Indan Insttute of Technology Kanpur 2 Remedes for multcollnearty Varous technques have

More information

Chapter 8 Indicator Variables

Chapter 8 Indicator Variables Chapter 8 Indcator Varables In general, e explanatory varables n any regresson analyss are assumed to be quanttatve n nature. For example, e varables lke temperature, dstance, age etc. are quanttatve n

More information

Lecture 4 Hypothesis Testing

Lecture 4 Hypothesis Testing Lecture 4 Hypothess Testng We may wsh to test pror hypotheses about the coeffcents we estmate. We can use the estmates to test whether the data rejects our hypothess. An example mght be that we wsh to

More information

4 Analysis of Variance (ANOVA) 5 ANOVA. 5.1 Introduction. 5.2 Fixed Effects ANOVA

4 Analysis of Variance (ANOVA) 5 ANOVA. 5.1 Introduction. 5.2 Fixed Effects ANOVA 4 Analyss of Varance (ANOVA) 5 ANOVA 51 Introducton ANOVA ANOVA s a way to estmate and test the means of multple populatons We wll start wth one-way ANOVA If the populatons ncluded n the study are selected

More information

Primer on High-Order Moment Estimators

Primer on High-Order Moment Estimators Prmer on Hgh-Order Moment Estmators Ton M. Whted July 2007 The Errors-n-Varables Model We wll start wth the classcal EIV for one msmeasured regressor. The general case s n Erckson and Whted Econometrc

More information

1. Inference on Regression Parameters a. Finding Mean, s.d and covariance amongst estimates. 2. Confidence Intervals and Working Hotelling Bands

1. Inference on Regression Parameters a. Finding Mean, s.d and covariance amongst estimates. 2. Confidence Intervals and Working Hotelling Bands Content. Inference on Regresson Parameters a. Fndng Mean, s.d and covarance amongst estmates.. Confdence Intervals and Workng Hotellng Bands 3. Cochran s Theorem 4. General Lnear Testng 5. Measures of

More information

Econ107 Applied Econometrics Topic 9: Heteroskedasticity (Studenmund, Chapter 10)

Econ107 Applied Econometrics Topic 9: Heteroskedasticity (Studenmund, Chapter 10) I. Defnton and Problems Econ7 Appled Econometrcs Topc 9: Heteroskedastcty (Studenmund, Chapter ) We now relax another classcal assumpton. Ths s a problem that arses often wth cross sectons of ndvduals,

More information

ECONOMICS 351*-A Mid-Term Exam -- Fall Term 2000 Page 1 of 13 pages. QUEEN'S UNIVERSITY AT KINGSTON Department of Economics

ECONOMICS 351*-A Mid-Term Exam -- Fall Term 2000 Page 1 of 13 pages. QUEEN'S UNIVERSITY AT KINGSTON Department of Economics ECOOMICS 35*-A Md-Term Exam -- Fall Term 000 Page of 3 pages QUEE'S UIVERSITY AT KIGSTO Department of Economcs ECOOMICS 35* - Secton A Introductory Econometrcs Fall Term 000 MID-TERM EAM ASWERS MG Abbott

More information

Chapter 11: Simple Linear Regression and Correlation

Chapter 11: Simple Linear Regression and Correlation Chapter 11: Smple Lnear Regresson and Correlaton 11-1 Emprcal Models 11-2 Smple Lnear Regresson 11-3 Propertes of the Least Squares Estmators 11-4 Hypothess Test n Smple Lnear Regresson 11-4.1 Use of t-tests

More information

Systems of Equations (SUR, GMM, and 3SLS)

Systems of Equations (SUR, GMM, and 3SLS) Lecture otes on Advanced Econometrcs Takash Yamano Fall Semester 4 Lecture 4: Sstems of Equatons (SUR, MM, and 3SLS) Seemngl Unrelated Regresson (SUR) Model Consder a set of lnear equatons: $ + ɛ $ + ɛ

More information

x i1 =1 for all i (the constant ).

x i1 =1 for all i (the constant ). Chapter 5 The Multple Regresson Model Consder an economc model where the dependent varable s a functon of K explanatory varables. The economc model has the form: y = f ( x,x,..., ) xk Approxmate ths by

More information

Durban Watson for Testing the Lack-of-Fit of Polynomial Regression Models without Replications

