1、線(xiàn)性聯(lián)立方程組估計(jì)線(xiàn)性聯(lián)立方程組.PROC MODEL也可以達(dá)計(jì)線(xiàn)性聯(lián)立方程.SY5UN過(guò)程提供 以下估計(jì)方法OLS. 2SLS. LIML.SCR. I ISLR, 3SLS, IBSLS.HMLa 不炎供 GMM 方 法（MODEL過(guò)程提供）.1介紹OLS Estimation加果不指定估il方法，默認(rèn)用OLS.proc syslin data-in;demand: nodel q = p y 8； supply: nodel q p u;run;這相當(dāng)于兩個(gè)獨(dú)立的OLS估計(jì),但是由于聯(lián)立性,估fl結(jié)果可能會(huì)有大的偏差.Two-Stage Least Squares Estimation價(jià)

2、格P足內(nèi)生變量.外生變量為收入Y,單位成本U,替代品價(jià)格S.這時(shí)候用2SLS：proc syslin data-in 2sls； endogenous p;instruments y u a;demand： model q p y ; supply: nodel q = p u;run;這里告訴系狡p是內(nèi)生變址.雖然q也是內(nèi)生變色,但是q不作為解釋變外,沒(méi) 有必要列出來(lái).往往所有前定受量都用作工具變量,這里Y,U,S都用作工具變量,用在第一步回 IQ中慎#IP.輸出結(jié)果只有第二步.如果要看第一步結(jié)果,在MODEL語(yǔ)句中用FIRST選項(xiàng)。Analysis of VarianceSourceDFS

3、um ofSquaresMeanSquareF ValuePr FModel39.6708823.223627115.58 FModel29.6460984.823049253.96 |t|Variable LabelIntercept1-0.518780.490999-1.060.2952InterceptP11.3330800.05927122.49 0001Priceu1-1.146230.243491-4.71 |t|Variable LabelIntercept11.9802611.1691691.690.0959InterceptP1-1.176540.605012-1.940.0

4、568Pricey10.4041150.1171793.450.0011Incomes10.3592040 0850774.22 |t|Variable LabelIntercept1-0.518780.4909991.060.2952InterceptP11.3330800.05927122.49.0001Pncru1-1.146230.2434914.71 It|Variable SbelIntercept11.9885291.2336251.610.1126InterceptP1-1.181470.652274-1.810.0755Pricey10.4023100.1072693.750

5、.0004Incomes10.3613450.1038163.480.0010Price ofSubstitutesModelSUPPLYDependent VariableqLabelQuantityParameter EstimatesVariableDFPara meter EstimateStandardErrort ValuePr It|Variable LabelIntercept1-0.524430.479522-1.090.2787InterceptP11.3360830.05793923.06.0001Priceu11.148040.2377934.83.0001Unit C

6、ostFigure 19.7: FIML ResultsComputing Reduced Form Estimates可以用REDUCED選項(xiàng)輸出帙型的REDUCED形式.proc ayalln dat*in 31ls reduced; endo9anouA p;inatrument， y u 8； demand: model q = p y s;supply: nodel qP U；Restricting Parameter Estimates可以用restrict語(yǔ)句和srestrict語(yǔ)句對(duì)賽數(shù)魄加線(xiàn)性約束。Restrict語(yǔ)句對(duì) 前面的nodel語(yǔ)句中的參數(shù)通加方程內(nèi)的約率,而sre

7、strict用來(lái)謔加方程網(wǎng)的 ，數(shù)約束,這需要告訴摹統(tǒng)是哪個(gè)方程的冷數(shù)。有西種方法,如果方程有l(wèi)abel, 直接用label.變最衣示相應(yīng)的參數(shù).如果沒(méi)有l(wèi)abel. E以用這個(gè)方程的因變量 作為labels輸出加應(yīng)拉格朗H乘數(shù),proc ayalln data*in 3als;ndoganous p;insrrumnt y U |t|Variable LabelInterceptIntercept11.9802611.1691691.690.0959P1-1.176540.605012-1.940.0568PriceY10.4041150.1171793.450.0011Incomes10.

