一3←
Onthedeficienqroftestsforstructuralchangeina
non−COintegrationregressionmodel*
Hiroyuki Hisamatsu Contents lIntroduction2 Modeland te$t Statistics
3 Asymptoticdistributionsoftheteststatistics 4 Monte Carlo experiment
5 Summaryandremarks Abstract WeconsiderthetandFtypetests払IteStingastIuCtuIalchangein economicrelationshipwhosedatageneratingpIOCeSSisnon−COintegration WeshowthatthedistIibutionsoft,hesetandFtestsdonothavestan− dardasymptoticdistributionsandinsmallsampletheypeめrmo㈹r rq】eCtionfor thenul1hypothesis ofnostruCturalchange・Tbdo so
wederivethenonstandaIdasymptoticdistributionsofthenoImalized thesetestsandshowthefinitesampledistributionsbyMonteCalk) expeIiment
ーThispape工isaleVisedvelSiono=4l1am酢atehltoKoidliMaekawafoI、belp餌1 commentsonthe丘IStVerSionofthispapeェ
−34− 香川大学経済学部 研究年報 40 ご(丸)(フ 1 Introduction Theproblemsofuni七‡00tandcointegrationtestfbrthemodelincluding StruCturalchangewereexaminedinmanyarticles。Fbrexample,SeeCampos, Ericssonand王壬endIy(1996),Leybourne,MillsandNewbold(1998),Grego‡y andHansen(1996),Ha七anakaandYamada(1998)・TheseminalwoIkwas Perron(1989)whoarguedthatifthereisabreakinthedeterministictrend, thenunit roo七testswillleadto amisleadingconclusionAnoverviewof theseproblemsaregiveninMaddalaandKim(1998)‖Thesepapersmadea StruCturalchangeasagivenconditionontheunitrootandcointegration tests Inthispaperwedisc11SSthatthetraditionaltes七sofstructuralchange willleadtoamisleadingconclusionundertheno−COintegrationdatageneI− atingprocess(DGP)・TheoIganizationofthispaperisasfbllows.Insection 2wedevelopethemodelsincludingstructuralchangeanddefinethetest Statistics..Insecもion3wedeIivetheasymptoticdist‡ibutionsofthesetests basedonthenonstandardasymptotics”Insection4weexaminethesmalI SampledistributionsofthesetestsbyMonteCarloexperiment.,Finally,in SeCtion5weclosewithsomeconcludingremarks
2 Modeland test statistics Weconsiderthefo1lowingmodels 訓仁=α+β∬亡+叫,≠=1,2,.=‖,r ごt=勘−1+uゎ Model(A) 祝電=β祝ト1+∈ゎ (1) wheIeVt∼N(0,0・雲),et∼N(0,q2),Xo=0,andp=1 Weassumethatthemodel(A)istheDGPof(yt)”Thisisnon−StruCtural Changeoveral1sampleperiod.Ifp<1thisiscointegrationsystemandif P=1thisisno−COintegration WeconsideIthethreetypesofmodelsinvolvingthestruCturalchange Onlyonconstant七erm,Onlyongradientcoe伍cient,Simultaneous1yonboth COnStantte工mandgIadientcoefRcient
