Financial Econometrics

Tutorial Exercise 4 Solutions

(a) Engle-Granger method

. regress lrdiv lrprice

Source | SS df MS Number of obs = 34

-------------+------------------------------ F( 1, 32) = 120.90
Model | 2.10335804 1 2.10335804 Prob > F = 0.0000
Residual | .556735447 32 .017397983 R-squared = 0.7907
-------------+------------------------------ Adj R-squared = 0.7842
Total | 2.66009349 33 .080608894 Root MSE = .1319

------------------------------------------------------------------------------
lrdiv | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lrprice | .4324629 .0393316 11.00 0.000 .352347 .5125787
_cons | -4.045398 .0538516 -75.12 0.000 -4.15509 -3.935705
------------------------------------------------------------------------------

The cointegrating regression is:

ln(divt/rpit) = –4.0454 + 0.4325 ln(pricet/rpit) + et

. predict e1, resid
. dfuller e1, lags(1) noconstant regress

Augmented Dickey-Fuller test for unit root Number of obs = 32

---------- Interpolated Dickey-Fuller ---------

Test 1% Critical 5% Critical 10% Critical
Statistic Value Value Value
------------------------------------------------------------------------------
Z(t) -2.135 -2.649 -1.950 -1.603

------------------------------------------------------------------------------
D.e1 | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
e1 |
L1 | -.4343184 .2034732 -2.13 0.041 -.8498661 -.0187706
LD | -.2898719 .1750436 -1.66 0.108 -.6473587 .067615
------------------------------------------------------------------------------

The test of H0:=0 against H1:<0 is as follows:

ρ^ −0 −0. 4343
τ=
se( ^ρ ) = 0 .2899 = –2.135 (from the Stata output)

Decision rule: Accept H0 if   EG0.05  et is non-stationary

Reject H0 if  < EG0.05  et is stationary

The 5% critical value is EG0.05 = –3.461 (EG0.05 = –3.461 is the closest available value in the
Engle-Granger table, for T=50 and k–1=1.

The decision is to accept H0  et is non-stationary

The Engle-Granger test suggests the log-dividend and log-price series are not cointegrated.
Despite this result, we will estimate the error correction model:
. regress D1.lrdiv D1.lrprice L1.e1, noconstant

Source | SS df MS Number of obs = 33

-------------+------------------------------ F( 2, 31) = 14.43
Model | .092314502 2 .046157251 Prob > F = 0.0000
Residual | .099176931 31 .003199256 R-squared = 0.4821
-------------+------------------------------ Adj R-squared = 0.4487
Total | .191491434 33 .005802771 Root MSE = .05656

------------------------------------------------------------------------------
D.lrdiv | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lrprice |
D1 | .0862645 .0333204 2.59 0.015 .0183072 .1542218
e1 |
L1 | -.421615 .081102 -5.20 0.000 -.5870235 -.2562065
------------------------------------------------------------------------------

The estimated error correction model is as follows (standard errors of estimated coefficients in
parentheses):

yt = 0.0863xt – 0.4216

v^ t−1 +  where v^ t−1 = y + 4.0454 – 0.4325x
t t–1 t–1
(0.0333) (0.0811)

(b) Johansen method

. var lrdiv lrprice, lags(1 2 3)

. var lrdiv lrprice if year>=1973, lags(1 2)
. var lrdiv lrprice if year>=1973, lags(1)

From the Stata output (not shown here) the values of MAIC are as follows:

Lag-length, p MAIC

1 -2.395
2 -2.334
3 -2.122

The cointegration tests will be implemented below using a VECM derived from a VAR(1)
model.
. vecrank lrdiv lrprice, lags(1) max

Johansen tests for cointegration

Trend: constant Number of obs = 33
Sample: 1971 2003 Lags = 1
-------------------------------------------------------------------------------
5%
maximum trace critical
rank parms LL eigenvalue statistic value
0 2 31.194642 . 32.6568 15.41
1 5 47.392763 0.62533 0.2605* 3.76
2 6 47.523028 0.00786
-------------------------------------------------------------------------------
5%
maximum max critical
rank parms LL eigenvalue statistic value
0 2 31.194642 . 32.3962 14.07
1 5 47.392763 0.62533 0.2605 3.76
2 6 47.523028 0.00786
-------------------------------------------------------------------------------

The cointegration test results are as follows:

trace max

Stage 1 H0:r=0 against H1:r>0 H0:r=0 against H1:r=1

trace = 32.6568 max = 32.3962
C.V. = 15.41 C.V. = 14.07
Decision is reject H0 Decision is reject H0

Stage 2 H0:r1 against H1:r=2 H0:r1 against H1:r=2

trace = 0.2605 max = 0.2605
C.V. = 3.76 C.V. = 3.76
Decision is accept H0 Decision is accept H0

