You are on page 1of 7

ECON 6306 Applied Econometrics

Homework 11

1. Use brumm.dta
2. Estimate inflat = f(money output)
Contains data from e:/econ4355/HillGriffithsLim/datafiles/brumm.dta
.d
obs:
76
vars:
7
size:
4,256
------------------------------------------------------------------------------------------------storage display
value
variable name
type
format
label
variable label
------------------------------------------------------------------------------------------------money
double %10.0g
growth rate of money supply = change ln(m)
inflat
double %10.0g
growth rate of prices = change ln(p)
output
double %10.0g
growth rate of output = change ln(q)
initial
double %10.0g
initial level of GDP per capita
poprate
double %10.0g
average population growth rate
inv
double %10.0g
average investment share of GDP
school
double %10.0g
a measure of the population educational attainment
------------------------------------------------------------------------------------------------Sorted by:
. su
Variable |
Obs
Mean
Std. Dev.
Min
Max
-------------+-------------------------------------------------------money |
76
29.58947
56.16827
2.5
356.7
inflat |
76
25.35395
58.94767
-.6
374.3
output |
76
2.997368
2.02695
-.6
9.6
initial |
76
1.703421
1.765252
.237
7.38
poprate |
76
2.011842
1.021759
.1
3.7
inv |
76
.2239513
.0666283
.0868
.421
school |
76
.8259211
.319229
.14
1.64
. regress inflat money output
Source |
SS
df
MS
-------------+-----------------------------Model | 259256.055
2 129628.027
Residual | 1356.03409
73 18.5758094
-------------+-----------------------------Total | 260612.089
75 3474.82785

Number of obs
F( 2,
73)
Prob > F
R-squared
Adj R-squared
Root MSE

=
76
= 6978.32
= 0.0000
= 0.9948
= 0.9947
=
4.31

-----------------------------------------------------------------------------inflat |
Coef.
Std. Err.
t
P>|t|
[95% Conf. Interval]
-------------+---------------------------------------------------------------money |
1.033131
.0090422
114.26
0.000
1.01511
1.051152
output | -1.662006
.2505661
-6.63
0.000
-2.161383
-1.162628
_cons | -.2342135
.9799249
-0.24
0.812
-2.187201
1.718774
------------------------------------------------------------------------------

a. Test the strong hypothesis that b1=0, b2=1 and b3=-1 jointly. (use
STATA command test)
. test _b[_cons]=_b[money]-1=_b[output]+1=0
( 1)
( 2)
( 3)

- money + _cons = -1
- output + _cons = 1
_cons = 0

ECON 6306 Applied Econometrics


Homework 11
F(

3,
73) =
Prob > F =

10.52
0.0000

Reject this hypothsis.


b. Test the weaker hypothesis that b2=1 and b3=-1.
. test _b[money]-1=_b[output]+1=0
( 1)
( 2)

money - output = 2
money = 1
F(

2,
73) =
Prob > F =

12.64
0.0000

Reject this hypothesis.


3. Examine the least squares residuals for the presence of heteroscedasticity
with respect to the variable money (use Breusch-Pagan). Note the STATA
command estat hettest requires the iid option to use the LaGrange multiplier
test (N*R-squared).
. * test for heteroskedasticity
. gen err2 = err*err
. regress err2 money
Source |
SS
df
MS
-------------+-----------------------------Model | 90381.0816
1 90381.0816
Residual | 294683.561
74 3982.21028
-------------+-----------------------------Total | 385064.642
75 5134.19523

Number of obs
F( 1,
74)
Prob > F
R-squared
Adj R-squared
Root MSE

=
=
=
=
=
=

76
22.70
0.0000
0.2347
0.2244
63.105

-----------------------------------------------------------------------------err2 |
Coef.
Std. Err.
t
P>|t|
[95% Conf. Interval]
-------------+---------------------------------------------------------------money |
.6180408
.12973
4.76
0.000
.3595481
.8765335
_cons | -.4449475
8.193454
-0.05
0.957
-16.77076
15.88087
-----------------------------------------------------------------------------. scalar lm = e(N)*e(r2)
. di lm
17.838465
. scalar plm = chi2tail(1,lm)
. di plm
.00002405