Onthedeficiencyoftestsforstructuralchangeina non−COintegrationregr−eSSionmodel ー35−− 的=αβ+ββ∬土+7か2亡+祝わま=1,2,…,r β2t=1if壬>n, =Oiff≦れ, Model(B) (2) whereD2tisadummyvaIiableand71ref6rtothe七imeofbreak・This modelinvolvesthestIuCturalchangeonlyonthecon$tantterm Model(C) yl=αc+βcxi+∈D2まXt+ul,ま=1,2,…=,T (3) Thismodelinvolvesthestructu‡alchangeonlyonthegradientcoe乱 cient。 Model(D) 肌∵=α1ヱ)1£+β1ヱ)1が紆+α2か2£+β2上)2t∬t+勒,壬=1,2,・・1・・,r (4) か1f =1ifま≦71, = Oifま>Tl
where Dltis a dummy vaIiableり This modelinvolves the stIuCtural ChangeOnbothconstanttermandgradien七COe覿cient WejudgethestructuIalchangebytheま−ValueofDummyvariable払r themodel(B)and(C),andfo1lowingF−Statisticsfbrthemodel(D)against (A) Thet−Statisticsinthemodel(B)isdefinedas where?isderivedas
−36− 香川大学経済学部 研究年報 40 2(フ00 [軍]=[ 1∬£ 1 ( ∑勃 ∑寓肋 ∑荒+1研 OntheotheIhand,thet−Statisticsinthemodel(C)isde丘nedas 舌ざ・=ぞ/ = 計 (e♭ec/(ア−3))(r(∑桔ト(∑勘)2) ( ∬t)2
WhereEisderivedas
−1 ご.′・・ ご.′・ご∑諷.1
声]= r =・∫・∫ ∑芸+1£毛 ( ∑初 ご・′・・J/′ ∑諷.1瑚 TheF−S七atisticsfor七hemodel(D)against(A)isde丘nedas ダ=((eムeA−ebeェ〉)/G)/(ebeヱ)/(ア−2G)) (7)Wheree』eDisthesumofsquareofresidualsof(D)andeheAisof
(A)andGisanumberoflinearIeStrictionsonparametersunderthenull hypothesisthatHo:α=α1=α2,β=β1=β2.InthiscaseGbecomes2 3 Asymptoticdistributionsofthetest statistics FbllowingMaekawaandHisamatsu(1999),Phillips(1986),PaIkandPhillips (1988),PhiliipsandDurlauf(1986),Wederive七heasymp七Oticdi由ibutions Ofthet−Statisticsinthemodel(B)and(C)andF−Statisticsfbrthemodel (D)喝ahst(A) lemmalUndertheDGP(A),Wehavethefo1lowingresults: 1去妾£亡⇒湖上榊γ7
On the deficiency of tests for structural change in a non−COintegrationr■egreSSionmodel −3芦−−− 1
宗畠∬誉⇒現勒(γ)2dγ,
1詰責祝t⇒招上榊,
1 纏祝誉⇒現町γ)2dγ}妄妾珊⇒席汀晰)榊,
1 1去妾γt⇒β湖上醐+招上嘲れ
1妄妾瑚
⇒β現1晰)2d叶席上叫γ)W細
1島套yぎ⇒動糎榊+2β席上晰)榊
十現1咄γ)2dγ,
1姦彰⇒湖上醐γ,
r 姦t去1∬t⇒凋.上1
Ⅵち2(γ)か,WheIe7も=r−れ,韻∬誉⇒J雲.上1勒1(γ)2か,
⇒ 2t ∫ l r∑和 ⊥α雲。.上1一戦2(γ)2dγ,
志貴明⇒湖上1晰)dr,