Durban Watson for Testing the Lack-of-Fit of Polynomial Regression Models without Replications Durban Watson for Testng the Lack-of-Ft of Polynomal Regresson Models wthout Replcatons Ruba A. Alyaf, Maha A. Omar, Abdullah A. Al-Shha ralyaf@ksu.edu.sa, maomar@ksu.edu.sa, aalshha@ksu.edu.sa Department

More information

STAT 3008 Applied Regression Analysis

STAT 3008 Applied Regression Analysis STAT 3008 Appled Regresson Analyss Tutoral : Smple Lnear Regresson LAI Chun He Department of Statstcs, The Chnese Unversty of Hong Kong 1 Model Assumpton To quantfy the relatonshp between two factors,

More information

Chapter 5 Multilevel Models

Chapter 5 Multilevel Models Chapter 5 Multlevel Models 5.1 Cross-sectonal multlevel models 5.1.1 Two-level models 5.1.2 Multple level models 5.1.3 Multple level modelng n other felds 5.2 Longtudnal multlevel models 5.2.1 Two-level

More information

Chapter 2 - The Simple Linear Regression Model S =0. e i is a random error. S β2 β. This is a minimization problem. Solution is a calculus exercise.

Chapter 2 - The Simple Linear Regression Model S =0. e i is a random error. S β2 β. This is a minimization problem. Solution is a calculus exercise. Chapter - The Smple Lnear Regresson Model The lnear regresson equaton s: where y + = β + β e for =,..., y and are observable varables e s a random error How can an estmaton rule be constructed for the

More information

Statistics for Economics & Business

Statistics for Economics & Business Statstcs for Economcs & Busness Smple Lnear Regresson Learnng Objectves In ths chapter, you learn: How to use regresson analyss to predct the value of a dependent varable based on an ndependent varable

More information

ANSWERS. Problem 1. and the moment generating function (mgf) by. defined for any real t. Use this to show that E( U) var( U)

ANSWERS. Problem 1. and the moment generating function (mgf) by. defined for any real t. Use this to show that E( U) var( U) Econ 413 Exam 13 H ANSWERS Settet er nndelt 9 deloppgaver, A,B,C, som alle anbefales å telle lkt for å gøre det ltt lettere å stå. Svar er gtt . Unfortunately, there s a prntng error n the hnt of

More information

Factor models with many assets: strong factors, weak factors, and the two-pass procedure

Factor models with many assets: strong factors, weak factors, and the two-pass procedure Factor models wth many assets: strong factors, weak factors, and the two-pass procedure Stanslav Anatolyev 1 Anna Mkusheva 2 1 CERGE-EI and NES 2 MIT December 2017 Stanslav Anatolyev and Anna Mkusheva

More information

j) = 1 (note sigma notation) ii. Continuous random variable (e.g. Normal distribution) 1. density function: f ( x) 0 and f ( x) dx = 1

j) = 1 (note sigma notation) ii. Continuous random variable (e.g. Normal distribution) 1. density function: f ( x) 0 and f ( x) dx = 1 Random varables Measure of central tendences and varablty (means and varances) Jont densty functons and ndependence Measures of assocaton (covarance and correlaton) Interestng result Condtonal dstrbutons

More information

Department of Quantitative Methods & Information Systems. Time Series and Their Components QMIS 320. Chapter 6

Department of Quantitative Methods & Information Systems. Time Series and Their Components QMIS 320. Chapter 6 Department of Quanttatve Methods & Informaton Systems Tme Seres and Ther Components QMIS 30 Chapter 6 Fall 00 Dr. Mohammad Zanal These sldes were modfed from ther orgnal source for educatonal purpose only.

More information

Economics 130. Lecture 4 Simple Linear Regression Continued

Economics 130. Lecture 4 Simple Linear Regression Continued Economcs 130 Lecture 4 Contnued Readngs for Week 4 Text, Chapter and 3. We contnue wth addressng our second ssue + add n how we evaluate these relatonshps: Where do we get data to do ths analyss? How do

More information

Linear Regression Analysis: Terminology and Notation

Linear Regression Analysis: Terminology and Notation ECON 35* -- Secton : Basc Concepts of Regresson Analyss (Page ) Lnear Regresson Analyss: Termnology and Notaton Consder the generc verson of the smple (two-varable) lnear regresson model. It s represented

More information

Dr. Shalabh Department of Mathematics and Statistics Indian Institute of Technology Kanpur

