{[ promptMessage ]}

Bookmark it

{[ promptMessage ]}

# panelslides - Multiple Equation GMM with Common Coecients...

This preview shows pages 1–13. Sign up to view the full content.

Multiple Equation GMM with Common Coe cients: Panel Data Eric Zivot November 25, 2009

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Multi-equation GMM with common coe cients Example (panel wage equation) LW 69 i = φ + βS 69 i + γIQ i + πEXPR 69 i + ε i 1 LW 80 i = φ + 80 i + i + 80 i + ε i 2 Note: common coe cients across two equations and cov ( ε i 1 i 2 ) 6 =0
The general multi-equation model with common coe cients δ is y im = z 0 im δ + ε im ,i =1 ,...,n ; m ,...,M E [ z im ε im ] 6 = 0 fo rsomee lemento f z im Themode lfo rthe m th equation is y m n × 1 = Z m δ + ε m ,m

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
The stacked system (across equations) is y 1 = Z 1 δ + ε 1 . . . y M = Z M δ + ε M y 1 . . . y M = Z 1 . . . Z M δ + ε 1 . . . ε M This is in the form of a giant regression y Mn × 1 = Z δ + ε
Assume each equation m has k m instruments x im satisfying E [ x im ε im ]=0 Very often the same instruments x i are used for each equation. The moment conditions are g i ( δ )= x i 1 ( y i 1 z 0 i 1 δ ) . . . x iM ( y iM z 0 iM δ ) There are a total of K = P M m =1 k m moment conditions.

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Identi f cation E [ g i ( δ )] = 0 for δ = δ 0 6 = 0 for δ 6 = δ 0 Identi f cation implies that E [ x i 1 y i 1 ] . . . E [ x iM y iM ] E [ x i 1 z 0 i 1 ] . . . E [ x iM z 0 iM ] δ 0 = 0 or σ xy Σ xz δ 0 = 0 For δ 0 to be the unique solution requires the rank condition rank( Σ xz )= L
Rema rk :Inthemode lw ithcommoncoe cients, some equations may be indi- vidually unidenti f ed but identi f ed withing the system.

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
The moments g i ( δ ) are a MDS with E [ g i ( δ 0 ) g i ( δ 0 ) 0 ]= S = S 11 ··· S 1 M . . . . . . . . . S M 1 S MM E cient GMM estimation min δ J ( δ , ˆ S 1 )= n g n ( δ ) 0 ˆ S 1 g n ( δ ) g n ( δ S xy S xz δ S xy = S x 1 y 1 . . . S x M y M , S xz = S x 1 z 1 . . . S x M z M Straightforward algebra gives δ ( ˆ S 1 )=( S 0 xz ˆ S 1 S xz ) 1 S 0 xz ˆ S 1 S xy
Special cases 1. 3SLS with common coe cients conditional homoskedasticity x i 1 = x i 2 = ··· = x iM = x i S = S 3 SLS = Σ E [ x i x 0 i ] 2. SUR with common coe cients (Random e f ects) conditional homoskedasticity x i 1 = x i 2 = = x iM = x i x i = union ( z i 1 ,..., z iM ) S = S SUR = Σ E [ z i z 0 i ]

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Random E f ects Estimator The SUR model with common coe cients is the giant regression y Mn × 1 = Z δ + ε E [ εε 0 ]= Σ I n Σ = σ 11 ··· σ 1 M . . . . . . . . . σ M 1 σ MM The FGLS estimator is the random e f ects (RE) estimator and has the form ˆ δ RE =( Z 0 ( ˆ Σ 1 I n ) Z ) 1 Z 0 ( ˆ Σ 1 I n ) y
The elements of ˆ Σ are usually estimated using the pooled OLS estimator of the giant regression ˆ δ OLS =( Z 0 Z ) 1 Z 0 y and forming ˆ σ mh y m Z m ˆ δ OLS ) 0 ( y h Z h ˆ δ OLS ) /n

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Simplifying the RE estimator It turns out that the RE estimator may also be derived from an alternative representation of the giant regression.
This is the end of the preview. Sign up to access the rest of the document.

{[ snackBarMessage ]}

### Page1 / 53

panelslides - Multiple Equation GMM with Common Coecients...

This preview shows document pages 1 - 13. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online