1r 姦意1祝t⇒招上勒2(γ)か,
2000 香川大学経済学部 研究年報 40 −3β− 1
韻祝誉⇒現晰)2dγ,
毒真髄ぎ⇒J躯(γ)2dγノ,
1 1孟春⇒β招上醐γ・招上醐γ,
r 1志意1眺⇒β湖上1醐γ+招上嘲γ,
1諸物⇒β現1勒1(γ)2d什、席上晰)醐γ,
1 1毒泉瑚⇒β現明2(γ)2dγ・席上勒2(γ)晰)dγ,
1 1 縫針ぎ⇒飾雲上勒1(γ)2dγ+2β席ぎょ晰)醐γ十現1晰)2dγ,
毒泉yぎ⇒飾雲上1勒2(γ)2れ2β招上1勒刷冊
巾ぎ.上1勒2(γ)2dγ,
WecanderivetheseIeSultsinlinewiththefo1lowlngeXample去妾∬t=妄招套(表見モ)
=招£ニ′r義榊
⇒招上1一戦(γ)dγ,
Onthedeficiencyoftestsforstructuralchangeina non−COintegrationregr−eSSionmodel −39−
去妾机=妄凋を遠島t)
=招震′γ議=榊
⇒招上1晰)か,
wheIe【1denotestheintegerpartofitsargumen七and“⇒”signifiesconver− genceindistributionandⅥ′(r)isastandardBrownianmOtion FiIStWederivetheasymptoticdistributionofthenormalizedsumof squaresofIeSiduals,eLeA/T2,eもeB/T2,e♭ec/T2andebeD/T2byusing lemmal宗eゝeA=宗妾{…ト如一翔2
=妄妾yぎー感鼻)2
増産套瑚−(詰責∬£)烏を)} +飢妄重か(去畠∬t)2} 1 1⇒β現1嘲2れ現町γ)2机2席上町γ)榊r
−(β招上1−嘲机招上1榊γ)2−神君.上1勒(γ)2か十鱒上1晰)醐γ
−(湖上1榊γ)(β招上1榊γ寸招上1榊γ)} ・ぐ2{J雲上1晰)2か−(招上1勒(γ)か)2}香川大学経済学部 研究年報 40 ご()0() −イ(L 1 =舶−く)2{上1一戦(γ)2か(上醐γ)2} 竹上1町γ)2dγ−(上1榊γ)2} +2招和一く){上1町γ)晰)か−(上1榊γう(上1一戦(γ)dγ)} = Q、 wheI・eβ⇒く(seeAppendix) 基eらeβ=宗主((研一謬)一弘一語)増ゃ2モー玩))2 f=1
=緩封ぎー島お)2
f=1 −2如詰物一缶妾諾t)(姦重訂t)} 2+媚皇か烏麦刷+欝 t=1
−2卜
塙撃姦姜y電
r れ乃、/笥1 て− ∑飢) 壬=n+1 J、 了1、一に乃nJ弔1き
+2如(− ヱ11、耳 r 1⇒β現−1Ⅵち(γ)2机2β減市上1嘲γ)W冊+ゼ.上晰)2か
Onthedeficiencyoftestsfor’StruCturalchangeina non−COintegrationregressionmodel −(β招上1w冊+招上1榊γ)2 1
+畔上1蜘)2打席上榊γ)2}+撫一入)
−2甜現1勒(γ)2dγ+凋帝王1晰)榊γ −(湖上1醐γ)(β招上1榊γ+招上1醐γ)} 1 −2〃(小人)入山(β招上1勒冊+招上晰刺 ー4J− 1 ▼−−1 J。パ/〟。‖圧川バ/ 川 ︸ ︶ γ d ︶ γ ︵ 2 叫 l /L川 招 +入(1一入)府(β∨筍 Ⅵ勺2(γ)か+ +次β〃(−(ト入)入√丈∨筍 l鳩1(γ)dγ 勒2(γ)町現1一明2(γ)dγ)} +入(1一入)府(β∨筍 = Qβ,where事→入,草→1一入,PB⇒くB,?⇒P(seeAppendix)
轟ec=宗畠{(祈璃卜応極一句一拍毎町」扇)}2 =纏封ぎー烏麦yt)2−2翫宗套瑚一烏麦∬痘妾yt)}
華較−(去套∬f)2}
十嫁毒麦か巽(志麦㌦}
−2峰毒瑚)
−4a− 香川大学経済学部 研究年報 40 2000 r 左∬亡)(姦か)} 二・ニラ
+2あ峰毒か
一等島鳴擾)}
1 1 ⇒β2戎上勒(γ)射2β、β野上1町γ糊)恒現町榊
−(β招.上1w冊+招上1榊γ)2
・鋤雲上1嘲2dγ−(招J■1榊γ)2} 1+榊一入)2J2.上1勒2(γ)2d叫一入)3(招上勒2(γ)か)2}
1−2繍現1勒(γ陶+、/剰町γ)榊γ
−(凋上1勒(嘲(β招上1勒(γ)か丁招上1榊γ)}一2榊一入)2(β現1明2(γ)2d叶、/利1勒2(γ)明2(γ)か)
−(1一入)何滴.上1一戦2(γ)dγ
×(β招上1w冊十招上1町潮)
.上1
+2把可(トス)2戎 Ⅵち2(γ)2か 上1勒2(潮岬上1醐γ)) ー(1一入)府(∨筍Qc,
where応⇒くC,f⇒り(seeAppendix)。Onthede負ciencyoftestsforstructuralchangeina nonrcointegrationregressionmodel −43− 宗ebeβ=宗主抽壬一宮卜毎一明2 f=1 +宗主((研一謬卜録一明2 f=1
