Download presentation
Presentation is loading. Please wait.
1
Tutorial 2: Autocorrelation
Matthew Robson University of York Econometrics 2
2
Autocorrelation Autocorrelation emerges when the errors in different time-periods are correlated. When ๐๐๐ฃ ๐ข ๐ , ๐ข ๐ โ 0, ๐โ ๐ Positive Autocorrelation Negative Autocorrelation (Gujarati and Porter, 2009)
3
Assignment 7 Estimate the log-linear consumption function:
๐ฟ๐๐ ๐ถ ๐ก = ๐ฝ 0 + ๐ฝ 1 ๐ฟ๐๐ ๐ผ ๐ก + ๐ฝ 2 ๐ฟ๐๐ ๐ ๐ก + ๐ฝ 4 ๐ ๐ก + ๐ข ๐ก Where: ๐ถ ๐ก = consumption, ๐ผ ๐ก = real disposable income ๐ ๐ก = wealth, ๐ ๐ก = interest rate For the period 1967q1 โ 2002q4. (1)
4
Descriptive Statistics
5
Results ๐ฟ๐๐ ๐ถ ๐ก = ๐ฟ๐๐ ๐ผ ๐ก ๐ฟ๐๐ ๐ ๐ก +โ ๐ ๐ก + ๐ข ๐ก
6
Predicted Values
7
Autocorrelation
8
Question a) Test for autocorrelation using the Durbin-Watson test statistic given by PC-GIVE. What are the limitations of this test? How does the Breusch-Godfrey test overcome some of these limitations?
9
Durbin-Watson Test Defined as: ๐= ๐ก=2 ๐ก=๐ ๐ข ๐ก โ ๐ข ๐กโ ๐ก=1 ๐ก=๐ ๐ข ๐ก 2
10
Durbin-Watson Test Durbin-Watson Statistic: ๐ = 0.439 ๐=144, ๐ = 3
๐๐๐๐๐โ ๐๐๐ ๐กโ ๐๐๐ ๐กโฆโ๐
๐๐ ๐๐๐ข๐๐ ๐๐ข๐ก๐๐๐๐๐๐๐๐๐ก๐๐๐๐ , ๐๐๐๐ก๐๐๐๐ก๐๐๐ข ๐๐๐ ๐ท๐ Durbin-Watson Statistic: ๐ = 0.439 ๐=144, ๐ = 3 ๐ผ=0.05 โ ๐ ๐ข =1.774, ๐ ๐ฟ =1.693 ๐ผ=0.01โ ๐ ๐ข =1.665, ๐ ๐ฟ =1.584 ๐ป 0 : no +ve autocorrelation, ๐ป 0 โ : no -ve autocorrelation The ๐ statistic is less than the critical ๐ ๐ฟ value โด we reject the null hypothesis ( ๐ป 0 ) of no +ve correlation at both 5% and 1% levels.
11
Question a) Limitations of Durbin-Watson statistic
Lagged residuals only to first order Zones of indecision Not appropriate when lagged dependant variable is included The Breusch-Godfrey test allows: Higher order autocorrelation Still appropriate when a lagged dependant variable is included
12
Question b) Test for autocorrelation using the Breusch-Godfrey test statistic given by PC-GIVE. What (default) order of autocorrelation is being tested for here?
13
Question b) ๐๐๐๐๐โ ๐๐๐ ๐กโ ๐๐๐ ๐กโฆโ ๐ธ๐๐๐๐ ๐๐ข๐ก๐๐๐๐๐๐๐๐๐ก๐๐๐ ๐ก๐๐ ๐ก
The order of the autocorrelation being tested here is 5th, e.gโฆ ๐ข ๐ก = ๐ 1 ๐ข ๐กโ1 + ๐ 2 ๐ข ๐กโ2 + ๐ 3 ๐ข ๐กโ3 + ๐ 4 ๐ข ๐กโ4 + ๐ 5 ๐ข ๐กโ5 + ๐ ๐ก Test statistic is ~ ๐ ๐ = , ๐ = โด We reject ๐ป 0 (of no autocorrelation) at both 5% and 1% levels.
14
Question c) Construct the Breusch-Godfrey test for up to second order autocorrelation and test using the F statistic.
15
Breusch-Godfrey Test Method
Estimate the model and save the residuals ( ๐ข ๐ก ) Estimate: ๐ข ๐ก = ๐ผ 0 + ๐ผ 1 log ๐ผ ๐ก + ๐ผ 2 log ๐ ๐ก + ๐ผ 3 ๐ ๐ก + ๐ผ 4 ๐ข ๐กโ1 + ๐ผ 5 ๐ข ๐กโ2 + ๐ ๐ก Note the ๐
2 from Step 2 and calculate the ๐ 2 test statistic as: ๐โ๐ ๐
2 ~ ๐ ๐ 2 Where: ๐ = ๐๐ข๐๐ ๐๐. ๐๐ ๐๐๐ ๐๐๐ฃ๐๐ก๐๐๐๐ ๐ = ๐๐ข๐๐๐๐ ๐๐ ๐๐๐๐ (๐๐ ๐กโ๐ ๐๐๐๐๐ ๐๐ ๐๐ข๐ก๐๐๐๐๐๐๐๐๐ก๐๐๐) Compare the test statistic from Step 3 with the ๐ 2 critical values at the 5% and 10% levels.
