Financial Econometrics
Tutorial Exercise 9 Solutions
Question 1
AR(8) model is as follows:
ARIMA regression
Sample: 2 - 3907 Number of obs = 3906
Wald chi2(8) = 229.32
Log likelihood = 11708.15 Prob > chi2 = 0.0000
------------------------------------------------------------------------------
| OPG
[Link] | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lftse |
_cons | -.0000123 .000184 -0.07 0.947 -.0003729 .0003483
-------------+----------------------------------------------------------------
ARMA |
ar |
L1. | -.0450127 .010497 -4.29 0.000 -.0655865 -.0244389
L2. | -.0500389 .0090027 -5.56 0.000 -.0676839 -.032394
L3. | -.0831168 .0091547 -9.08 0.000 -.1010596 -.0651739
L4. | .0544616 .0092167 5.91 0.000 .0363972 .0725259
L5. | -.0508978 .0091385 -5.57 0.000 -.0688089 -.0329866
L6. | -.0361694 .0099771 -3.63 0.000 -.0557241 -.0166147
L7. | .0298347 .0102067 2.92 0.003 .0098299 .0498395
L8. | .0258334 .0106135 2.43 0.015 .0050313 .0466356
-------------+----------------------------------------------------------------
/sigma | .0120773 .0000733 164.87 0.000 .0119337 .0122208
------------------------------------------------------------------------------
Note: The test of the variance against zero is one sided, and the two-sided
confidence interval is truncated at zero.
1
LM test for GARCH effects is based on the following auxiliary regression:
2 2 2
e t =δ 0 +δ 1 et−1 +. . .+δ 12 et−12+v t
Source | SS df MS Number of obs = 3894
-------------+------------------------------ F( 12, 3881) = 92.89
Model | .000144178 12 .000012015 Prob > F = 0.0000
Residual | .000501973 3881 1.2934e-07 R-squared = 0.2231
-------------+------------------------------ Adj R-squared = 0.2207
Total | .000646152 3893 1.6598e-07 Root MSE = .00036
------------------------------------------------------------------------------
e2 | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
e2 |
L1. | .0106259 .0160427 0.66 0.508 -.0208271 .0420789
L2. | .1376461 .0160188 8.59 0.000 .1062401 .1690522
L3. | .1900582 .0160727 11.82 0.000 .1585465 .22157
L4. | .1341395 .0163377 8.21 0.000 .1021082 .1661708
L5. | .1246511 .0164634 7.57 0.000 .0923734 .1569288
L6. | -.0087989 .0165814 -0.53 0.596 -.0413081 .0237103
L7. | -.0185014 .0165813 -1.12 0.265 -.0510103 .0140075
L8. | -.0442693 .0164625 -2.69 0.007 -.0765453 -.0119934
L9. | .0519385 .0163379 3.18 0.001 .0199068 .0839701
L10. | .1093726 .0160733 6.80 0.000 .0778596 .1408856
L11. | .0551023 .0159913 3.45 0.001 .0237501 .0864546
L12. | .0319695 .0160181 2.00 0.046 .0005648 .0633742
|
_cons | .0000327 6.82e-06 4.79 0.000 .0000193 .0000461
------------------------------------------------------------------------------
The null hypothesis is: H0:1=...=12=0.
The test statistic is: = TR2 ~ 2(12) if H0 is true.
T = 3894 R2 = 0.2231 = 38940.2231 = 868.9
Decision rule: Accept H0 if χ 20 .05
2
Reject H0 if > χ 0 .05
χ 20 .05 from 2(12) is 21.03 Decision is reject H0
ARCH or GARCH effects are present.
2
Question 2
The three estimations required in this question are shown below.
