(OWPTERTWELVE. ALTOCORRICATION 471
va J
we
' J ba
125 RELATIONSHIP BETWEEN WAGES AND PRODUCTIVITY 4 ef
ii BUSINESS SECTOR OFTHE UNTED STATES 15 ane ; i
Now that we have discussed the consequences of autocorrelation, the obvi- er e
cus guesioni, How dowedtas and do econ fr 2 :
Sunt the ops tinue to conser once sample ee
Gresdaon
os
°
ai
FIGURE 12.7 xf campansat (¥ end index prodcty (Unie Eats, 1059-108
TABLE 124
although there is sume
tionship between the two
tocestimatea linear:
he
For discussion purposes, we will call
productivity regressions.v
(ATA. ex TWO! ReLARING THE ASSLNETIONS OFTHE CLASSICAL MODEL
oune 128
Alternatively, we ean plot the standardized residuals against time, which
arealso shown in Fi
Simply the residual
(2B it they
‘mately unit variance.” In large
distributed with zero mean and u
‘Examining the time sequence
{hrm puna test
OAPTERTWELE: AUTOCORRELTION 475
‘Table 12.5. As this figure reveals, most of the residuals are bunched in the
ure 12.32. A prior, then, Figure 12.8 woul
the residuals.
CERT owt po aunts abot ie pasa) dacbton4476 PART TWO RELAKNG THE ASSUMPTIONS OF HE CLASSICAL UODEL
Now let
tables, the reader can verify that the residuals in our wages-prodctvty re-
‘Then under the null hypothesis that the successive outcomes (here, resid F is te a ‘actually they are positively correlated.
als) are independent, and assuming that N= 10, the umber
of ransisasyptoialy) normal dibaied wi
H.Durtin-wateon Tea A.
tot, celebrate tet fr detecing serial corclation la that developed
(126.2) ! Hanis aes Destin wal Goene opel kabens te Dates
Watson d statistic, which is defined as
i? 0263)
ll hypothesis of randomness is sustainable, following the proper-
‘normal ion, we should expect tha
RSEIR)+1960—)=095 (12.6.3)
Decision Rue. nt ec hen potest andomnes wit 95% cotonce FF
‘ers crn, fs tho precedngconiseace inva cs ea cnn 40,
‘train fs ose tao es (No: You can choose any loa Contre tne
very :
(12.6.4)
reject the hypoth.
ur wages-productivity regression are random7
(470 exe RELIG THE ASQUMPTIONS OF HE GLASSCAL MODEL
cenit regression model does not include the lagged value(s) of the de-
Pendent variable as one of the explanatory variables Thus,
Dlicable in models ofthe following type:
Ym A+ BN + Bika t+ Xe t Morty (12.66)
Ire ¥- the one period lagged value of ¥; Such models are known as
sampling or probability distribution of the d statistic given in
difficult to derive because, as Durbin and have shown, it
‘2 complicated way on the X values
an BRA 2B aan
Sines Ei and Sa differ in only one observation, they are approxi
tnutely equal. Therefore, setting Fa! , SG, (12.6.7) may be wiles o¢
2 (122 BABens
am2(1- 288-1) (126.8)
where ~ means approximately.
sata tn: fi Gar Koro, Lando Mate, sd van Suey Pr
ecm by Fr agin 199 po Re ae
"at se the scsson onthe “eac”DaWatn tat given ater ihe section,
dea Ci-3) magaeay
But since -1 < p < 1, (12,6.10) implies that
dees posse caste
limits.
Itis apparent from
‘comparison with
‘mains a unique value for any given regression.400. me Twor Renna HE ARSUNETIONS
mau 128
ymparatively larger than the
ics of the Durbin-Watson test are as fol
Si ae and
he computed df 0.1299hes below de
there pote seal conelaton ne red
Nanagaie craton
No astern, pote rage
»
Curren TWELYE: AITOCORRELATION 48%
Ss Retl p 0 versus His <0. Reject Hy at a level if the estimated
{oi dys that is, there Is statistically significant evidence of negative
‘autocorrelation.
Shere = 0 versus Hi: 4 0.Reject He at 2a level ifd