Dr. Shalabh Department of Mathematics and Statistics Indian Institute of Technology Kanpur Analyss of Varance and Desgn of Experment-I MODULE VII LECTURE - 3 ANALYSIS OF COVARIANCE Dr Shalabh Department of Mathematcs and Statstcs Indan Insttute of Technology Kanpur Any scentfc experment s performed

More information

Joint Statistical Meetings - Biopharmaceutical Section

Joint Statistical Meetings - Biopharmaceutical Section Iteratve Ch-Square Test for Equvalence of Multple Treatment Groups Te-Hua Ng*, U.S. Food and Drug Admnstraton 1401 Rockvlle Pke, #200S, HFM-217, Rockvlle, MD 20852-1448 Key Words: Equvalence Testng; Actve

More information

x = , so that calculated

x = , so that calculated Stat 4, secton Sngle Factor ANOVA notes by Tm Plachowsk n chapter 8 we conducted hypothess tests n whch we compared a sngle sample s mean or proporton to some hypotheszed value Chapter 9 expanded ths to

More information

e i is a random error

e i is a random error Chapter - The Smple Lnear Regresson Model The lnear regresson equaton s: where + β + β e for,..., and are observable varables e s a random error How can an estmaton rule be constructed for the unknown

More information

Composite Hypotheses testing

Composite Hypotheses testing Composte ypotheses testng In many hypothess testng problems there are many possble dstrbutons that can occur under each of the hypotheses. The output of the source s a set of parameters (ponts n a parameter

More information

A TEST FOR SLOPE HETEROGENEITY IN FIXED EFFECTS MODELS

A TEST FOR SLOPE HETEROGENEITY IN FIXED EFFECTS MODELS A TEST FOR SLOPE HETEROGEEITY I FIXED EFFECTS MODELS TED JUHL AD OLEKSADR LUGOVSKYY Abstract. Typcal panel data models make use of the assumpton that the regresson parameters are the same for each ndvdual

More information

2016 Wiley. Study Session 2: Ethical and Professional Standards Application

2016 Wiley. Study Session 2: Ethical and Professional Standards Application 6 Wley Study Sesson : Ethcal and Professonal Standards Applcaton LESSON : CORRECTION ANALYSIS Readng 9: Correlaton and Regresson LOS 9a: Calculate and nterpret a sample covarance and a sample correlaton

More information

Basically, if you have a dummy dependent variable you will be estimating a probability.

Basically, if you have a dummy dependent variable you will be estimating a probability. ECON 497: Lecture Notes 13 Page 1 of 1 Metropoltan State Unversty ECON 497: Research and Forecastng Lecture Notes 13 Dummy Dependent Varable Technques Studenmund Chapter 13 Bascally, f you have a dummy

More information

Inner Product. Euclidean Space. Orthonormal Basis. Orthogonal

Inner Product. Euclidean Space. Orthonormal Basis. Orthogonal Inner Product Defnton 1 () A Eucldean space s a fnte-dmensonal vector space over the reals R, wth an nner product,. Defnton 2 (Inner Product) An nner product, on a real vector space X s a symmetrc, blnear,

More information

First Year Examination Department of Statistics, University of Florida

First Year Examination Department of Statistics, University of Florida Frst Year Examnaton Department of Statstcs, Unversty of Florda May 7, 010, 8:00 am - 1:00 noon Instructons: 1. You have four hours to answer questons n ths examnaton.. You must show your work to receve

More information

Psychology 282 Lecture #24 Outline Regression Diagnostics: Outliers

Psychology 282 Lecture #24 Outline Regression Diagnostics: Outliers Psychology 282 Lecture #24 Outlne Regresson Dagnostcs: Outlers In an earler lecture we studed the statstcal assumptons underlyng the regresson model, ncludng the followng ponts: Formal statement of assumptons.