鍼一定)2}
−2毎瑚一年志
∬t)(姦yt)} +鎖毒∬ぎ一(姦∬t)2}]+鍼劾−(姦か)2}
−2鳩麦瑚−(志麦刷姦か
+釣毒か−(志麦㌦}】
=⇒ 入2(?1+(1一入)2Q2 1舶一く1)2{、上1一戦1(γ)2dγ一(上Ⅵ細か)2)
+2(β−く・1)、禰 ×{上1 晰)明1(γ)か一上1町)dγ上1一触桝 竹上1晰)2か−(上1晰)㈹ 舶−く2)2{上1Ⅵ鮒2か(上1一戦2(嘲2) WheI・eQl Q2香川大学経済学部 研究年報 40 +2(β一く2)席{亮一1勒2(γ)晰)dγ ご(〃ヤ) −4多−
、/二:
l仇2(γ)か l鳩2(γ)か) +♂ぎ{上1勒2(γ)2dγ−(上1勒2(γ)㈹ 夙 ⇒ く1,庖⇒く2(seeAppendix) S讐OndwederivetheasymptoticdistributionsofVar(今)of(B)and Var(E)of(C)byusingtheaboveresults e忘eβ r(∑≡1功一(∑≡1諾電)2 Ⅴαγ(守)= r−3 れ乃(∑た1∬2)+2筑(∑≡1∬壬)(∑諷.1∬f) 一丁(∑完.1諾t)2一策(∑己1£t)2 ブ、ト r(r−3) T3((姦∑た1∬ぎ)−(字詰∑≡1∬t)2) 一− _ 甘〟明(γ)2か−(∨領ぷ明(γ)呵2 r4 ==さQβ _,2 ぷ勒(γ)2dγ一(〟勒(γ)叫2 = Qβ ,2 =1包(γ)・Onthede丘ciencyoftestsforstruCturalchangeina nonMCOintegrationregressionmodel −ペは−
rx叫∈)=ア無
r(∑た1措ト(∑≡1勘)2 × _ J ) }− が((量∑た1尋)−(字詰∑≡1勘)2〉=T2{糸
ー ̄テ ニーー __、一幕(毒∑完.1戎)2
一業(姦∑己1∬t)2(寿∑完+1尋) J3⇒Q。「.一、讐㌣∴ご‥
ぷ勒(γ)2dγ−(〟勒(γ)叫2=‘Qc r.一JU‥∴、(∴_.′、ご∴.二∴∴′.し。−、「
戎 −(ト入)4(小梅(γ)2呵2 −(1−入)2(甘職(γ)dγ)2(か侮(γ)2dγ)_ 1七(γ)り一46− 香川大学経済学部 研究年報 40 2(フ00 FinauywedeIivetheasymptoticdistributionst亨/、庁of(B)1tざ/、斤of (C)andF/Tof(D)against(A) Tbeoremlmeαβγ叩わ如d壷β土加わ那桝/げ好例,fぎ/躇扉/q ・川・ナノ:、/J’イJ上‖りJJり川バ/トIJりい申・川りバ/り/Jり小 こ・、了− /′ 孟?/、庁= ⇒ 、/呵苛 ′v研’ り て まぞ・/斤= (eムeA/r2−eらe上)/r2)/2 (eゝeA−ebe上))/2 ノ・、r= ebeβ/(r−4) ebeβ/ア(r−4) 1(2−(入2(?1+(1一入)2(22) − ̄ ̄■■ 入2(21+(1一入)2(22 Theoremlshowsthat t守andtfdonotconvergeonthenoImaldistIi−  ̄∴ =・一 nothaveFdistributionandthenoImalizeFte$t,F/T,COnVergeSOnthe nonstandard asymptoticdistribution。Oneoftheimportant factor ofde− terminlng the asymptotic distributionis入whichis alimit ofthe br・eak f上・aCtionれ/TThuseVeninsmausample,thesedistributionareaf董bcted bythebI・eakf土aCtion71/T・ 4MonteCarloexperiment  ̄ 、三 (1)r=22,れ=9,出005(19)=士1・・729,凡05(2,18)=3。.555い(β=0: COintegIation)
Onthedeficiencyoftestsforstr’uCturalchangeina non−COintegrationr−egreSSionmodel −4アー ThedistributionsareshowninFig・・11fbr壬?,Fig・1・2fbrtf,Fig・1・3fbr FりTheIqiectionIateSOftheseteststhatexceedtheabovecIiticalvaluesat 5%1evelofthetandFdistributionsaIeSummarizedasfbllows 古今‥0一・1026々:0・・1090,ダ:0・0540 (2)r=22,れ=9,出005(19)=士1・729,昂05(2,18)=3・555い(β=1: no−COintegration) こ ‥− 壬今‥0・6150々:0・4233,ダ:06000 (3)r=240,n=120,出005(237)=士1・・651,昂05(2,236)=3・034 (β=0) ‥ ミ 壬?