内生性问题：处理方法与进展 - Stata

1 Stata .. 2017-09-17.. ( ). ( ). ( ).. IV-GMM. (Panel Data).. Heckman Treatment effect . (DID) (PSM). (RDD). (SCM). (SEM).. ! .. ( ability fat). ( ). - ( ). (self-selection). (self-selection). : .. y = 0 + 1 x1 + 2 x2 + + k xk + . : Pose ! (y , x1 , x1 , ,xk ). rank (X ' X ) = k Cov ( X , ) 0=. = or E[ x1 , x1 , , xk ] 0.. OLS (MLE) ( ). ( ).. Tobin's Q. \ .. Fazzari et al. (1988, JEL): - . Investit = i + 1 Qit + 2 CashFlowit + it Refs Fazzari et al. (1988) |JEL| Kaplan and Zingales (1997) |QJE| . Fazzari et al. (2000) |QJE| Kaplan and Zingales (2000) |QJE| . Erickson and Whited (2000) |JPE| Alti (2003) |JF|.

2 \ . Omitted Variable or missing value bias: . True : + 1 x1 + 2 x2 + u1. y=.. + 1 x1. Estimate : y = + u2. if Corr ( x2 , x1 ) 0, then Corr (u2 , x1 ) 0 Endog! . | |. | | | |.. ( ). IV or GMM ( ). : . Omitted Variable bias: .. + 1 Controlsi + 2 Educationi + i Incomei =. Q1: . A1: Maybe yes, maybe not. C1: ( - - ). C2: (0/1). Q2: . A2: IV ; ; ;. A2: RDD, 600 599 601 .. A3: PSM, .. Measurement Error (ME): . + x* + u True model : y =. x* + , E[ ] =. x= 0. Empirical model : y = + x + . + x* + u y=. = + ( x ) + u = + x + ( u ). Cov ( x, ) 0 ? = + x + . Stata commands: eivreg | sem | logitem | simex | cme | Ewreg | XTEWreg.

3 Measurement Error (ME): . - . Fazzari et al. (1988) |JEL| Kaplan and Zingales (1997) |QJE| . Fazzari et al. (2000) |QJE| Kaplan and Zingales (2000) |QJE| . Erickson and Whited (2000) |JPE| Alti (2003) |JF|, Erickson and Whited (2012) |RFS|. I it CFit . = 0 + Q. 1 it + 2 + it Kit 1 Kit 1 . T. Whited : Higher Order Moments GMM (HGMM) | Signs Estimator (SigE). Erickson and Whited(2012) |RFS| Average q Marginal q HGMM, Dynamic Panel Data, IV. Minimum Distance Technique ( Stata codes). Stata commands: | Ewreg | XTEWreg |.. GMM (IV-GMM). (Panel Data Models). Heckman Treatment effect.

4 (DID). (PSM). (RDD). (SCM). SEM . : .. ( ). ( ).. ( 11 ). ( ). ( ). IV-GMM . y = a + X + . Z. IV Corr(Z, ) = 0 . 2 SLS Corr(Z, ) = 0 . Z1, Z2 X . GMM. E[Z1' ] = 0, E[Z2' ] = 0, . Stata commands: ivregress | ivreg2 | gmm : GMM. w1. w2. OLS w1. w2w3. x w3. IV. 2 SLS. IV-2 SLS . IV. 2 SLS. Stage1: reg X on Z, get X_hat Stata2: reg Y on X_hat, get This is wrong! (SE is biased). : ivregress 2sls y x1 x2 (x3 x4 = z1 z2 z3). GMM . Moment Condition (MC, ). Lars Peter Hansen (SMC).. Fixed Effects Model (FE).. FE : yit = 0 + i + X it' + it 0 + X it' + uit POLS : yit =.. uit= i + it i : , CEO.

5 FE . Stata commands: xtreg, fe | xi: regress | areg : 20. 15. 10. 5. 0. -10 -5 0 5 10. x 30. 25. 20. 15. 10. 5. 0 5 10 15 20 25. x 4. 2. 0. -2. -4. -10 -5 0 5 10. x . Fixed Effects Model (FE).. FE : yit = 0 + i + X it' + it FE . : . yit = X it' + it ( yit yit ) = ( X it' X it' ) + ( it it ). 1.. Ti where, yit = t =1. yit Ti . Fixed Effects Model (FE).. Flannery and Rangan (2006) |JFE| . Lemmon et al. (2008) |JF| . Malmendier et al.(2011) |JF| ( ) . Graham et al.(2012) |RFS| . (2012) | | . Petersen(2009) |RFS| . Cameron and Miller (2015) |JHR| . : . Dynamic Panel Data Models.