Clear evidence of heteroskedasticity

ECON 6306 Applied Econometrics


Homework 11

4. Obtain robust standard errors and compare them to the standard errors from
2.

inflat |
Coef.
Std. Err.
t
P>|t|
[95% Conf. Interval]
-------------+---------------------------------------------------------------money |
1.033131
.0090422
114.26
0.000
1.01511
1.051152
output | -1.662006
.2505661
-6.63
0.000
-2.161383
-1.162628
_cons | -.2342135
.9799249
-0.24
0.812
-2.187201
1.718774
|
Robust
inflat |
Coef.
Std. Err.
t
P>|t|
[95% Conf. Interval]
-------------+---------------------------------------------------------------money |
1.033131
.0236942
43.60
0.000
.9859086
1.080354
output | -1.662006
.1759137
-9.45
0.000
-2.012601
-1.31141
_cons | -.2342135
.619615
-0.38
0.707
-1.469104
1.000677

The robust standard errors are larger for money. The robust standard errors are smaller for the
intercept and for output (this is unusual).
5. The variable output may be endogenous. Use the four instruments initial,
school, inv and poprate in an instrument variable regression with robust
standard errors.
. ivregress 2sls inflat money (output = initial school inv poprate),robust small

Instrumental variables (2SLS) regression

Number of obs =
F(

2,

76

73) = 1001.46

Prob > F

0.0000

R-squared

0.9947

Adj R-squared =

0.9946

Root MSE

4.3436

-----------------------------------------------------------------------------|
inflat |

Robust
Coef.

Std. Err.

P>|t|

[95% Conf. Interval]

-------------+---------------------------------------------------------------output |

-1.3942

.3341747

-4.17

0.000

-2.060209

-.7281911

money |

1.035059

.0247109

41.89

0.000

.9858106

1.084308

_cons |

-1.093985

1.21889

-0.90

0.372

-3.523229

1.33526

------------------------------------------------------------------------------

ECON 6306 Applied Econometrics


Homework 11
Instrumented:

output

Instruments:

money initial school inv poprate

6. Test the strong and weak hypotheses from 2.


. test _b[_cons]=_b[money]-1=_b[output]+1=0
( 1)
( 2)
( 3)

- money + _cons = -1
- output + _cons = 1
_cons = 0
F(

3,
73) =
Prob > F =

9.75
0.0000

. test _b[money]-1=_b[output]+1=0
( 1)
( 2)

- output + money = 2
money = 1
F(

2,
73) =
Prob > F =

2.30
0.1072

The strong hypothesis is still rejected, but, using robust standard errors, the weak hypothesis is no
long rejected.
7. Use the Hausman test to check the endogeneity of output. Use robust
standard errors when estimating the necessary regressions. First regress
output on money initial school inv and poprate. Then run the regression from
2. adding these residuals as an additional variable. Test the statistical
significance of the coefficient of the residuals. The null hypothesis is that
output is endogenous. Do we reject the endogeneity of output?

ECON 6306 Applied Econometrics


Homework 11

. regress output money initial school inv poprate,robust


Linear regression

Number of obs =
F( 5,
70) =
Prob > F
=
R-squared
=
Root MSE
=

76
3.53
0.0066
0.2411
1.8277

-----------------------------------------------------------------------------|
Robust
output |
Coef.
Std. Err.
t
P>|t|
[95% Conf. Interval]
-------------+---------------------------------------------------------------money | -.0047965
.0029347
-1.63
0.107
-.0106496
.0010566
initial | -.2709296
.1418189
-1.91
0.060
-.5537785
.0119193
school | -1.160768
1.001055
-1.16
0.250
-3.157309
.8357734
inv |
12.90258
4.560579
2.83
0.006
3.80679
21.99836
poprate | -.4389086
.2683989
-1.64
0.106
-.9742133
.096396
_cons |
2.552969
1.317966
1.94
0.057
-.0756307
5.181569
-----------------------------------------------------------------------------. predict hr,residuals
. regress inflat money output hr,robust
Linear regression

Number of obs =
F( 3,
72) =
Prob > F
=
R-squared
=
Root MSE
=

76
753.52
0.0000
0.9948
4.3308

-----------------------------------------------------------------------------|
Robust
inflat |
Coef.
Std. Err.
t
P>|t|
[95% Conf. Interval]
-------------+---------------------------------------------------------------money |
1.035059
.0247066
41.89
0.000
.9858078
1.084311
output |
-1.3942
.328338
-4.25
0.000
-2.04873
-.7396704
hr |
-.338844
.3562418
-0.95
0.345
-1.048999
.371311
_cons | -1.093985
1.205469
-0.91
0.367
-3.497044
1.309075
------------------------------------------------------------------------------

We are testing the coefficient of the residuals (hr) for significance. If the residual coefficient is
statistically significant we would reject the null hypothesis that the residuals and output are
uncorrelated, that is, cov(output,e)=0.
The coefficient is not significant at the 5% level, so we cannot reject the hypothesis that output and
e are correlated, so we have no evidence that output is endogenous.

