Upload
tobias-tufts
View
218
Download
2
Embed Size (px)
Citation preview
Christopher Dougherty
EC220 - Introduction to econometrics (chapter 9)Slideshow: two-stage least squares
Original citation:
Dougherty, C. (2012) EC220 - Introduction to econometrics (chapter 9). [Teaching Resource]
© 2012 The Author
This version available at: http://learningresources.lse.ac.uk/135/
Available in LSE Learning Resources Online: May 2012
This work is licensed under a Creative Commons Attribution-ShareAlike 3.0 License. This license allows the user to remix, tweak, and build upon the work even for commercial purposes, as long as the user credits the author and licenses their new creations under the identical terms. http://creativecommons.org/licenses/by-sa/3.0/
http://learningresources.lse.ac.uk/
TWO-STAGE LEAST SQUARES
wuxUpw 4321
1
In a second variation, we shall consider the model shown above. x is the rate of growth of productivity, assumed to be exogenous. w is now hypothesized to be a function of p, U, and x. For simplicity, p has reverted to being a simple function of w.
puwp 21
wuxUpw 4321
22
22423211
1
wp uuxUp
22
243121
1
wp uuxUw
puwp 21
2
The reduced form equations are now as shown. The equation for w is underidentified because there is no exogenous variable to act as an instrument for p. Both U and x are already in the equation in their own right and are therefore unavailable.
TWO-STAGE LEAST SQUARES
wuxUpw 4321
22
22423211
1
wp uuxUp
22
243121
1
wp uuxUw
puwp 21
3
However, the equation for p is now overidentified. An equation is said to be overidentified if the number of available instruments exceeds the number of endogenous explanatory variables used as explanatory variables.
TWO-STAGE LEAST SQUARES
wuxUpw 4321
22
22423211
1
wp uuxUp
22
243121
1
wp uuxUw
puwp 21
4
We could use U as an instrument.
TWO-STAGE LEAST SQUARES
wuxUpw 4321
22
22423211
1
wp uuxUp
22
243121
1
wp uuxUw
puwp 21
5
Equally, we could use x as an instrument. Both estimators are consistent. Hence, in large samples they will yield the same estimate - the true value. But in finite samples they will be different because they have different error components.
TWO-STAGE LEAST SQUARES
wuxUpw 4321
22
22423211
1
wp uuxUp
22
243121
1
wp uuxUw
puwp 21
6
If we had to choose between them, it would be rational to choose the more efficient, that is, that with the smaller population variance. We would therefore choose the instrument more highly correlated with w.
wwUU
ppUUb
ii
iiIV2
2,
2
2
2 1IV2
wUw
u
b rnp
2,
2
2
2 1IV2
wxw
u
b rnp
wwxx
ppxxb
ii
iiIV2
TWO-STAGE LEAST SQUARES
wuxUpw 4321
22
22423211
1
wp uuxUp
22
243121
1
wp uuxUw
puwp 21
7
2,
2
2
2 1IV2
wUw
u
b rnp
2,
2
2
2 1IV2
wxw
u
b rnp
But we do not need to choose between them. Instead, we will combine them into a super-instrument more highly correlated with w than either individually.
wwUU
ppUUb
ii
iiIV2
wwxx
ppxxb
ii
iiIV2
TWO-STAGE LEAST SQUARES
wuxUpw 4321
22
22423211
1
wp uuxUp
22
243121
1
wp uuxUw
puwp 21
8
Let us denote this super-instrument Z. It will be a linear function of U and x. How do we choose h1, h2, and h3 so as to maximize the correlation between Z and w?
xhUhhZ 321
TWO-STAGE LEAST SQUARES
wuxUpw 4321
22
22423211
1
wp uuxUp
22
243121
1
wp uuxUw
puwp 21
9
Very simple. Regress the reduced form equation for w using OLS. When you do this, the coefficients are chosen so as to minimize the sum of the squares of the residuals.