16
Breusch-Godfrey Test Construct the Breusch-Godfrey test, for second order autocorrelation, e.g. ๐ข ๐ก = ๐ 1 ๐ข ๐กโ1 + ๐ 2 ๐ข ๐กโ2 ๐ป 0 : ๐๐ ๐๐ข๐ก๐๐๐๐๐๐๐๐๐ก๐๐๐, ๐ 1 = ๐ 2 =0 ๐ป 1 : ๐ป 0 ๐๐ ๐๐๐๐ ๐ Test Statistic: ๐โ๐ ๐
2 = 44โ ~ ๐ 2 2 ๐ =5.991 ๐ =9.210 โด We reject the null hypothesis of no autocorrelation at both 5% and 1% ๐ข ๐ก = ๐ผ 0 + ๐ผ 1 log ๐ผ ๐ก + ๐ผ 2 log ๐ ๐ก + ๐ผ 3 ๐ ๐ก + ๐ผ 4 ๐ข ๐กโ1 + ๐ผ 5 ๐ข ๐กโ2 + ๐ ๐ก
17
Breusch-Godfrey Test (F-test)
Method Estimate the model and save the residuals ( ๐ข ๐ก ) Estimate two factor models: RES: ๐ข ๐ก = ๐ผ 0 + ๐ผ 1 log ๐ผ ๐ก + ๐ผ 2 log ๐ ๐ก + ๐ผ 3 ๐ ๐ก + ๐ ๐ก UNRES: ๐ข ๐ก = ๐ผ 0 + ๐ผ 1 log ๐ผ ๐ก + ๐ผ 2 log ๐ ๐ก + ๐ผ 3 ๐ ๐ก + ๐ผ 4 ๐ข ๐กโ1 + ๐ผ 5 ๐ข ๐กโ2 + ๐ ๐ก Over the same sample i.e. ๐โ2 = ๐โ๐ = 144โ2 Undertake the F-test for: ๐ป 0 : ๐ผ 4 = ๐ผ 5 =0 ๐น= ๐
๐๐๐
๐ธ๐ 2 โ ๐
๐
๐ธ๐ 2 ๐ 1โ ๐
๐๐๐
๐ธ๐ 2 ๐ โ โ๐พ Where: ๐ =๐๐. ๐๐ ๐๐๐ ๐ก๐๐๐๐ก๐๐๐๐ ๐ = ๐๐. ๐๐ ๐๐๐๐๐๐๐ก๐๐๐ ๐๐ ๐๐๐
๐ธ๐ ๐ โ = ๐๐. ๐๐ ๐๐๐ ๐๐๐ฃ๐๐ก๐๐๐๐ ๐๐ ๐กโ๐ ๐๐๐
๐ธ๐ ๐ ๐๐๐๐๐ (๐โ๐)
18
Breusch-Godfrey Test (F-test)
Construct the Breusch-Godfrey test, for second order autocorrelation. ๐ป 0 : ๐๐ ๐๐ข๐ก๐๐๐๐๐๐๐๐๐ก๐๐๐, ๐ผ 4 = ๐ผ 5 =0 ๐ป 1 : ๐ป 0 ๐๐ ๐๐๐๐ ๐ Test Statistic: ๐น= ๐
๐๐๐
๐ธ๐ 2 โ ๐
๐
๐ธ๐ 2 ๐ 1โ ๐
๐๐๐
๐ธ๐ 2 ๐ โ โ๐พ ๐น= โ ร 10 โ โ โ6 ๐น= ~ ๐น ๐, ๐ โ โ๐พ ๐น 2, โ3.07, ๐น 2, โ4.79 โด We reject the null hypothesis of no autocorrelation at both 5% and 1% ๐
๐ธ๐: ๐ข ๐ก = ๐ผ 0 + ๐ผ 1 log ๐ผ ๐ก + ๐ผ 2 log ๐ ๐ก + ๐ผ 3 ๐ ๐ก + ๐ ๐ก ๐๐๐
๐ธ๐: ๐ข ๐ก = ๐ผ 0 + ๐ผ 1 log ๐ผ ๐ก + ๐ผ 2 log ๐ ๐ก + ๐ผ 3 ๐ ๐ก + ๐ผ 4 ๐ข ๐กโ1 + ๐ผ 5 ๐ข ๐กโ2 + ๐ ๐ก
19
Question c) ๐๐๐๐๐โ ๐๐๐ ๐กโ ๐๐๐ ๐กโฆโ ๐ธ๐๐๐๐ ๐๐ข๐ก๐๐๐๐๐๐๐๐๐ก๐๐๐ ๐ก๐๐ ๐ก (โฒ๐ก๐ ๐๐๐โฒ = 2) Test Statistic: ~ ๐ 2 2 ๐ =5.991 ๐ =9.210 โด We reject the null hypothesis of no autocorrelation at both 5% and 1%
20
Question d) What are the consequences of your findings for the usefulness of the standard Ordinary Least Squares results for the consumption function above?
21
Question d) Consequences of autocorrelation
OLS estimators are LUE but not BLUE (most efficient and unbiased) The estimated variances of OLS estimators are biased Usual ๐ก and ๐น tests are unreliable The usual formula to compute the error variance is a biased estimator of the true ๐ 2 The conventionally computed ๐
2 may be an unreliable measure of the true ๐
2
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.