More information

Notes on Frequency Estimation in Data Streams

Notes on Frequency Estimation in Data Streams Notes on Frequency Estmaton n Data Streams In (one of) the data streamng model(s), the data s a sequence of arrvals a 1, a 2,..., a m of the form a j = (, v) where s the dentty of the tem and belongs to

More information

Now we relax this assumption and allow that the error variance depends on the independent variables, i.e., heteroskedasticity

Now we relax this assumption and allow that the error variance depends on the independent variables, i.e., heteroskedasticity ECON 48 / WH Hong Heteroskedastcty. Consequences of Heteroskedastcty for OLS Assumpton MLR. 5: Homoskedastcty var ( u x ) = σ Now we relax ths assumpton and allow that the error varance depends on the

More information

Chapter 13: Multiple Regression

Chapter 13: Multiple Regression Chapter 13: Multple Regresson 13.1 Developng the multple-regresson Model The general model can be descrbed as: It smplfes for two ndependent varables: The sample ft parameter b 0, b 1, and b are used to

More information

Computing MLE Bias Empirically

Computing MLE Bias Empirically Computng MLE Bas Emprcally Kar Wa Lm Australan atonal Unversty January 3, 27 Abstract Ths note studes the bas arses from the MLE estmate of the rate parameter and the mean parameter of an exponental dstrbuton.

More information

Interval Estimation in the Classical Normal Linear Regression Model. 1. Introduction

Interval Estimation in the Classical Normal Linear Regression Model. 1. Introduction ECONOMICS 35* -- NOTE 7 ECON 35* -- NOTE 7 Interval Estmaton n the Classcal Normal Lnear Regresson Model Ths note outlnes the basc elements of nterval estmaton n the Classcal Normal Lnear Regresson Model

More information

Discussion of Extensions of the Gauss-Markov Theorem to the Case of Stochastic Regression Coefficients Ed Stanek

Discussion of Extensions of the Gauss-Markov Theorem to the Case of Stochastic Regression Coefficients Ed Stanek Dscusson of Extensons of the Gauss-arkov Theorem to the Case of Stochastc Regresson Coeffcents Ed Stanek Introducton Pfeffermann (984 dscusses extensons to the Gauss-arkov Theorem n settngs where regresson

More information

Hidden Markov Models & The Multivariate Gaussian (10/26/04)

Hidden Markov Models & The Multivariate Gaussian (10/26/04) CS281A/Stat241A: Statstcal Learnng Theory Hdden Markov Models & The Multvarate Gaussan (10/26/04) Lecturer: Mchael I. Jordan Scrbes: Jonathan W. Hu 1 Hdden Markov Models As a bref revew, hdden Markov models

More information

ECONOMETRICS II (ECO 2401S) University of Toronto. Department of Economics. Winter 2017 Instructor: Victor Aguirregabiria

ECONOMETRICS II (ECO 2401S) University of Toronto. Department of Economics. Winter 2017 Instructor: Victor Aguirregabiria ECOOMETRICS II ECO 40S Unversty of Toronto Department of Economcs Wnter 07 Instructor: Vctor Agurregabra SOLUTIO TO FIAL EXAM Tuesday, Aprl 8, 07 From :00pm-5:00pm 3 hours ISTRUCTIOS: - Ths s a closed-book

More information

Linear Approximation with Regularization and Moving Least Squares

Linear Approximation with Regularization and Moving Least Squares Lnear Approxmaton wth Regularzaton and Movng Least Squares Igor Grešovn May 007 Revson 4.6 (Revson : March 004). 5 4 3 0.5 3 3.5 4 Contents: Lnear Fttng...4. Weghted Least Squares n Functon Approxmaton...

More information

Limited Dependent Variables

Limited Dependent Variables Lmted Dependent Varables. What f the left-hand sde varable s not a contnuous thng spread from mnus nfnty to plus nfnty? That s, gven a model = f (, β, ε, where a. s bounded below at zero, such as wages

More information

LINEAR REGRESSION ANALYSIS. MODULE VIII Lecture Indicator Variables

LINEAR REGRESSION ANALYSIS. MODULE VIII Lecture Indicator Variables LINEAR REGRESSION ANALYSIS MODULE VIII Lecture - 7 Indcator Varables Dr. Shalabh Department of Maematcs and Statstcs Indan Insttute of Technology Kanpur Indcator varables versus quanttatve explanatory

More information

See Book Chapter 11 2 nd Edition (Chapter 10 1 st Edition)

See Book Chapter 11 2 nd Edition (Chapter 10 1 st Edition) Count Data Models See Book Chapter 11 2 nd Edton (Chapter 10 1 st Edton) Count data consst of non-negatve nteger values Examples: number of drver route changes per week, the number of trp departure changes

More information

/ n ) are compared. The logic is: if the two

/ n ) are compared. The logic is: if the two STAT C141, Sprng 2005 Lecture 13 Two sample tests One sample tests: examples of goodness of ft tests, where we are testng whether our data supports predctons. Two sample tests: called as tests of ndependence

