what is the difference between IV and 2SLS.
when they become equal.
2SLS and IV
Moderators: EViews Gareth, EViews Moderator
-
- Non-normality and collinearity are NOT problems!
- Posts: 3775
- Joined: Wed Sep 17, 2008 2:25 pm
Re: 2SLS and IV
2SLS is a generalization of IV. The two are the same when the number of instruments equals the number of coefficients.
Re: 2SLS and IV
can we say that they are equal when the equation is exactly identified .
In 2SLS we replace the endogenous variables with there estimated values. but in eviews , for 2SLS we are giving instruments.
In 2SLS we replace the endogenous variables with there estimated values. but in eviews , for 2SLS we are giving instruments.
-
- Non-normality and collinearity are NOT problems!
- Posts: 3775
- Joined: Wed Sep 17, 2008 2:25 pm
Re: 2SLS and IV
ramzan wrote:can we say that they are equal when the equation is exactly identified .
Yes, that's exactly the right thing to say.
ramzan wrote:In 2SLS we replace the endogenous variables with there estimated values. but in eviews , for 2SLS we are giving instruments.
The two methods are just equivalent ways of expressing the same underlying mathematics.
Re: 2SLS and IV
One more thing that IVLS can only be applied to exactly identified equation where as the 2SLS can be applied to exactly and over-identified equation as well.
Re: 2SLS and IV
GDP M2 GPDI FEDEXP TB6
3578.00 626.40 436.20 198.60 6.56
3697.70 710.10 485.80 216.60 4.51
3998.40 802.10 543.00 240.00 4.47
4123.40 855.20 606.50 259.70 7.18
4099.00 901.90 561.70 291.20 7.93
4084.40 1015.90 462.20 345.40 6.12
4311.70 1151.70 555.50 371.90 5.27
4511.80 1269.90 639.40 405.00 5.51
4760.60 1365.50 713.00 444.20 7.57
4912.10 1473.10 735.40 489.60 10.02
4900.90 1599.10 655.30 576.60 11.37
5021.00 1754.60 715.60 659.30 13.78
4913.30 1909.50 615.20 732.10 11.08
5132.30 2126.00 673.70 797.80 8.75
5505.20 2309.70 871.50 856.10 9.80
5717.10 2495.40 863.40 924.60 7.66
5912.40 2732.10 857.70 978.50 6.03
6113.30 2831.10 879.30 1018.40 6.05
6368.40 2994.30 902.80 1066.20 6.92
6591.90 3158.40 936.50 1140.30 8.04
6707.90 3277.60 907.30 1228.70 7.47
6676.40 3376.80 829.50 1287.60 5.49
6880.00 3430.70 899.80 1418.90 3.57
7062.60 3484.40 977.90 1471.50 3.14
7347.70 3499.00 1107.00 1506.00 4.66
7543.80 3641.90 1140.60 1575.70 5.59
7813.20 3813.30 1242.70 1635.90 5.09
8159.50 4028.90 1393.30 1678.80 5.18
8515.70 4380.60 1566.80 1705.00 4.85
8875.80 4643.70 1669.70 1750.20 4.76
3578.00 626.40 436.20 198.60 6.56
I have the following model
Y_1t=β_10+β_11 Y_2t+ β_12 X_(1t+) β_13 X_(2t)+ u_1t
Y_2t=β_20+β_21 Y_1t+u_2t
The first equation is the income equation where as the second equation is the money supply equation.
Y1:income
Y2: stock of money
X1:investment expenditure
X2: Government expenditure on goods and expenditure.
2SLS manually
Regress endogenous variable on the predetermined variable. In the first step we regress the Yit on the Xit and X2t.
Dependent Variable: Y1
Method: Least Squares
Date: 11/09/10 Time: 18:29
Sample: 1970 1999
Included observations: 30
Variable Coefficient Std. Error t-Statistic Prob.
C 2587.351 72.00106 35.93491 0.0000
X1 1.670732 0.164621 10.14897 0.0000
X2 1.969327 0.098368 20.02005 0.0000
R-squared 0.994756 Mean dependent var 5794.517
Adjusted R-squared 0.994368 S.D. dependent var 1520.432
S.E. of regression 114.1062 Akaike info criterion 12.40678
Sum squared resid 351546.3 Schwarz criterion 12.54690
Log likelihood -183.1016 F-statistic 2560.941
Durbin-Watson stat 0.427806 Prob(F-statistic) 0.000000
then I calculate the predicated value of the Y1t, using following eviews command.
forecast y1hat
to estimate the money supply function results ,Ireplace the Y1t with the y1hat
equation eq02.ls y2 c y1hat
Dependent Variable: Y2
Method: Least Squares
Date: 11/09/10 Time: 18:33
Sample: 1970 1999
Included observations: 30
Variable Coefficient Std. Error t-Statistic Prob.
