Scribd
Upload
2 views9 pages

Autocorrelation 2024

Uploaded by

HK gamer
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
2 views9 pages

Autocorrelation 2024

Uploaded by

HK gamer
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 9

18/12/2024

Autocorrelation

Dr Arshad Ali Bhatti


Fall, 2024

Autocorrelation

 That the error term from one time period


depends, in some systematic way, on error
terms from other time periods.
 E(UtUt-1)≠0 for all t ≠ t-1
 It is not the relationship b/w different
variables rather b/w the successive values of
the same variable.

Dr Arshad Ali Bhatti/ Spring 2019

1
18/12/2024

Reasons

 Spatial Autocorrelation
 Prolonged influence of shocks
 Inertia (Tendency to remain unchanged)
 Data manipulation
 Misspecification

Dr Arshad Ali Bhatti/ Spring 2019

Consequences

If AC is ignored and OLS is used to estimate


the parameters then:
 The estimates and forecasts will still be
unbiased and consistent.
 The estimates and forecasts will be
inefficient.
 The estimated variances of regression
coefficients will be biased, and hence tests
of hypotheses are invalid.

Dr Arshad Ali Bhatti/ Spring 2019

2
18/12/2024

Detection

 Durbin-Watson d-test
 Durbin-h
 Artificial Regression

Dr Arshad Ali Bhatti/ Spring 2019

Durbin Watson d-Test

Limitations
 Test is for first order serial correlation.

 Test is inconclusive if the computed value of


d- statistic lies b/w dL and dU.
 Test cannot be applied in models with lagged
dependent variables.

Dr Arshad Ali Bhatti/ Spring 2019

3
18/12/2024

Cont.

 TEST statistic (d):


Σ tn= 2 (et − et −1 ) 2
d=
Σet2
d ≈ 2(1 − ρˆ )
Σe e
where ρˆ = t 2t −1
Σet

Dr Arshad Ali Bhatti/ Spring 2019

Cont.

How to check AC?


 ρ^=0 suggests d=2 No AC
 ρ^=1 suggests d=0 +ve AC
 ρ^=-1 suggests d=4 -ve AC

 When the calculated value of d-statistic is close to 2,


we accept the H0 of no AC.
 When it is close to zero or 4, we reject H0.

Dr Arshad Ali Bhatti/ Spring 2019

4
18/12/2024

Durbin Watson d Statistic

Dr Arshad Ali Bhatti/ Spring 2019

Durbin h-statistic

 Whenever one of the explanatory variable is the


lagged value of the dependent variable e.g. Yt-1, Yt-2
etc.

n
h = ρˆ
1 − n. var(bi )
where var(bi ) = Least square estimate of
the coefficient of lagged variable.
ρˆ = 1 − 0.5 d

Dr Arshad Ali Bhatti/ Spring 2019

10

5
18/12/2024

Cont.

 h Statistic is normally distributed with mean


zero and variance of one.
 If the p-value for h Statistic is low, we reject
H0 of NO Autocorrelation.

Dr Arshad Ali Bhatti/ Spring 2019

11

Problem with h-statistic

 When n{var(bi)} >1


i.e negative sign in the sq-root , h Statistic
cannot be calculated from the above formula.
 One possible solution is to run the artificial
regression.

Dr Arshad Ali Bhatti/ Spring 2019

12

6
18/12/2024

Artificial Regression

 Regress the LS residual et on et-1 plus all


the independent variables, including the
lagged dependent variable.
e.g.: Yt =b1 +b2 X2t + b3 Yt-1+ et (Actual Reg.)
et =α1 + α2 et-1+ α3 X2t + α4 Yt-1+ vt
(Artificial Regression)
 Test H0: No AC,
by the significance of the coefficient of et-1
using t-test.

Dr Arshad Ali Bhatti/ Spring 2019

13

Solutions

Dr Arshad Ali Bhatti/ Spring 2019

14

7
18/12/2024

Generalized Least Square (GLS)

Theory:
 Say the original model is:
Yt = β1+ β2 X2t + ut
with first-order autoregressive specification on Ut.
i.e.

u t = ρ u t −1 + vt where − 1 ≤ ρ ≤ +1

Dr Arshad Ali Bhatti/ Spring 2019

15

Cont.

 Steps:
 Transform the model with the autoregressive
disturbance term in to a model with a non-
autoregressive disturbance term.
 That is Yt*= Yt - ρ^Yt-1 and X2t*=X2t-ρ^X2t-1
 Regress Y* on X* using OLS
 The GLS estimates from this transformed
model are efficient and consistent.

Dr Arshad Ali Bhatti/ Spring 2019

16

8
18/12/2024

Applications:

 In practice we add AR(1) specification as an


explanatory variable in to the actual model
and re-estimate it……. d Statistic is likely to
improve unless there is any other problem
like non-stationarity etc.
 Discuss Interest rate model (EX47 Pindyck)
 Discuss dynamic consumption function (data
Ex67) and use Durbin h Statistic to check AC.

Dr Arshad Ali Bhatti/ Spring 2019

17

References:

 Gujrati D. N., Basic Econometrics, 4th ed., McGraw


Hill.
 Kennedy Peter, A Guide to Econometrics, 4th ed.,
McGraw Hill.
 Pindyck & Rubinfeld, Econometric Models and
Economic Forecasts, 4th ed., McGraw Hill.
 Thomas R.L., Introductory Econometrics, 2nd ed.,
Longman.
 Johnson & Johnson, Basic and Applied
Econometrics, MacMillan.

Dr Arshad Ali Bhatti/ Spring 2019

18

You might also like