1. AR(8)-ARCH(1) model:
ARIMA regression
Sample: 2 - 3907 Number of obs = 3906
Wald chi2(8) = 229.32
Log likelihood = 11708.15 Prob > chi2 = 0.0000
------------------------------------------------------------------------------
| OPG
[Link] | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lftse |
_cons | -.0000123 .000184 -0.07 0.947 -.0003729 .0003483
-------------+----------------------------------------------------------------
ARMA |
ar |
L1. | -.0450127 .010497 -4.29 0.000 -.0655865 -.0244389
L2. | -.0500389 .0090027 -5.56 0.000 -.0676839 -.032394
L3. | -.0831168 .0091547 -9.08 0.000 -.1010596 -.0651739
L4. | .0544616 .0092167 5.91 0.000 .0363972 .0725259
L5. | -.0508978 .0091385 -5.57 0.000 -.0688089 -.0329866
L6. | -.0361694 .0099771 -3.63 0.000 -.0557241 -.0166147
L7. | .0298347 .0102067 2.92 0.003 .0098299 .0498395
L8. | .0258334 .0106135 2.43 0.015 .0050313 .0466356
-------------+----------------------------------------------------------------
/sigma | .0120773 .0000733 164.87 0.000 .0119337 .0122208
------------------------------------------------------------------------------
Note: The test of the variance against zero is one sided, and the two-sided
confidence interval is truncated at zero.
3
2. AR(8)-ARCH(13) model:
ARCH family regression
Sample: 10 - 3907 Number of obs = 3898
Distribution: Gaussian Wald chi2(8) = 20.30
Log likelihood = 12499.32 Prob > chi2 = 0.0093
------------------------------------------------------------------------------
| OPG
[Link] | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lftse |
lftse |
LD. | -.0507088 .0164901 -3.08 0.002 -.0830289 -.0183887
L2D. | -.0310939 .0172248 -1.81 0.071 -.064854 .0026662
L3D. | -.0343304 .0169922 -2.02 0.043 -.0676346 -.0010263
L4D. | .0132373 .0166647 0.79 0.427 -.0194249 .0458995
L5D. | -.0271206 .0158592 -1.71 0.087 -.0582041 .0039628
L6D. | -.009055 .0161069 -0.56 0.574 -.040624 .0225139
L7D. | .0121692 .0159335 0.76 0.445 -.0190599 .0433984
L8D. | -.0083213 .0164358 -0.51 0.613 -.0405348 .0238923
|
_cons | .0004391 .0001313 3.34 0.001 .0001817 .0006964
-------------+----------------------------------------------------------------
ARCH |
arch |
L1. | .0644141 .0148145 4.35 0.000 .0353783 .0934499
L2. | .1306674 .0202931 6.44 0.000 .0908936 .1704412
L3. | .1210375 .0202321 5.98 0.000 .0813833 .1606916
L4. | .1117587 .0172343 6.48 0.000 .07798 .1455373
L5. | .0832166 .0186993 4.45 0.000 .0465666 .1198666
L6. | .0809189 .0182995 4.42 0.000 .0450525 .1167853
L7. | .0378929 .0134227 2.82 0.005 .0115849 .0642008
L8. | .0602592 .0159361 3.78 0.000 .0290251 .0914933
L9. | .0291812 .0147068 1.98 0.047 .0003565 .058006
L10. | .0335833 .01459 2.30 0.021 .0049873 .0621793
L11. | .0703043 .0153343 4.58 0.000 .0402497 .1003589
L12. | .0405259 .0134845 3.01 0.003 .0140968 .0669551
L13. | .035576 .0149039 2.39 0.017 .0063649 .0647871
|
_cons | .0000172 1.64e-06 10.44 0.000 .0000139 .0000204
------------------------------------------------------------------------------
4
3. AR(3)-ARCH(12) model:
ARCH family regression
Sample: 5 - 3907 Number of obs = 3903
Distribution: Gaussian Wald chi2(3) = 16.64
Log likelihood = 12508.43 Prob > chi2 = 0.0008
------------------------------------------------------------------------------
| OPG
[Link] | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lftse |
lftse |
LD. | -.0506947 .0163397 -3.10 0.002 -.0827199 -.0186695
L2D. | -.0297629 .0171462 -1.74 0.083 -.063369 .0038431
L3D. | -.0358244 .0169617 -2.11 0.035 -.0690687 -.0025801
|
_cons | .0004239 .0001304 3.25 0.001 .0001684 .0006794