ECON 6306 Applied Econometrics


Homework 11

8. Test the validity of the overidentifying restrictions. First run an ivregress of


inflat on money and output instrumented by the four instruments. Predict the
residuals. Run a regression of these residuals on money, initial school inv and
poprate. Then calculate the LR statistic and test it for significance using the
number of instruments minus one (the number of overidentifying
instruments) as the degrees of freedom. The null hypothesis is that all the
surplus instruments are valid.
. ivregress 2sls inflat money (output = initial school inv poprate),robust small
Instrumental variables (2SLS) regression

Number of obs
F( 2,
73)
Prob > F
R-squared
Adj R-squared
Root MSE

=
76
= 1001.46
= 0.0000
= 0.9947
= 0.9946
= 4.3436

-----------------------------------------------------------------------------|
Robust
inflat |
Coef.
Std. Err.
t
P>|t|
[95% Conf. Interval]
-------------+---------------------------------------------------------------output |
-1.3942
.3341747
-4.17
0.000
-2.060209
-.7281911
money |
1.035059
.0247109
41.89
0.000
.9858106
1.084308
_cons | -1.093985
1.21889
-0.90
0.372
-3.523229
1.33526
-----------------------------------------------------------------------------Instrumented: output
Instruments:
money initial school inv poprate
. predict ivr,residuals
. regress ivr money initial school inv poprate,robust
Linear regression

Number of obs =
F( 5,
70) =
Prob > F
=
R-squared
=
Root MSE
=

76
1.59
0.1731
0.0323
4.3634

-----------------------------------------------------------------------------|
Robust
ivr |
Coef.
Std. Err.
t
P>|t|
[95% Conf. Interval]
-------------+---------------------------------------------------------------money |
.0016071
.0246392
0.07
0.948
-.0475341
.0507484
initial |
.5087449
.2191058
2.32
0.023
.0717521
.9457377
school | -1.235133
1.715624
-0.72
0.474
-4.656838
2.186571
inv |
7.288907
5.277609
1.38
0.172
-3.236953
17.81477
poprate |
.5622483
.5112346
1.10
0.275
-.4573771
1.581874
_cons | -2.657553
2.775906
-0.96
0.342
-8.193924
2.878817
-----------------------------------------------------------------------------. scalar ivlr = e(N)*e(r2)
. di ivlr
2.4552178
. scalar pivlr = chi2tail(3,ivlr)
. di pivlr
.48343861

ECON 6306 Applied Econometrics


Homework 11
We do not reject the null hypothesis that all the surplus instruments are valid.
9. Test the relevance of the instruments using a joint F-test. Regress output on
money and the instruments. Test whether the coefficients for the instruments
are jointly equal to zero.

. regress output money initial school inv poprate


Source |
SS
df
MS
-------------+-----------------------------Model |
74.297195
5
14.859439
Residual | 233.842279
70 3.34060398
-------------+-----------------------------Total | 308.139474
75 4.10852632

Number of obs
F( 5,
70)
Prob > F
R-squared
Adj R-squared
Root MSE

=
=
=
=
=
=

76
4.45
0.0014
0.2411
0.1869
1.8277

-----------------------------------------------------------------------------output |
Coef.
Std. Err.
t
P>|t|
[95% Conf. Interval]
-------------+---------------------------------------------------------------money | -.0047965
.0038462
-1.25
0.217
-.0124675
.0028746
initial | -.2709296
.1567111
-1.73
0.088
-.5834801
.0416208
school | -1.160768
.8839082
-1.31
0.193
-2.923667
.6021316
inv |
12.90258
3.602194
3.58
0.001
5.718228
20.08693
poprate | -.4389086
.2797947
-1.57
0.121
-.9969416
.1191243
_cons |
2.552969
1.231741
2.07
0.042
.0963384
5.0096
-----------------------------------------------------------------------------. test
( 1)
( 2)
( 3)
( 4)

_b[initial]=_b[school]=_b[inv]=_b[poprate]=0
initial - school = 0
initial - inv = 0
initial - poprate = 0
initial = 0
F(

4,
70) =
Prob > F =

4.64
0.0022

We reject the null hypothesis that the coefficients are zero.


However, we do not pass the hurdle for strong instruments, which is that the F-statistic is greater
than 10.

You might also like