6 I + yit 1 + X it' + it yit = (1) || . i yit 2 + X i't 1 + it 1 (2) || . yit 1 =+. yit yit 1 + X it' + it = (3) || , . =. yit 1 yit 1 yit 2 || OLS, FE GMM.. =. it it it 1 || IVs for yit 1: yit 2 yit 3 yit 4 yit 2. Corr ( yit 1 , it ) 0 || OLS, FE GMM. Stata commands: xtabond | xtdpdsys | xtdpd | xtlsdvc | xtregdhp | xtabond2.. Dynamic Panel Data Models . Aghion et al.(2009) |JM| ( ). Brown et al.(2009) |JF| ( ). Wintoki et al.(2012) |JFE| . ( ). Flannery and Hankins(2013) |JCF| : . Seo and Shin (2017) |JoE| . : . Dynamic Panel Data Models . (long-difference, LD). Hahn et al.

7 (2007) |JE| T y . Huang and Ritter(2009) |JFQA| . Han-Phillips dynamic panel data model Han and Phillips(2010) |ET| Linear Dynamic Panel Data Regression y Panel Unit Root Test (Quantile Dynamic Panel Data). Galvao(2011) |ET| Quantile regression for dynamic panel data VAR (Panel VAR models). Holtz-Eakin et al.(1988) |E~trica| Arellano and Bond(1991) |RES| . Love and Zicchino(2006) |QREF| Canova and Ciccarelli (2013, Survey). Abrigo and Love (2016, Stata Journal). Stata commands: xtregdhp | gmm | pvar | pvar2 | xtvar . Spatial Dynamic Panel Data Models Lee, , J. Yu, 2010, A spatial dynamic panel data model with both time and individual fixed effects, Econometric Theory, 26 (02), pp.

8 564-597. Yu, J., R. de Jong, Lee, 2012, Estimation for spatial dynamic panel data with fixed effects: The case of spatial cointegration, Journal of Econometrics, 167 (1), pp. 16-37. Lee, , J. Yu, 2010, Some recent developments in spatial panel data models, Regional Science and Urban Economics, 40 (5), pp. 255-271. ( ). Yu, J., Lee, 2012, Convergence: A spatial dynamic panel data approach, Global Journal of Economics, 1 (1), pp. forthcoming. ( ). Lee, , J. Yu, 2011, Estimation of spatial panels, Now Publishers Inc. (Book).. Difference-In-Difference (DID). ( ) . 2009 2011 Difference ( ) 16,000 20,000 4,000.

9 ( ) 12,000 17,000 5,000 . Difference -1,000.. 50.. 45. 40. : (ADA). 1992 7 .. 35. 30 . 25. 20. 15 . 10. 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996.. Source: Acemoglu and Angrist (2001, JPE) . Difference-In-Difference (DID).. yit = 0 + 1 Treati + 2 Postit + Treati Postit + 3 X it + it . yit = 0 + 1 Treati + 2 Yeart + Treati Yeart + 3 X it + 4 X it Yeart + it Stata commands ( DID ). global controls z1 z2 z3 z4 . reg y Treat # ##($controls). Stata commands: diff | ddid | regress . Difference-In-Difference (DID).. PSM + DID. (Treat) help ddid : . Difference-In-Difference (DID).

10 Cooper et al. (2005) |JF| . Villalonga (2004) |FM| DID Heckman Chhaochharia and Grinstein (2009) |JF| CEO . Fr sard (2010) |JF| . Black and Kim (2012) |JF|, , DID, 2 SLS, 3 SLS. Tsoutsoura (2015) |JF|, .. Propensity Score Matching Method (PSM).. Yi = + Di + X i' + i =. E[Yi | Di = 1, X i =. x ] E[Yi | Di = 0, X i =. x]. {observable} {unobservable}. ( ). PSM Logit PS . Stata commands: teffects psmatch | kmatch | psmatch2 | optmatch2. | ccmatch | cem . Propensity Score Matching Method (PSM).. : Propensity Score (PS ). Logit(Size, Industry, ROA, Leverage, Ownership, .) PS.