You are on page 1of 4

University of East London Dr. Derick Boyd Business School Finance, Economics and Accounting FE !!

! A""lied Econometrics #$E B%EUS&$'()DF%E* LA(%A+(E ,UL#-.L-E% #ES# F)% SE%-AL &)%%ELA#-)+ %eferences The EViews User's Guide available Online. Studenmund, A.H. (200 !, Usin" E#$n$metri#s % A &ra#ti#al Guide, Addis$n%'esle(%)$n"man, **.+ +%+ ,. Th$mas -.). .ntr$du#t$r( E#$n$metri#s/ The$r( and A**li#ati$ns 2nd editi$n, )$n"man 00+ (**. 01 % 0,, especially equations 5.31 - 5.342 **. 13 % 3 , especially pp. 70 - 71! -ntroduction 'e have been intr$du#ed t$ the 4urbin%'ats$n test 5$r 5irst $rder serial #$rrelati$n and the 4urbin h%test 5$r serial #$rrelati$n in the *resen#e $5 a la""ed de*endent variable (see &ind(#6 and -ubin5eld 00,, 7th edn, **. 10% 30 5$r a "$$d e8*lanati$n and an a""re"ate #$nsum*ti$n re"ressi$n e8am*le!. The )9 test is *arti#ularl( use5ul be#ause it is n$t $nl( suitable 5$r testin" 5$r aut$#$rrelati$n $5 an( $rder, but als$ suitable 5$r m$dels with $r with$ut la""ed de*endent variables. The )9 test, as $ther tests, uses the estimated residuals in #$nstru#tin" the test. .t is w$rth remindin" $urselves, h$wever, that aut$re"ressi$n #an be the result $5 mis% s*e#i5i#ati$n $5 the m$del and n$t "enuine aut$#$rrelati$n due t$ the behavi$ural #hara#teristi#s $5 the residuals. Studenmund (200 / + +%+ ,! *r$vides a "$$d e8*lanati$n $5 aut$#$rrelati$n as indi#atin" *r$blems in the s*e#i5i#ati$n $5 a m$del. Essentially, this rests on the fact that, economic variables are usually autocorrelated and if such a relevant variable effect is included in the stochastic term, then the stochastic term ill to that e!tent become autocorrelated. .5 mis%s*e#i5i#ati$n is the real #ause $5 the aut$#$rrelati$n, then usin" :$#hrane ; Or#utt, Hildreth%)iu, Generalised )east S<uares $r s$me $ther aut$re"ressive meth$d t$ #$rre#t 5$r aut$re"ressi$n is n$t a**r$*riate, rather the m$del sh$uld be #$rre#tl( s*e#i5ied. .n dia"n$sti# tests, aut$#$rrelati$n $5 $rder p is #h$sen as 5$ll$ws/ p= 5$r undated and annual data p = 2 5$r hal5 (earl( data p = 7 5$r <uarterl( data p = 2 5$r m$nthl( data Other values 5$r p #an usuall( be s*e#i5ied usin" an $*ti$n.

%7

2 #he Breusch'(odfrey Lagrange ,ulti"lier #est for Serial &orrelation 'e will illustrate this test with re5eren#e t$ a se#$nd $rder aut$re"ressive s#heme. Su**$se that we have a m$del, >t = ? 2 @2t ? + @+t ? t .......... ( ! .............. (2! .............. (+!

and we sus*e#t a se#$nd $rder aut$re"ressive s#heme/ t = t% ? 2 t%2 ? t Then the m$del #$uld be written as/ >t = ? 2 @2t ? + @+t ? t% ? 2 t%2 ? t

This we #$uld term the unrestricted 5$rm $5 the m$del. .t is unrestri#ted be#ause we d$ n$t restri#t the 5$rm the err$r term ma( ta6e, i5 it is an inde*endent rand$m err$r the value will be Aer$, i5 the err$r is aut$%#$rrelated u* t$ a se#$nd $rder then $ne $r b$th $5 the will be n$n%Aer$. .5 we estimated the e<uati$n as, >t = ? 2 @2t ? + @+t ? t ................ (7!