‥0・1027々‥0・・0997}ダ‥0・・0527 (4)r=240,れ=120,出005(237)=土1・651,昂05(2,236)=3」・034 (p=1) − ㌻・三 t?‥0・8947々:0・8330,F‥0・9700 (5)r=22,n=9(β=1) ThedistributionsofthenormalizedtestsareshowninFig・511fbrt琴/、庁 ,Fig52払Ⅰ々/、庁,Fig∴5・3foIダ/7∵ (6)r=240,れ=120(β=1) ThedistributionsofthenormalizedtestsareshowninFigl611for句/、庁 ,Fir tan。Ftestshawg。。。。erformanCe When p=O but these te$tS perform over rqjection when p=1。These
result,SShowthatifDGPisno−COintegIation,thetIaditionaltandFtests
forstIuCturalchangeperformoverrqjectionfbrthenu11hypothesisofno−
−4&− 香川大学経済学部 研究年報 40 2(フ00 5 Summaryandremarks WeconsideIedthedistr・ibutionalpropertiesof■tandFtypete$tSforstIuC− turalchangeinasimplelinearregressionmodelwhoseDGPisno−COintegration OurresultsshowthatthetandFtypetestsdonotconvergeonthestan− darddistIibutionswhenthesamplesizebecomeslargeh Thenon−StruCtural
ChangehypothesisisoverlyrQjectedand thenoImalized these testshave
nonstandardasymptoticdistributions Refbrences [1]Campos,Jl,Ericsson,N”R,andDHendry,1996,Cointegrationtestsin thepresenceofstructuralbreaks,Joumalqf’Economletrics,70,187−220 [2】Gregory,A”WI,andB思Hansen,1996,Residual−basedtestsfbrcoin− tegrationinmodelswithIegimeshifts,Journalqf’Ecorwmetrics,70, 99−126
【3]Hatanaka,Mu,and K=Yamada,1998,Aunit root testin thepI■eS− enceofstruCtuIalchangesinI(0)andI(1)models,inEngle・RF‥,and H.White(eds・),Cointqgration,仇usality,andFbrecastin9,256−282,Ox− fbrdUniversityPIeSS [4】Hisama七Su,Hl,2000,Onthedeficiencyoftestsfbrstructuralchange inanon−COintegrationr・egreSSionmodel,WbrkingpqperseriesNol35, 血βf加古eq/βco几Om豆cf‡eβ餓rCん,飢ダαぴα仇査脚β軸 【5】Leybourne,MillsandNewbold,1998,SpuIiousr−q5ectionsbyDickey− FhllertestsinthepresenceofabreakundeIthenull,JournalqfEcono− mefねcざ,87,191−203 【6]Maddala,G・S”,andKim,In−Moo,1998,仇iltrOOtS,COintegration,and structuralchange,CambIidgeUniveISitypress 【7]Maekawa,K”,andH”Hisamatsu,1999,Seeminglyunrelatedregression modelswithintegratedIegreSSOrS,PqPerPreSentedatthe凡rEastem 〟eef吻q′兢eβ00循Omefわcβocまety,Jタタ9 【8]Park,JY”andP‖C恩Phillips,1988,Statisticalinf6r’enCeSinregressions withintegratedprocesses:partl)Econometricneory,4}468−497