xhUhhZ 321 xhUhhw 321ˆ
TWO-STAGE LEAST SQUARES
wuxUpw 4321
22
22423211
1
wp uuxUp
22
243121
1
wp uuxUw
puwp 21
10
But, as we saw in Chapter 1, the OLS coefficients are optimal with respect to two other criteria of goodness of fit. They maximize R2, and they maximize the correlation between the actual and fitted values of the dependent variable.
xhUhhZ 321 xhUhhw 321ˆ
TWO-STAGE LEAST SQUARES
wuxUpw 4321
22
22423211
1
wp uuxUp
22
243121
1
wp uuxUw
puwp 21
11
Since the fitted values are calculated as the linear function of U and x most highly correlated with w, they are exactly what we need.
wZ ˆ
xhUhhZ 321 xhUhhw 321ˆ
TWO-STAGE LEAST SQUARES
wuxUpw 4321
22
22423211
1
wp uuxUp
22
243121
1
wp uuxUw
puwp 21
xhUhhZ 321 xhUhhw 321ˆ
wZ ˆ
12
This is the Two-Stage Least Squares estimator. The first stage is to regress the reduced form equation for the endogenous variable being used as an explanatory variable. The second is to save the fitted values from this regression and use them as the instrument.
wwww
ppwwb
ii
ii
ˆˆ
ˆˆTSLS2
TWO-STAGE LEAST SQUARES
wuxUpw 4321
22
22423211
1
wp uuxUp
22
243121
1
wp uuxUw
puwp 21
13
The TSLS estimator satisfies the usual requirements. It is correlated with w - indeed it is the function of U and x most highly correlated with w. Since U and x are both exogenous, it is independent of up. And since U and x are not regressors, it is available as an instrument.
wZ ˆ
xhUhhZ 321 xhUhhw 321ˆ
wwww
ppwwb
ii
ii
ˆˆ
ˆˆTSLS2
TWO-STAGE LEAST SQUARES
wuxUpw 4321
22
22423211
1
wp uuxUp
22
243121
1
wp uuxUw
puwp 21
14
Finally, note that the TSLS estimator, and indeed any IV estimator, will yield consistent estimates only if the instruments are truly exogenous. If they turn out to be endogenous, they are unlikely to be independent of the disturbance terms in the model.
wZ ˆ
xhUhhZ 321 xhUhhw 321ˆ
wwww
ppwwb
ii
ii
ˆˆ
ˆˆTSLS2
TWO-STAGE LEAST SQUARES
pumwp 321 wuUpw 321
22
2323211
1
wp uumUp
22
2323121
1
wp uumUw
15
In the previous example, is m likely to be exogenous? Probably not. In the long run, at least, an increase in prices is likely to be one of the determinants of an increase in the money supply.
TWO-STAGE LEAST SQUARES
pumwp 321 wuUpw 321
22
2323211
1
wp uumUp
22
2323121
1
wp uumUw
16
Finding truly exogenous variables to act as instruments is often a serious problem when fitting simultaneous equations models.
TWO-STAGE LEAST SQUARES
Copyright Christopher Dougherty 2011.
These slideshows may be downloaded by anyone, anywhere for personal use.
Subject to respect for copyright and, where appropriate, attribution, they may be
used as a resource for teaching an econometrics course. There is no need to
refer to the author.
The content of this slideshow comes from Section 9.3 of C. Dougherty,
Introduction to Econometrics, fourth edition 2011, Oxford University Press.
Additional (free) resources for both students and instructors may be
downloaded from the OUP Online Resource Centre
http://www.oup.com/uk/orc/bin/9780199567089/.
Individuals studying econometrics on their own and who feel that they might
benefit from participation in a formal course should consider the London School
of Economics summer school course
EC212 Introduction to Econometrics
http://www2.lse.ac.uk/study/summerSchools/summerSchool/Home.aspx
or the University of London International Programmes distance learning course
20 Elements of Econometrics
www.londoninternational.ac.uk/lse.
11.07.25