This w$uld be a restricted 5$rm $5 the e<uati$n, sin#e im*li#it in this 5$rm is the restri#ti$n that = 2 = 0. 'e #an de5ine a 2 variable with h de"rees $5 5reed$m as2 (SS-- % SS-U ! /h ............... (B! C2 'here h is de"ree $5 5reed$m, whi#h is the number $5 restri#ti$ns (in this 2nd $rder e8am*le h=2!, SS- the sum $5 s<uares residuals 5$r the restri#ted and unrestri#ted e<uati$ns and C2the estimated varian#e $5 the restri#ted e<uati$n. 'e #an 5urther sh$w that, (SS-- % SS-U ! = C2SST % SSSSTDn = n-2 = #%/

'here -2 #$mes 5r$m au8iliar( re"ressi$n % "an e!planation of hy this relationship holds can be found in #homas p. 70, but you $ust need to remember the n-2 = T-2 result%. 'e ma(, there5$re, $btain a )9 test statisti#s in T-2 in $rder t$ #arr( $ut the test. The *r$#edure 5$r $btainin" the ab$ve test statisti#s is as 5$ll$ws. . Estimate the restricted e<uati$n and retain the residuals (et!2

+ 2. Use the residuals (et! as the de*endent variable 5$r the unrestricted re"ressi$n in an au!iliary re&ression/ et = ? 2 @2t ? + @+t ? et% ? 2 et%2 +. Ta6e T-2 and #$m*are it with the relevant #riti#al value 5$r 2 where the de"rees $5 5reed$m, h, is the order of the autoregressive scheme.

The null hypothesis of the test is, Ho: no autocorrelation. We reject the null hypothesis if TR2 > 2h critical value. &lass E0ercise 'e will #arr( $ut this e8er#ise 5$r s$me$ne's m$del wh$ #an *r$vide us with <ui#6 a##ess t$ the estimati$n $5 their #$untr( e<uati$nEE #utorial 1uestion A 9$ne( 4emand 5un#ti$n l$"( ' t ! = + 2 l$"(()*t ! + + +,t + 7 l$" *+t + t (+. ! was estimated usin" <uarterl( data 5$r the United States $ver the *eri$d 0B2/ % 002/7 (T= 1+ $bservati$ns!. 'here/ l$"( ' t ! is the l$"arithm $5 the narr$w m$ne( su**l( in time *eri$d t2 l$"(()*t ! is the l$"arithm $5 the real G4&2 +,t is the interest rate2 l$" *+t is the rate $5 in5lati$n measured here as the di55eren#es in the l$"arithm $5 the #$nsumer *ri#e inde82 t is the residual. The e#$n$metri#ian wanted t$ #$ndu#t the Freus#h%G$d5re( Serial :$rrelati$n )a"ran"e 9ulti*lier and she ran the e<uati$n/ (+.2! et = + 2 l$"(()*t ! + + +,t + 7 l$" *+t + et + vt the results $5 whi#h are re*$rted bel$w.

Test Equation: Dependent Variable: RESID Method: Least Squares Date: 11/28/02 Ti e: 1!:21 "resa ple issin# $alue la##ed residuals set to %ero& Variable 'oe((i)ient Std& Error t*Statisti) ' *0&00!+,, 0&01+0+1 *0&-8.!8+ L/010D"2 0&00033. 0&002!-, 0&+.!323 RS *0&000,!. 0&001018 *0&,,!.-8 DL/01"R2 0&-0-1-+ 0&+81!.! 1&0,88!RESID1*12 0&320+0! 0&0+22.! 28&,1+2! R*squared 0&8+.282 Mean dependent $ar 4d5usted R*squared 0&8++1!+ S&D& dependent $ar S&E& o( re#ression 0&022-,2 46ai6e in(o )riterion Su squared resid 0&0.3!-3 S)h7ar% )riterion Lo# li6elihood +30&0,8, 8*statisti) Durbin*9atson stat 1&..03!, "rob18*statisti)2

"rob& 0&!2!, 0&.0!. 0&,.8, 0&231+ 0&0000 1&21E*1, 0&0,-3!3 *-&.2-!-*-&!23.-20+&2,1, 0&000000

'here, et are the residual $5 the 9$ne( 4emand e<uati$n (+. !, et are residuals la""ed $ne *eri$d, vt are white n$ise err$rs in this e<uati$n, and the $ther variables are de5ined as ab$ve. (a! E8*lain the Freus#h%G$d5re( Serial :$rrelati$n )a"ran"e 9ulti*lier test ma6in" #lear what $rder $5 test is bein" #arried $ut. (b! :arr( $ut the test and e8*lain the #$n#lusi$ns ($u arrive at with re"ard t$ the serial #$rrelati$n. GGGGG HmmmE
:reus)h*0od(re; Serial 'orrelation LM Test: 8*statisti) 81+&00!0 "robabilit; /bs<R*squared 1+!&-..0 "robabilit; 0&000000 0&000000

You might also like