More information

CHAPTER 8. Exercise Solutions

CHAPTER 8. Exercise Solutions CHAPTER 8 Exercse Solutons 77 Chapter 8, Exercse Solutons, Prncples of Econometrcs, 3e 78 EXERCISE 8. When = N N N ( x x) ( x x) ( x x) = = = N = = = N N N ( x ) ( ) ( ) ( x x ) x x x x x = = = = Chapter

More information

Lecture 12: Discrete Laplacian

Lecture 12: Discrete Laplacian Lecture 12: Dscrete Laplacan Scrbe: Tanye Lu Our goal s to come up wth a dscrete verson of Laplacan operator for trangulated surfaces, so that we can use t n practce to solve related problems We are mostly

More information

An (almost) unbiased estimator for the S-Gini index

An (almost) unbiased estimator for the S-Gini index An (almost unbased estmator for the S-Gn ndex Thomas Demuynck February 25, 2009 Abstract Ths note provdes an unbased estmator for the absolute S-Gn and an almost unbased estmator for the relatve S-Gn for

More information

Statistics for Managers Using Microsoft Excel/SPSS Chapter 13 The Simple Linear Regression Model and Correlation

Statistics for Managers Using Microsoft Excel/SPSS Chapter 13 The Simple Linear Regression Model and Correlation Statstcs for Managers Usng Mcrosoft Excel/SPSS Chapter 13 The Smple Lnear Regresson Model and Correlaton 1999 Prentce-Hall, Inc. Chap. 13-1 Chapter Topcs Types of Regresson Models Determnng the Smple Lnear

More information

9. Binary Dependent Variables

9. Binary Dependent Variables 9. Bnar Dependent Varables 9. Homogeneous models Log, prob models Inference Tax preparers 9.2 Random effects models 9.3 Fxed effects models 9.4 Margnal models and GEE Appendx 9A - Lkelhood calculatons

More information

Properties of Least Squares

Properties of Least Squares Week 3 3.1 Smple Lnear Regresson Model 3. Propertes of Least Squares Estmators Y Y β 1 + β X + u weekly famly expendtures X weekly famly ncome For a gven level of x, the expected level of food expendtures

More information

DO NOT OPEN THE QUESTION PAPER UNTIL INSTRUCTED TO DO SO BY THE CHIEF INVIGILATOR. Introductory Econometrics 1 hour 30 minutes

DO NOT OPEN THE QUESTION PAPER UNTIL INSTRUCTED TO DO SO BY THE CHIEF INVIGILATOR. Introductory Econometrics 1 hour 30 minutes 25/6 Canddates Only January Examnatons 26 Student Number: Desk Number:...... DO NOT OPEN THE QUESTION PAPER UNTIL INSTRUCTED TO DO SO BY THE CHIEF INVIGILATOR Department Module Code Module Ttle Exam Duraton

More information

Here is the rationale: If X and y have a strong positive relationship to one another, then ( x x) will tend to be positive when ( y y)

Here is the rationale: If X and y have a strong positive relationship to one another, then ( x x) will tend to be positive when ( y y) Secton 1.5 Correlaton In the prevous sectons, we looked at regresson and the value r was a measurement of how much of the varaton n y can be attrbuted to the lnear relatonshp between y and x. In ths secton,

More information

Difference Equations

Difference Equations Dfference Equatons c Jan Vrbk 1 Bascs Suppose a sequence of numbers, say a 0,a 1,a,a 3,... s defned by a certan general relatonshp between, say, three consecutve values of the sequence, e.g. a + +3a +1

More information

III. Econometric Methodology Regression Analysis

III. Econometric Methodology Regression Analysis Page Econ07 Appled Econometrcs Topc : An Overvew of Regresson Analyss (Studenmund, Chapter ) I. The Nature and Scope of Econometrcs. Lot s of defntons of econometrcs. Nobel Prze Commttee Paul Samuelson,

More information

Chapter 5. Solution of System of Linear Equations. Module No. 6. Solution of Inconsistent and Ill Conditioned Systems

Chapter 5. Solution of System of Linear Equations. Module No. 6. Solution of Inconsistent and Ill Conditioned Systems Numercal Analyss by Dr. Anta Pal Assstant Professor Department of Mathematcs Natonal Insttute of Technology Durgapur Durgapur-713209 emal: anta.bue@gmal.com 1 . Chapter 5 Soluton of System of Lnear Equatons