8、3592040.0850774.22 F11136790.0104Figure 19.10: TEST Statement Results跨方程的愴驗(yàn)要用proc syslln data*in 3als;endo9enoua p;instrumantfl y u s;demand: nodel q p y 0;supply: nodel q = p u;ateatl: atest demand.y .01 supply.u; run;The SYSUN Procedure Three-Stage Least Squares Estimation System Weighted MSEDegre

9、es of freedom|113System Weighted R-Square 0.9627ModelDEMANDDependent VariableqLabelQuanblyParameter EstimatesVariableDFParameter EstimateStandardErrort ValuePr a |t|Variable LabelIntercept11.9802611.1691691.690.0959InterceptP1-1.176540.605012-1.940.0568Pricey10.4041150.1171793.450.0011Incomenn COCC7

10、7a noSubstitutesModelSUPPLYDependent VariableqLabelQuantityParameter EstimatesVariableDFParameter EstimateStandard Errort ValuePr It|VHnble LabelIntercept1-0.518780.4909991.060.2952InterceptP11.3330800.05927122.49.0001Priceu11.146230.2434914.71 F111322.460.0001J以保存同測(cè)伯和殯希.在“Tin語(yǔ)句用。ut數(shù)融集,然后用在相應(yīng)方程后血 用o

11、utput語(yǔ)句飾出要輸出的變顯和保存變顯名.proc syslln dataln out*pred 3sls;endogenous p;inatrumontd y u s;demand: model q p y s;output predlctedqdcmand;supply: nodl q - p u；output predicted=1_suppLy;run;2語(yǔ)法:PROC SYSL1N options ；fiY variables;ENDOGENOUS variables;IDENTITY identities ;INSTRUMENTS variables ;MODEL respons

12、e = regressors / options ;OUTPUT PREDICTED= variable RESIDUAL= variable ;RESTRICT restrictions ;SRESTRICT restrictions ；SIESI equations ;TESY equations ；VAR variables ;WEIGHT vanable;PROC SYSLIN StatementPROC SYSLIN options；DATA= SAS-data-setOUT= SAS-data-set指定一個(gè)數(shù)據(jù)篥出及差和預(yù)色值.和后面3PW誨句配合使用.OUTEST= SAS-d

13、ata-set崎的叁數(shù)估計(jì)結(jié)果.OUTCOVCOVOUT紀(jì)參數(shù)方外矩內(nèi)珈出H。心i= 里面.OUTCOV3COV3OUT使用3s1s.bg timl方出的時(shí)候,T而個(gè)方程的協(xié)方注矩田輸出到ouy估計(jì)方法2SLS3SLSFIMLITSURIT3SLSLIMLOLSSUR出，REDUCED 抬出陵個(gè)的REDUCED FORM 必狽用IDENTITY lA句使俺解殳成方的ENDOGENOLIS StatementENDOGENOUS variables ;坦定第一步需要用，只交0件預(yù)知的內(nèi)，殳S不一定是殳部的內(nèi)牛女IDENTITY StatementIDEHTTrf equation ; 許出定義式

14、，怛等式.label: IDENTITY ;INSTRUMENTS StatementINSTRUMENTS variables ;MODEL StatementMODEL response = regressors / options ; 立力如下：CORRBCOVB DW NOINT OVERID過(guò)僮識(shí)別為奴檢5ft.OUTPUT StatementOUTPUT PREDICTED= variable RESIDUAL=variable ;和SYSLIN中的OUT=數(shù)摳*配合隹用.PREDICTED= variableIRRECT-, PRFD- or I*-.RESIDUAL= vari

15、ableRESID-or R-proc syslln dataa outb; model y s xl x2;output p=yhat ryrsid; run;proc yslln data*in outbred; decaand: model q p y ;output p=q_demand rr_demand; supply: model q p u;output p*q_aupply rr_aupply; run;RESTRICT StatementRESTRICT equabon，.equation ;施加紋忖的蟲(chóng).可以用NTERCEPT做板M1J配可以使用LABEL婦：label : RESTRICT 一 ;restrictreatriatinter8Pt x4 = 0;xl x2 , #3估。方法是加拉格朗H乘數(shù).注意.RESTRICT語(yǔ)句不能用在FIML估諱方法中,SRESTRICT StatementSRESTRICT equation，.equation ;和RESTRICT爰不多.只是用于的方杵下數(shù)的七 用LARELSRIABLE譏加半缸 沒(méi)書(shū) LABEL的方程用該方程的因變件LAREL,只要沒(méi)ff沖突.q以用LABEL 也不能在 FIML中用.label: SRESTRICT ;proc sydlln dataa 3sls