IV estimation and interpretation of the set of results

For econometric discussions not necessarily related to EViews.

Moderators: EViews Gareth, EViews Moderator

oksanakim
Posts: 39
Joined: Wed Apr 04, 2012 10:38 am

IV estimation and interpretation of the set of results

Postby oksanakim » Fri Nov 17, 2017 11:22 am

Dear Moderators
May I ask for assistance on the set of results related to an IV estimation. I am running a regression whereby a firm value LNMB is regressed on a number of variables, of which foreigners, dexpats, is believed to be an endogenous variables because of a two-directional causality between LNMB and dexpats. I estimated my model using an OLS regression and found a significantly positive coefficient on dexpats, as predicted. Next, I found an instrument - a dummy variable equal to one if a company is headquartered in a large city (dlargecity). This variable is not correlated with LNMB but is correlated with dexpats, according to the correlation matrix. Hence, I use it as an IV via a TSLS estimation. Upon estimation, the coefficient on the dexpats is no longer significant. What does this mean? That the result of the OLS estimation does not hold (bad news), or that endogeneity of dexpats is not an issue and OLS results can be considered valid (good news)? Please see the output below on Dexpats ( I skipped industries and years):


Dependent Variable: LNMB
Method: Two-Stage Least Squares
Date: 11/17/17 Time: 12:13
Sample (adjusted): 1 2116
Included observations: 1725 after adjustments
White heteroskedasticity-consistent standard errors & covariance
Instrument specification: C DLARGECITY BOARDSIZE BOARDAGE
DCEOWOMAN DFEMALEDIR INDEPDIR CHANGEDIRECTOR
DIRECTORSHARE CEOAGE CEODUALITY DPOLITCON DBIG4 SIZE
LEV CHSALES RETURNS DIFRS DLISTEDOVERSEAS DLONDON
STATEOWN DSTRATEGICIND @EXPAND(YEAR, @DROPFIRST)
@EXPAND(INDUSTRY, @DROPFIRST)

Variable Coefficient Std. Error t-Statistic Prob.

C -3.169766 3.236134 -0.979492 0.3275
DEXPATS -2.248857 1.646050 -1.366214 0.1721
BOARDSIZE 0.002805 0.018745 0.149620 0.8811
BOARDAGE 0.010234 0.017608 0.581232 0.5612
DCEOWOMAN -0.246019 0.228716 -1.075656 0.2822
DFEMALEDIR -0.154095 0.106534 -1.446443 0.1482
INDEPDIR 0.034364 0.015673 2.192635 0.0285
CHANGEDIRECTOR -0.006046 0.017198 -0.351577 0.7252
DIRECTORSHARE -0.007146 0.002660 -2.686083 0.0073
CEOAGE -0.006914 0.003949 -1.750692 0.0802
CEODUALITY 0.133780 0.181356 0.737663 0.4608
DPOLITCON 0.029842 0.092729 0.321814 0.7476
DBIG4 0.711978 0.367297 1.938425 0.0527
SIZE -0.053262 0.075066 -0.709532 0.4781
LEV 2.220114 0.289600 7.666147 0.0000
CHSALES -3.88E-06 3.03E-06 -1.279954 0.2007
RETURNS 0.000631 0.000560 1.126737 0.2600
DIFRS 0.400530 0.153046 2.617049 0.0089
DLISTEDOVERSEAS -0.325734 0.314060 -1.037171 0.2998
DLONDON 1.094213 0.511984 2.137203 0.0327
STATEOWN -0.006651 0.003603 -1.845682 0.0651
DSTRATEGICIND 0.192295 0.256007 0.751133 0.4527

R-squared -0.252955 Mean dependent var -0.007941
Adjusted R-squared -0.289608 S.D. dependent var 1.125641
S.E. of regression 1.278288 Sum squared resid 2736.985
F-statistic 10.30680 Durbin-Watson stat 0.547960
Prob(F-statistic) 0.000000 Second-Stage SSR 1359.189
J-statistic 0.000000 Instrument rank 50


Next, the negative R-squared in the above estimation bother me somewhat. Further, the regressor endogeneity test rejects the null that dexpat is exogneous at 5 percent, if I correctly understand the below output. If my understanding of the below test is correct, then this does not correspond to the fact that the coefficient on dexpats in the above estimation is insignificant meaning that endogeneity is not an issue. I am very confused by the overall interpretation of the below and the above outputs.

Endogeneity Test
Null hypothesis: DEXPATS are exogenous
Equation: DEXPATSTSLS
Specification: LNMB C DEXPATS BOARDSIZE BOARDAGE DCEOWOMAN
DFEMALEDIR INDEPDIR CHANGEDIRECTOR DIRECTORSHARE
CEOAGE CEODUALITY DPOLITCON DBIG4 SIZE LEV CHSALES
RETURNS DIFRS DLISTEDOVERSEAS DLONDON STATEOWN
DSTRATEGICIND @EXPAND(YEAR, @DROPFIRST)
@EXPAND(INDUSTRY, @DROPFIRST)
Instrument specification: C DLARGECITY BOARDSIZE BOARDAGE
DCEOWOMAN DFEMALEDIR INDEPDIR CHANGEDIRECTOR
DIRECTORSHARE CEOAGE CEODUALITY DPOLITCON DBIG4 SIZE
LEV CHSALES RETURNS DIFRS DLISTEDOVERSEAS DLONDON
STATEOWN DSTRATEGICIND YEAR=2 YEAR=3 YEAR=4 YEAR=5
YEAR=6 YEAR=7 YEAR=8 YEAR=9 YEAR=10 YEAR=11 YEAR=12
YEAR=13 YEAR=14 YEAR=15 YEAR=16 YEAR=17 YEAR=18
INDUSTRY=2 INDUSTRY=3 INDUSTRY=4 INDUSTRY=5
INDUSTRY=6 INDUSTRY=7 INDUSTRY=8 INDUSTRY=9
INDUSTRY=10 INDUSTRY=11 INDUSTRY=12
Endogenous variables to treat as exogenous: DEXPATS

Value df Probability
Difference in J-stats 5.846789 1 0.0156


And finally, I do not understand how to interpret the Instrument weakness test for the dlargecity that I used as an IV: what does the Cragg-Donlad stats means in relation to the Stock-Yogo critical values? I did check the Eviews manual and the previous posts but am still uncertain I understand this part:

Weak Instrument Diagnostics
Equation: DEXPATSTSLS

Cragg-Donald F-stat: 5.723933

Stock-Yogo bias critical values not available for
models with less than 3 instruments.

Stock-Yogo critical values (size):
10% 16.38
15% 8.96
20% 6.66
25% 5.53

Moment selection criteria:

SIC-based: 0.000000
HQIC-based: 0.000000
Relevant MSC: 94.04930


I will be immensely grateful for your help with the above!! :?

Return to “Econometric Discussions”

Who is online

Users browsing this forum: No registered users and 6 guests