More information

Chapter 14 Simple Linear Regression

Chapter 14 Simple Linear Regression Chapter 4 Smple Lnear Regresson Chapter 4 - Smple Lnear Regresson Manageral decsons often are based on the relatonshp between two or more varables. Regresson analss can be used to develop an equaton showng

More information

Department of Statistics University of Toronto STA305H1S / 1004 HS Design and Analysis of Experiments Term Test - Winter Solution

Department of Statistics University of Toronto STA305H1S / 1004 HS Design and Analysis of Experiments Term Test - Winter Solution Department of Statstcs Unversty of Toronto STA35HS / HS Desgn and Analyss of Experments Term Test - Wnter - Soluton February, Last Name: Frst Name: Student Number: Instructons: Tme: hours. Ads: a non-programmable

More information

The Granular Origins of Aggregate Fluctuations : Supplementary Material

The Granular Origins of Aggregate Fluctuations : Supplementary Material The Granular Orgns of Aggregate Fluctuatons : Supplementary Materal Xaver Gabax October 12, 2010 Ths onlne appendx ( presents some addtonal emprcal robustness checks ( descrbes some econometrc complements

More information

Efficient nonresponse weighting adjustment using estimated response probability

Efficient nonresponse weighting adjustment using estimated response probability Effcent nonresponse weghtng adjustment usng estmated response probablty Jae Kwang Km Department of Appled Statstcs, Yonse Unversty, Seoul, 120-749, KOREA Key Words: Regresson estmator, Propensty score,

More information

January Examinations 2015

January Examinations 2015 24/5 Canddates Only January Examnatons 25 DO NOT OPEN THE QUESTION PAPER UNTIL INSTRUCTED TO DO SO BY THE CHIEF INVIGILATOR STUDENT CANDIDATE NO.. Department Module Code Module Ttle Exam Duraton (n words)

More information

2.3 Nilpotent endomorphisms

2.3 Nilpotent endomorphisms s a block dagonal matrx, wth A Mat dm U (C) In fact, we can assume that B = B 1 B k, wth B an ordered bass of U, and that A = [f U ] B, where f U : U U s the restrcton of f to U 40 23 Nlpotent endomorphsms

More information

Computation of Higher Order Moments from Two Multinomial Overdispersion Likelihood Models

Computation of Higher Order Moments from Two Multinomial Overdispersion Likelihood Models Computaton of Hgher Order Moments from Two Multnomal Overdsperson Lkelhood Models BY J. T. NEWCOMER, N. K. NEERCHAL Department of Mathematcs and Statstcs, Unversty of Maryland, Baltmore County, Baltmore,

More information

The Ordinary Least Squares (OLS) Estimator

The Ordinary Least Squares (OLS) Estimator The Ordnary Least Squares (OLS) Estmator 1 Regresson Analyss Regresson Analyss: a statstcal technque for nvestgatng and modelng the relatonshp between varables. Applcatons: Engneerng, the physcal and chemcal

More information

Asymptotic Properties of the Jarque-Bera Test for Normality in General Autoregressions with a Deterministic Term

Asymptotic Properties of the Jarque-Bera Test for Normality in General Autoregressions with a Deterministic Term Asymptotc Propertes of the Jarque-Bera est for Normalty n General Autoregressons wth a Determnstc erm Carlos Caceres Nuffeld College, Unversty of Oxford May 2006 Abstract he am of ths paper s to analyse

More information

Dr. Shalabh Department of Mathematics and Statistics Indian Institute of Technology Kanpur

Dr. Shalabh Department of Mathematics and Statistics Indian Institute of Technology Kanpur Analyss of Varance and Desgn of Exerments-I MODULE III LECTURE - 2 EXPERIMENTAL DESIGN MODELS Dr. Shalabh Deartment of Mathematcs and Statstcs Indan Insttute of Technology Kanur 2 We consder the models

More information

Dr. Shalabh Department of Mathematics and Statistics Indian Institute of Technology Kanpur

Dr. Shalabh Department of Mathematics and Statistics Indian Institute of Technology Kanpur Analyss of Varance and Desgn of Experment-I MODULE VIII LECTURE - 34 ANALYSIS OF VARIANCE IN RANDOM-EFFECTS MODEL AND MIXED-EFFECTS EFFECTS MODEL Dr Shalabh Department of Mathematcs and Statstcs Indan