Both tests reject H0:r=0 and both tests fail to reject H0:r1. We infer that the log-dividend and
log-price series are cointegrated (r=1, so there is one cointegrating vector). In this case, the
Engle-Granger and Johansen cointegration tests give contradictory results.
. vec lrdiv lrprice, lags(1) rank(1)

Vector error-correction model

Sample: 1971 2003 No. of obs = 33

AIC = -2.569258
Log likelihood = 47.39276 HQIC = -2.492966
Det(Sigma_ml) = .0001939 SBIC = -2.342515

Equation Parms RMSE R-sq chi2 P>chi2

----------------------------------------------------------------
D_lrdiv 2 .058999 0.4365 24.01263 0.0000
D_lrprice 2 .297413 0.1059 3.671567 0.1595
----------------------------------------------------------------

------------------------------------------------------------------------------
| Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
D_lrdiv |
_ce1 |
L1. | -.344835 .0710633 -4.85 0.000 -.4841166 -.2055535
_cons | .0122468 .0103269 1.19 0.236 -.0079936 .0324872
-------------+----------------------------------------------------------------
D_lrprice |
_ce1 |
L1. | .676833 .3582296 1.89 0.059 -.0252841 1.37895
_cons | .0062395 .052058 0.12 0.905 -.0957922 .1082712
------------------------------------------------------------------------------

Cointegrating equations

Equation Parms chi2 P>chi2

-------------------------------------------
_ce1 1 244.2177 0.0000
-------------------------------------------

Identification: beta is exactly identified

Johansen normalization restriction imposed
------------------------------------------------------------------------------
beta | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
_ce1 |
lrdiv | 1 . . . . .
lrprice | -.5475658 .0350387 -15.63 0.000 -.6162404 -.4788913
_cons | 3.921397 . . . . .
------------------------------------------------------------------------------

The VECM based on the VAR(1) model is as follows:

yt = 0.0122 – 0.3448 (yt–1 + 3.9214 – 0.5476xt–1) + e1t

(0.0103) (0.0711)

xt = 0.0062 + 0.6768 (yt–1 + 3.9214 – 0.5476xt–1) + e2t

(0.0521) (0.3582)

The results of the cointegration tests based on the VECM derived from the VAR(2) model are
the same as before (although the precise numbers are different), and the Stata output is not
reproduced here. The estimated VECM is as follows:
. vec lrdiv lrprice, lags(2) rank(1)

Vector error-correction model

Sample: 1972 2003 No. of obs = 32

AIC = -2.400362
Log likelihood = 47.40579 HQIC = -2.263717
Det(Sigma_ml) = .0001771 SBIC = -1.988124

Equation Parms RMSE R-sq chi2 P>chi2

----------------------------------------------------------------
D_lrdiv 4 .059291 0.4600 23.85453 0.0001
D_lrprice 4 .293856 0.2088 7.388676 0.1167
----------------------------------------------------------------

------------------------------------------------------------------------------
| Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
D_lrdiv |
_ce1 |
L1. | -.4062898 .1193737 -3.40 0.001 -.6402578 -.1723217
lrdiv |
LD. | .1305146 .1459754 0.89 0.371 -.155592 .4166212
lrprice |
LD. | -.0570437 .0523316 -1.09 0.276 -.1596118 .0455244
_cons | .0106322 .0106027 1.00 0.316 -.0101487 .0314131
-------------+----------------------------------------------------------------
D_lrprice |
_ce1 |
L1. | .2875093 .5916379 0.49 0.627 -.8720798 1.447098
lrdiv |
LD. | .8772316 .7234813 1.21 0.225 -.5407658 2.295229
lrprice |
LD. | -.3458551 .2593652 -1.33 0.182 -.8542016 .1624914
_cons | .0150248 .052549 0.29 0.775 -.0879695 .118019
------------------------------------------------------------------------------

Cointegrating equations

Equation Parms chi2 P>chi2

-------------------------------------------
_ce1 1 181.9504 0.0000
-------------------------------------------

Identification: beta is exactly identified

Johansen normalization restriction imposed
------------------------------------------------------------------------------
beta | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
_ce1 |
lrdiv | 1 . . . . .
lrprice | -.5440621 .0403341 -13.49 0.000 -.6231154 -.4650088
_cons | 3.917029 . . . . .
------------------------------------------------------------------------------

The VECM based on the VAR(2) model is as follows:

yt = 0.0106 + 0.1305yt–1 – 0.0570xt–1 – 0.4063 (yt–1 + 3.9170 – 0.5441xt–1) + e1t

(0.0106) (0.1460) (0.0523) (0.1194)

xt = 0.0150 + 0.8772yt–1 – 0.3459xt–1 + 0.2875 (yt–1 + 3.9170 – 0.5441xt–1) + e2t

(0.0525) (0.7235) (0.2594) (0.5916)