C -2198.297 139.0986 -15.80388 0.0000
Y1HAT 0.791598 0.023248 34.05020 0.0000
R-squared 0.976419 Mean dependent var 2388.630
Adjusted R-squared 0.975577 S.D. dependent var 1214.819
S.E. of regression 189.8494 Akaike info criterion 13.39468
Sum squared resid 1009198. Schwarz criterion 13.48809
Log likelihood -198.9202 F-statistic 1159.416
Durbin-Watson stat 0.296090 Prob(F-statistic) 0.000000
but when I am doing this through Eviews built in method
equation eq03.tsls y2 c y1 @ c y1(-1)
Dependent Variable: Y2
Method: Two-Stage Least Squares
Date: 11/09/10 Time: 22:01
Sample (adjusted): 1971 1999
Included observations: 29 after adjustments
Instrument list: C Y1(-1)
Variable Coefficient Std. Error t-Statistic Prob.
C -2208.971 136.0338 -16.23840 0.0000
Y1 0.793461 0.022488 35.28357 0.0000
R-squared 0.978753 Mean dependent var 2449.397
Adjusted R-squared 0.977966 S.D. dependent var 1189.016
S.E. of regression 176.4973 Sum squared resid 841085.2
F-statistic 1244.930 Durbin-Watson stat 0.304181
Prob(F-statistic) 0.000000
the results are not same , why?
3578.00 626.40 436.20 198.60 6.56
3697.70 710.10 485.80 216.60 4.51
3998.40 802.10 543.00 240.00 4.47
4123.40 855.20 606.50 259.70 7.18
4099.00 901.90 561.70 291.20 7.93
4084.40 1015.90 462.20 345.40 6.12
4311.70 1151.70 555.50 371.90 5.27
4511.80 1269.90 639.40 405.00 5.51
4760.60 1365.50 713.00 444.20 7.57
4912.10 1473.10 735.40 489.60 10.02
4900.90 1599.10 655.30 576.60 11.37
5021.00 1754.60 715.60 659.30 13.78
4913.30 1909.50 615.20 732.10 11.08
5132.30 2126.00 673.70 797.80 8.75
5505.20 2309.70 871.50 856.10 9.80
5717.10 2495.40 863.40 924.60 7.66
5912.40 2732.10 857.70 978.50 6.03
6113.30 2831.10 879.30 1018.40 6.05
6368.40 2994.30 902.80 1066.20 6.92
6591.90 3158.40 936.50 1140.30 8.04
6707.90 3277.60 907.30 1228.70 7.47
6676.40 3376.80 829.50 1287.60 5.49
6880.00 3430.70 899.80 1418.90 3.57
7062.60 3484.40 977.90 1471.50 3.14
7347.70 3499.00 1107.00 1506.00 4.66
7543.80 3641.90 1140.60 1575.70 5.59
7813.20 3813.30 1242.70 1635.90 5.09
8159.50 4028.90 1393.30 1678.80 5.18
8515.70 4380.60 1566.80 1705.00 4.85
8875.80 4643.70 1669.70 1750.20 4.76
3578.00 626.40 436.20 198.60 6.56
I have the following model
Y_1t=β_10+β_11 Y_2t+ β_12 X_(1t+) β_13 X_(2t)+ u_1t
Y_2t=β_20+β_21 Y_1t+u_2t
The first equation is the income equation where as the second equation is the money supply equation.
Y1:income
Y2: stock of money
X1:investment expenditure
X2: Government expenditure on goods and expenditure.
2SLS manually
Regress endogenous variable on the predetermined variable. In the first step we regress the Yit on the Xit and X2t.
Dependent Variable: Y1
Method: Least Squares
Date: 11/09/10 Time: 18:29
Sample: 1970 1999
Included observations: 30
Variable Coefficient Std. Error t-Statistic Prob.