More information

NUMERICAL DIFFERENTIATION

NUMERICAL DIFFERENTIATION NUMERICAL DIFFERENTIATION 1 Introducton Dfferentaton s a method to compute the rate at whch a dependent output y changes wth respect to the change n the ndependent nput x. Ths rate of change s called the

More information

Lecture 3 Stat102, Spring 2007

Lecture 3 Stat102, Spring 2007 Lecture 3 Stat0, Sprng 007 Chapter 3. 3.: Introducton to regresson analyss Lnear regresson as a descrptve technque The least-squares equatons Chapter 3.3 Samplng dstrbuton of b 0, b. Contnued n net lecture

More information

Testing for Granger Non-causality in Heterogeneous Panels

Testing for Granger Non-causality in Heterogeneous Panels Testng for Granger on-causalty n Heterogeneous Panels Chrstophe Hurln y June 27 Abstract Ths paper proposes a very smple test of Granger (1969) non-causalty for heterogeneous panel data models. Our test

More information

Estimation: Part 2. Chapter GREG estimation

Estimation: Part 2. Chapter GREG estimation Chapter 9 Estmaton: Part 2 9. GREG estmaton In Chapter 8, we have seen that the regresson estmator s an effcent estmator when there s a lnear relatonshp between y and x. In ths chapter, we generalzed the

More information

Bayesian predictive Configural Frequency Analysis

Bayesian predictive Configural Frequency Analysis Psychologcal Test and Assessment Modelng, Volume 54, 2012 (3), 285-292 Bayesan predctve Confgural Frequency Analyss Eduardo Gutérrez-Peña 1 Abstract Confgural Frequency Analyss s a method for cell-wse

More information

Chapter 12 Analysis of Covariance

Chapter 12 Analysis of Covariance Chapter Analyss of Covarance Any scentfc experment s performed to know somethng that s unknown about a group of treatments and to test certan hypothess about the correspondng treatment effect When varablty

More information

Topic- 11 The Analysis of Variance

Topic- 11 The Analysis of Variance Topc- 11 The Analyss of Varance Expermental Desgn The samplng plan or expermental desgn determnes the way that a sample s selected. In an observatonal study, the expermenter observes data that already

More information

β0 + β1xi and want to estimate the unknown

β0 + β1xi and want to estimate the unknown SLR Models Estmaton Those OLS Estmates Estmators (e ante) v. estmates (e post) The Smple Lnear Regresson (SLR) Condtons -4 An Asde: The Populaton Regresson Functon B and B are Lnear Estmators (condtonal

More information

UNIVERSITY OF TORONTO Faculty of Arts and Science. December 2005 Examinations STA437H1F/STA1005HF. Duration - 3 hours

UNIVERSITY OF TORONTO Faculty of Arts and Science. December 2005 Examinations STA437H1F/STA1005HF. Duration - 3 hours UNIVERSITY OF TORONTO Faculty of Arts and Scence December 005 Examnatons STA47HF/STA005HF Duraton - hours AIDS ALLOWED: (to be suppled by the student) Non-programmable calculator One handwrtten 8.5'' x

More information

Exam. Econometrics - Exam 1

Exam. Econometrics - Exam 1 Econometrcs - Exam 1 Exam Problem 1: (15 ponts) Suppose that the classcal regresson model apples but that the true value of the constant s zero. In order to answer the followng questons assume just one

More information

PHYS 705: Classical Mechanics. Calculus of Variations II

PHYS 705: Classical Mechanics. Calculus of Variations II 1 PHYS 705: Classcal Mechancs Calculus of Varatons II 2 Calculus of Varatons: Generalzaton (no constrant yet) Suppose now that F depends on several dependent varables : We need to fnd such that has a statonary

More information

2E Pattern Recognition Solutions to Introduction to Pattern Recognition, Chapter 2: Bayesian pattern classification

2E Pattern Recognition Solutions to Introduction to Pattern Recognition, Chapter 2: Bayesian pattern classification E395 - Pattern Recognton Solutons to Introducton to Pattern Recognton, Chapter : Bayesan pattern classfcaton Preface Ths document s a soluton manual for selected exercses from Introducton to Pattern Recognton

More information

CSci 6974 and ECSE 6966 Math. Tech. for Vision, Graphics and Robotics Lecture 21, April 17, 2006 Estimating A Plane Homography