Onthede負ciencyoftestsforstructuralchangeina non−COintegrationr’egr−eSSionmodel −49−− [9】PeIrOn,P‖,1989,Thegreatcrash,theoilpriceshock,andtheunitroot hypothesis,βco†10metわcα,6,1361−1401 [10]Phi11ips,Pl・CB”1986,UnderstandingspuriousI■egreSSionsineconomet” Iics,Jo祝γ鶴αgげβ00循Omefれcβ,33,311−340 [11】Phillips,PC”B”andS‖NIDurlauf,1986,Multipletimeseriesregression withintegIatedprocesses,ReviewqfEconomicStudies,53,473−496 Appendix derivationofasymptotics: ∑≡1(∬モー司(研一謬)/r2 ∑≡1(勘一語)2/r2 ∑た1翔一(轟∑≡1∬f)(音義∑≡湖) 轟∑≡1∬芳一(姦∑≡1∬f)2
_((潮
J君克明(γ)2dγ−(、周√ぷ勒(γ)か)2 βα雲(甘明(γ)2dγ− l戦(γ)dり2) 勒(γ)dγ)(ガ勒(γ)か)) +v句帝(〟勒(ネ)勒(γ)dγ− J君(ぷ勒(γ)2か−(腋明(γ)dγ)2) 勒(γ)か−(〟勒(γ)呵(ぷ勒(γ)dγ) 甘明(γ)2dγ−(甘鵜(γ)dγ)2 一苅∑芝1∬亡∑た湖十n苅∑≡1動的−(r∑菟+1∬亡一基∑≡1∬t)∑諷.1的
[ 瑠詔悲 鉾′T42の0 香川大学経済学部 研究年報 40
 ̄苧∑≡1的’
ー50−− _剋竺 rヱ、テ (テ詰∑≡1∬t))(再議 争苧(姦∑≡1霊ぎ) 十2駕著∑≡1訂t)(再議∑完.1∬t)
.1諾t)2−(苧)(字詰∑≡1∬t)2
誌 ー(争)3(再 −(ト入)入バ∨筍ぷ勒1(γ)dγ ;慧二 .入(ト(潮×()か)
⇒_(γ)2
 ̄ ∵. ⊥人)仰〟勒(γ)dγ) ー((1一入)2〟勒2(γ)か−(1 乳ぷ勒2(γ刺 ×(β〟勒2(γ)dγ+(ア£
/(T4∼庁) 今/、庁= [Onthede負ciencyoftestsforstructuralchangeina non−COintegrationregr−eSSionmodel
肇荒想)
∑ −(讐浮(再議芸+1∬も)一筆(ラ嶺∑己1∬刷姦∑≡1瑚)十((轟∑≡1針−(音義∑≡1∬亡)2増穿(再議∑諷.1的)
ー5J−[争苧(彗款窓掛読芸喪≡き+1∬t)
{(1 ̄入)錘2(γ)dγ
_{(1_入)(潮
×仰㍑明(γ)2か+∨筍軍ぷ勒(γ)勒(γ)か)+(ト洲
=⇒ _γ,2 { −(㍉二言ぷ勒2(γ)か ̄ ̄
軍 . − ら(γ)か) (γ)l鳩(γ)dγ) ×(β、伺ぷ勒(ネ)2dγ+∨  ̄二 ̄ ∨ dγ)2) 巧ぷ勒2(γ)叫[佑
ニ ∵ ̄二 二‖ご()()() 香川大学経済学部 研究年報 40
 ̄{的
一丘2−/r5
、.‡・Tト
()
l J.γ ー((1一入)2(ノ帝〟り(γ)呵( (γ)2か) 2 tr 明月JO d −(1一入)3仰(∨筍.ぷ勒2(γ) Wら2(γ)2dγ)) ×(βv伺ぷ勒(γ)か +((1一入)2(グ雲 −(1一入)3(∨筍、モ
弟(γ)dγ) +蚕JO ︶ γ γ ︵ 2 勒 Ⅵち2(γ)dγ)2)×(βαγ)dγ)
−(ト入)仰(、/筍ぷ鞘2(γ)d項∨筍√ぷ勒(γ)dγ)) 周ぷ勒2(γ)勒2(γ)叫 ×仲雲〟勒2(γ)2dγ十 ==> (1一入)2(J雲甘職(γ)2dγ)(α雲 十2(1一入)3ノ王二1(∨筍」 ×(戎〟勒2(r)2dγ 一(1一入)3tん碩√〟勒2(γ)dγ)2(紘〟侮(γ)2呵 −(1一入)4(戎〟勒2(γ)2dり2 −(ト入)2(v筍〟勒(γ)呵2(♂雲〟明2(γ)2呵Onthedeficiencyoftestsforstructuralchangeina non−COintegrationr’egreSSionmodel −エ㌻− −((ぷ勒(r)dγ)(ぷ勒2(γ)2か) ー(ト入)荷(〟 l鳩2(γ)dγ) (ぷ明2(γ)2dγ)) 射両(γ匝) 入)(ぷ明2(γ)か)2) ×(β甘侮(γ沖+ +((〟勒2(γ)2dγ)−(1 射匝(γ)明(γ匝) ×(βぷ勒(γ)2dγ+ −((1一入)2
ミ.Jしさ