C 2587.351 72.00106 35.93491 0.0000
X1 1.670732 0.164621 10.14897 0.0000
X2 1.969327 0.098368 20.02005 0.0000
R-squared 0.994756 Mean dependent var 5794.517
Adjusted R-squared 0.994368 S.D. dependent var 1520.432
S.E. of regression 114.1062 Akaike info criterion 12.40678
Sum squared resid 351546.3 Schwarz criterion 12.54690
Log likelihood -183.1016 F-statistic 2560.941
Durbin-Watson stat 0.427806 Prob(F-statistic) 0.000000
then I calculate the predicated value of the Y1t, using following eviews command.
forecast y1hat
to estimate the money supply function results ,Ireplace the Y1t with the y1hat
equation eq02.ls y2 c y1hat
Dependent Variable: Y2
Method: Least Squares
Date: 11/09/10 Time: 18:33
Sample: 1970 1999
Included observations: 30
Variable Coefficient Std. Error t-Statistic Prob.
C -2198.297 139.0986 -15.80388 0.0000
Y1HAT 0.791598 0.023248 34.05020 0.0000
R-squared 0.976419 Mean dependent var 2388.630
Adjusted R-squared 0.975577 S.D. dependent var 1214.819
S.E. of regression 189.8494 Akaike info criterion 13.39468
Sum squared resid 1009198. Schwarz criterion 13.48809
Log likelihood -198.9202 F-statistic 1159.416
Durbin-Watson stat 0.296090 Prob(F-statistic) 0.000000
but when I am doing this through Eviews built in method
equation eq03.tsls y2 c y1 @ c y1(-1)
Dependent Variable: Y2
Method: Two-Stage Least Squares
Date: 11/09/10 Time: 22:01
Sample (adjusted): 1971 1999
Included observations: 29 after adjustments
Instrument list: C Y1(-1)
Variable Coefficient Std. Error t-Statistic Prob.
C -2208.971 136.0338 -16.23840 0.0000
Y1 0.793461 0.022488 35.28357 0.0000
R-squared 0.978753 Mean dependent var 2449.397
Adjusted R-squared 0.977966 S.D. dependent var 1189.016
S.E. of regression 176.4973 Sum squared resid 841085.2
F-statistic 1244.930 Durbin-Watson stat 0.304181
Prob(F-statistic) 0.000000
the results are not same , why?
Re: 2SLS and IV
in the Second data ,i have tried both manually and built in method. Parameters are same but the SE are different, please explain, File is attached .
thanks
thanks
- Attachments
-
- 2sls dimitrios.docx
- (27.5 KiB) Downloaded 529 times
-
- Non-normality and collinearity are NOT problems!
- Posts: 3775
- Joined: Wed Sep 17, 2008 2:25 pm
Re: 2SLS and IV
ramzan wrote:in the Second data ,i have tried both manually and built in method. Parameters are same but the SE are different, please explain, File is attached .
thanks
When you substitute the fitted values into the second stage, standard errors are computed incorrectly. It's not a software issue; doing this is called the "forbidden regression." That's why one should always use the TSLS command.
Re: 2SLS and IV
one more i want to ask is that is this correct that the IVLS can only be applied to exactly identified equations where as the 2SLS can be applied to over-identified equation and exactly identified equations.
-
- Non-normality and collinearity are NOT problems!
- Posts: 3775
- Joined: Wed Sep 17, 2008 2:25 pm
Re: 2SLS and IV
ramzan wrote:one more i want to ask is that is this correct that the IVLS can only be applied to exactly identified equations where as the 2SLS can be applied to over-identified equation and exactly identified equations.
That's a reasonable way to think about the difference, but some of this is just terminology. First, 2sls and IV are identical in the just identified case so 2sls in that case is just a different way of thinking about the estimator, it's not a different estimator. Second, you can redo 2sls as a straighforward instrumental variable estimator by using the fitted values from the first stage as instruments in IV.
-
- Posts: 1
- Joined: Fri May 17, 2019 7:57 pm
Re: 2SLS and IV
Dear all...
excuse me, how can I view the first stage result/report in eviews??
Thanks
excuse me, how can I view the first stage result/report in eviews??
Thanks
-
- Fe ddaethom, fe welon, fe amcangyfrifon
- Posts: 13317
- Joined: Tue Sep 16, 2008 5:38 pm
Re: 2SLS and IV
gilangwika wrote:Dear all...
excuse me, how can I view the first stage result/report in eviews??
Thanks
No software actually estimates in two stages. If you want to see results from a first stage regression, you have to run that regression separately.
Follow us on Twitter @IHSEViews
Return to “Econometric Discussions”
Who is online
Users browsing this forum: No registered users and 43 guests