CSci 6974 and ECSE 6966 Math. Tech. for Vision, Graphics and Robotics Lecture 21, April 17, 2006 Estimating A Plane Homography CSc 6974 and ECSE 6966 Math. Tech. for Vson, Graphcs and Robotcs Lecture 21, Aprl 17, 2006 Estmatng A Plane Homography Overvew We contnue wth a dscusson of the major ssues, usng estmaton of plane projectve

More information

Chapter 3. Two-Variable Regression Model: The Problem of Estimation

Chapter 3. Two-Variable Regression Model: The Problem of Estimation Chapter 3. Two-Varable Regresson Model: The Problem of Estmaton Ordnary Least Squares Method (OLS) Recall that, PRF: Y = β 1 + β X + u Thus, snce PRF s not drectly observable, t s estmated by SRF; that

More information

Limited Dependent Variables and Panel Data. Tibor Hanappi

Limited Dependent Variables and Panel Data. Tibor Hanappi Lmted Dependent Varables and Panel Data Tbor Hanapp 30.06.2010 Lmted Dependent Varables Dscrete: Varables that can take onl a countable number of values Censored/Truncated: Data ponts n some specfc range

More information

Lecture 3: Probability Distributions

Lecture 3: Probability Distributions Lecture 3: Probablty Dstrbutons Random Varables Let us begn by defnng a sample space as a set of outcomes from an experment. We denote ths by S. A random varable s a functon whch maps outcomes nto the

More information

Comparison of Regression Lines

Comparison of Regression Lines STATGRAPHICS Rev. 9/13/2013 Comparson of Regresson Lnes Summary... 1 Data Input... 3 Analyss Summary... 4 Plot of Ftted Model... 6 Condtonal Sums of Squares... 6 Analyss Optons... 7 Forecasts... 8 Confdence

More information

Projection estimators for autoregressive panel data models

Projection estimators for autoregressive panel data models Econometrcs Journal (2002), volume 5, pp 457 479 Projecton estmators for autoregressve panel data models STEPHEN BOND AND FRANK WINDMEIJER Nuffeld College, Unversty of Oxford, Oxford, OX1 1NF, UK CEMMAP,

More information

BOOTSTRAP METHOD FOR TESTING OF EQUALITY OF SEVERAL MEANS. M. Krishna Reddy, B. Naveen Kumar and Y. Ramu

BOOTSTRAP METHOD FOR TESTING OF EQUALITY OF SEVERAL MEANS. M. Krishna Reddy, B. Naveen Kumar and Y. Ramu BOOTSTRAP METHOD FOR TESTING OF EQUALITY OF SEVERAL MEANS M. Krshna Reddy, B. Naveen Kumar and Y. Ramu Department of Statstcs, Osmana Unversty, Hyderabad -500 007, Inda. nanbyrozu@gmal.com, ramu0@gmal.com

More information

Lecture 10 Support Vector Machines II

Lecture 10 Support Vector Machines II Lecture 10 Support Vector Machnes II 22 February 2016 Taylor B. Arnold Yale Statstcs STAT 365/665 1/28 Notes: Problem 3 s posted and due ths upcomng Frday There was an early bug n the fake-test data; fxed

More information

Chapter 6. Supplemental Text Material

Chapter 6. Supplemental Text Material Chapter 6. Supplemental Text Materal S6-. actor Effect Estmates are Least Squares Estmates We have gven heurstc or ntutve explanatons of how the estmates of the factor effects are obtaned n the textboo.

More information

Time-Varying Systems and Computations Lecture 6

Time-Varying Systems and Computations Lecture 6 Tme-Varyng Systems and Computatons Lecture 6 Klaus Depold 14. Januar 2014 The Kalman Flter The Kalman estmaton flter attempts to estmate the actual state of an unknown dscrete dynamcal system, gven nosy

More information

ECONOMETRICS - FINAL EXAM, 3rd YEAR (GECO & GADE)

ECONOMETRICS - FINAL EXAM, 3rd YEAR (GECO & GADE) ECONOMETRICS - FINAL EXAM, 3rd YEAR (GECO & GADE) June 7, 016 15:30 Frst famly name: Name: DNI/ID: Moble: Second famly Name: GECO/GADE: Instructor: E-mal: Queston 1 A B C Blank Queston A B C Blank Queston

More information