Ⅵち2(γ)2か) −(1一入)、/T=1(ぷ蔽2(γ)dγ)(ぷ明(γうか)) ×(β〟勒2(γ)2d付躇〟勒2(γ)勒2(γ桝 (甘明(γ)2dγ)(ぷ準2(γ)2呵1(
γ)(. +2(1一入)\/′i ̄= ×(〟堅2(γ)2d Ⅵ勺(γ)dγ) l穐2(γ)dγ) −(ト」)(〟勒2(γ)か)セ(〟勒(γ)2dγ) _(,/r5
= /r5ご(肋) 香川大学経済学部 研究年報 40
描(姦∑己1∬f)(毒∑諷.1功
一讐穿(亮誌∑諷.1∬£)(轟∑≡1功)毒∑≡1研
一牒(毒∑左.1∬苦)
) 一皮トー款轟∑≡1∬ぎ)(毒∑芸.1∬ぎ)
+瑠漂(姦∑≡1勅封琵.1動議冒∑芸.1∬t)
)_宗(毒.1∬誉,
−(1γ)}
小人)冊)}
×(β戎〟勒(γ)2か+v句搾ぷ勒(γ)勒(γ)か) て− 。ミ ==争 γ, −(1一入)4(J雲ぷ明2(γ)2か)2 一(ト入)2(、周ガ勒(γ)dγ)2(J雲〟勒2(γ)2呵Onthede負ciencyoftestsforstructuralchangeina non−COintegrationregressionmodel −55− } ×(β腋明(γ)か+ −−((1「入)(〟 l鳩2(γ)2dγ) (
×(,
+(ト入)((ぷ勒(γ)2呵−(ぷ勒(γ)呵2) 糾明2(γ)勒2(γ)か) ×(βぷ勒2(γ)2dγ+ _ −(1一入)4(.ノ言明2(γ)2dγ)2 一(1一入)(甘明(γ)か)2(〟勒2(γ)2dγ) =77, [ {(的 ]′が 戌= [ (莞.(動
]/r6 入3(1一入)3(グ君)2(ぷ勒2(γ)2d?−(ぷ勒2(γ)叫2) ×[β(ぷ勒1(γ)2dγ−(甘明1(γ)dγ)2) 射錘胡1(γ)か一針軋(γ)dγ勒1(γ匝)】 ⇒ 入3(1一入)3(戎)2(〟勒1(γ)2dγ−(〟勒1(γ)dγ)2) ×(ぷ勒2(γ)2か−り言明2(γ)dγ)2)[.融㌶豊盈㌫醐}
] (片桐1(γ)2dγ一(甘職1(γ)dγ)2)ご(つ(ノり 香川大学経済学部 研究年報 40 勒1(γ)勒1(γ)dγ−ぷ勒1(γ沖ガ勒1(γ)dγ‡ (〟勒1(γ)2か−(腋明1(γ)dγ)2† = く1,
{(∑聖曲芸「瑞:慧籠だ慧・1的
] .iご__・ [ (莞箆+(}
入3(1一入)3(α3)2(甘明1(γ)2dγ−(腋明1(γ)叫2† ×[β(ぷ勒2(γ)2dγ−(ぷ勒2(γ)dγ)2) 勒2(γ)勒2(γ)dγ−ノご勒2(γ)dγぷ勒2(γ)d珊 入3(1一入)3(戎)2(ぷ勒1(γ)2dγ−(〟勒1(γ沖)2‡ ×(ぷ勒2(γ)2d中一(〟勒2(γ)dγ)2) β(ぷ勒2(γ)2dγ−(ぷ勒2(γ)dγ)2) 新〟勒2(γ)勒2(γ)か−ぷ勒2(γ)かぷ勒2(γ沖) (〟勒2(γ)2dγ−(〟勒2(γ)dγ)2) 勒2(γ)勒2(γ)か−ぷ勒2(γ)dγぷ勒2(γ)か) (ぷ勒2(γ)2か−(〟勒2(γ)叫2)Onthedeficiencyoftestsforstructuralchangeina non−COintegrationregressionmodel FiglRel1:Distributionofthet−Statis七icsfbIModel(B) −「又声− lの 寸 0 寸 l√〉 M l=〉 M しn く勺 く⊃ (ヽ」 Ln ⊂) u) 0 0 く⊃
−16−12−8−40481216
note:T=22,71=9,P=0(cointegIation) Figurell2:DistributionoftheモーSta七istic$forModel(C) Uつ 寸 0 ヾ・ Ln 叩 く⊃ lり u■〉 N ⊂〉 N の くつ の くっ く⊃ ○ −16 −12 −8 −4 0 ヰ 8 12 16 note:r=22,n=9,β=0(cointegI・ation)香川大学経済学部 研究年報 40 Figurel.3:DistributionoftheF−Sta七istics 2000 ー5∠トーー 8 1〔) 2D こヨ○ ヰO note:T=22,71=9,P=0(cointegra七ion) Figure2“1:Distributionofthet−StatisticsforModel(B) 可・ くN くっ の 0 く0 0 づ・ く) N ⊂) ○ ○ −16 −12 −8 −4 0 ヰ 8 12 16 no七e:r=22,n=9,p=1(no−COin七egIation)
