序列相关性补救措施与广义小二乘法概述.pdfVIP

AdditionalReading

Theseslidescontainadditionreadingforstudents

seekingtodevelopadeeperunderstandingofTopic2.

NOTE–Thismaterialisnotassessable.

AdditionalRemediesforSerial-

Correlation

•TheNewey-Westadjustmentprovidesan

approximateadjustmenttoreducethebiasin

thestandarderrors.

•Themoreformalapproachistouse

GeneralizedLeastSquares.

•

•ερε

–

•ρ

•ρY

•

•“quasi

differenced”

GeneralizedLeastSquares

1.Theerrortermisnotseriallycorrelated.

2.Theslopecoefficientβisthesameastheslopecoefficientof

1

theoriginalseriallycorrelatedequation.Thuscoefficients

estimatedwithGLShavethesamemeaningasthoseestimated

withOLS.

3.Thedependentvariablehaschanged.ThismeansthattheGLS

isnotdirectlycomparabletotheOLS.

4.ToforecastwithGLS,adjustmentsmayberequired(seelater

lectures).

•CannotuseOLStoestimateaGLSmodelbecauseGLS

equationsareinherentlynonlinearinthecoefficients.Instead

useeitherCochrane–OrcuttorAR(1)method.

TheCochrane–OrcuttMethod

•PerhapsthebestknownGLSmethod

•Thisisatwo-stepiterativetechniquethatfirstproducesan

estimateofρandthenestimatestheGLSequationusingthat

estimate.

•Thetwostepsare:

1.Estimateρbyrunningaregressionbasedontheresidualsoftheequation

suspectedofhavingserialcorrelation:

e=ρe+u

tt–1t

wheretheesaretheOLSresidualsfromtheequ