Onthedeficiencyoftestsforstructuralchangeina non−COintegr・ationregressionmodel Figure22:Dis七ribu七ionofthet−StatisticsforModel(C) −5伊− 寸 N ⊂〉 ぐヾ く○ く} の ○ ヾ・ ○ (〇 ○ −16 −12 −8 −ヰ 8 ヰ 8 12 16 note:ア=22,れ=9,β=1(no−00inもegI・如ion) Figure23:DistributionoftheF−Statis七ics 寸 く\l O の く♪ 寸 N <⊃ 8 10(〕 200 380 ヰ0(〕 note:T=22,Tl=9,P=1(no−COintegIation)
香川大学経済学部 研究年報 40 Fig一∬e31:DistIibutionofthet−Statistics払rModel(B) ご(礼)() S.† ?寸 の.M ?叩 の.N ?M 椚.︻ 〇.L S.〇 ?○ −16 −12 −8 −ヰ 0 ヰ 8 12 16 n叫‥T=240,n=120,β=0(coin七egI血on) Figue32:Distribu七ionofthet−S七atisticsforModel(C) の.寸 〇.寸 の.M ?門 の.q ?N の.L ?† S.〇 −16 −12 −8 −4 (〕 ヰ 8 12 16 note=T=240,Tl=120,P=0(coin七egIation)
Onthede丘ciencyoftestsfor’StruCturalchangeina non−COintegrationr−egr−eSSionmodel Figure33:DistIibutionof七heF−Statistics −6J− ⊂〉 の 別 r■ヽ く¢ Ln 寸 ト1 ぐ、」 く⊃ o 10 20 30 ヰ8 no七e:T=240,Tl=120,P=0(coin七egration) Figure4,1‥DistIibutionof七het−StatisticsforModel(B) く」⊃ ⊂) の (⊃ 寸・ ⊂〉 n く:〉 ぐq く⊃ d O d −ヰ0 −20 O 20 ヰ8 note‥r=240,n=120,β=1(no−COintegration)
ヱ(木社) 香川大学経済学部 研究年報 40 Figure42:Distributionof七het−Sta七isticsforModel(C) −62− の く> 「\ ○ く△ く⊃ l√〉 ⊂) 寸 く⊃ l√〉 くっ ぐ、l くっ く⊃ (=〉 く=〉
ー30−2D−10
O 10 2(〕 .5D note:ア=240,n=120,β=1(no−COin七e騨ation) Figure43:DistributionoftheF−Sta七istics ・■・ ぐヾ C〉 の く) くJO O 寸 く> M (⊃ く〉 0 208 60(〕 1000 1ヰ80 note‥T=240,Tl=120,P=1(no−CO血egIation)Onthede丘ciencyoftestsforstructuralchangeina non−COintegr−ationregr’eSSionmodel Figure51:DistributionofthenoImalizedtforModel(B) −1丈㌻− −16 −12 −8 −ヰ 0 ヰ 8 12 16 note:T=22,れ=9,β=1(no−COintegI・ation) Figure52‥Distributionof七henormalizedtforModel(C) の Cロ トヽ く○ しn 寸 n ぐヾ ⊂〉 −16 −12 −8 −4 0 ヰ 8 12 16 note:T=22,Tl=9,P=1(no−COintegIation)
2(フ00 香川大学経済学部 研究年報 40 Fig一皿e53:DistIibutionofthenormalizedF−Statis七ics −イ凄−−−−− 8:) N 寸 ぐ、」 (:⊃ N く」> く勺 の 寸 ○ 0 10 20 30 4(〕 note:T=22,71=9,P=1(no−COintegIation) Figure61:Di$tIibutionofthenoImalized七forModel(B) ー16 −12 −8 −ヰ 0 4 8 12 16 note:T=240,71=120,P=1(no−COintegIation)
Onthedeficiencyoftestsfor・StrtlCturalchangeina non−COintegrationregreSSlOnmOdel FiglRe6r2‥DistributionofthenoImalizedtforModel(C) −65−−−− ー16 −12 −8 −ヰ 0 ヰ 8 no七e‥r=240,n=120,β=1(no−COin晦Iation) 12 16 Figure63:DistributionofthenormalizedF−Statistics 0 10 20 30 4(〕 note:T=240,7l=120,P=1(no−COintegIation)