-------------+----------------------------------------------------------------
ARCH |
arch |
L1. | .0628625 .0145925 4.31 0.000 .0342617 .0914633
L2. | .1328414 .0202547 6.56 0.000 .093143 .1725398
L3. | .1238245 .020145 6.15 0.000 .084341 .163308
L4. | .1120446 .0169238 6.62 0.000 .0788745 .1452147
L5. | .0877618 .0188854 4.65 0.000 .050747 .1247765
L6. | .0850323 .018756 4.53 0.000 .0482712 .1217934
L7. | .0455209 .0138337 3.29 0.001 .0184075 .0726344
L8. | .0602277 .0154362 3.90 0.000 .0299734 .090482
L9. | .0351049 .014796 2.37 0.018 .0061052 .0641046
L10. | .0355681 .0147577 2.41 0.016 .0066435 .0644926
L11. | .0713004 .0153886 4.63 0.000 .0411393 .1014615
L12. | .0415666 .0136439 3.05 0.002 .014825 .0683082
|
_cons | .000018 1.64e-06 10.98 0.000 .0000148 .0000212
------------------------------------------------------------------------------
5
Question 3
The AR(3)-GARCH(1,1) model is as follows:
ARCH family regression
Sample: 5 - 3907 Number of obs = 3903
Distribution: Gaussian Wald chi2(3) = 16.16
Log likelihood = 12515.7 Prob > chi2 = 0.0011
------------------------------------------------------------------------------
| OPG
[Link] | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lftse |
lftse |
LD. | -.053993 .0174066 -3.10 0.002 -.0881093 -.0198767
L2D. | -.0295708 .016792 -1.76 0.078 -.0624824 .0033409
L3D. | -.0330701 .0164831 -2.01 0.045 -.0653765 -.0007638
|
_cons | .0004082 .000131 3.12 0.002 .0001514 .0006651
-------------+----------------------------------------------------------------
ARCH |
arch |
L1. | .1036631 .0076451 13.56 0.000 .0886789 .1186472
|
garch |
L1. | .888038 .0079332 111.94 0.000 .8724892 .9035868
|
_cons | 1.38e-06 2.42e-07 5.72 0.000 9.09e-07 1.86e-06
------------------------------------------------------------------------------
The test to determine whether a fitted model containing an ARCH or GARCH component
adequately describes the time-path of the conditional variance, are based on the property that ut can
be written vtt, where vt = ut/t ~ N(0,1).
We can estimate vt using
v^ t =et / σ^ t = e t / √a^ 0 + a^ 1 e 2t−1 + b^ 1 σ^ 2t−1
v^
A Ljung-Box test can be used to determine whether t are white noise. The null hypthesis is:
H0:k=0 for k=1...12, where k is the ACF (autocorrelation function) for vt.
12 ^ρ2k
T (T +2) ∑
The test statistic is: = k =1 T −k ~ 2(12) if H is true.
0
Portmanteau test for white noise
---------------------------------------
Portmanteau (Q) statistic = 5.6995
Prob > chi2(13) = 0.9564
= 5.70 χ 20 .05 from 2(13) is 22.362 Decision is accept H0
v^
t are white noise, so the fitted model adequately describes the time-path of the
conditional
6
variance.
7
Question 4
The GJR threshold GARCH (or TARCH) model is as follows:
ARCH family regression
Sample: 5 - 3907 Number of obs = 3903
Distribution: Gaussian Wald chi2(3) = 9.83
Log pseudolikelihood = 12591.65 Prob > chi2 = 0.0200
------------------------------------------------------------------------------
| Semirobust
[Link] | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
lftse |
lftse |
LD. | -.0461048 .0165137 -2.79 0.005 -.078471 -.0137386
L2D. | -.0215618 .0176625 -1.22 0.222 -.0561797 .0130561
L3D. | -.0204926 .0171727 -1.19 0.233 -.0541505 .0131654
|
_cons | .0000192 .0001339 0.14 0.886 -.0002431 .0002816
-------------+----------------------------------------------------------------
ARCH |
arch |
L1. | .145324 .0204361 7.11 0.000 .10527 .185378
|
tarch |
L1. | -.152914 .0217008 -7.05 0.000 -.1954467 -.1103813
|
garch |
L1. | .9147153 .0122113 74.91 0.000 .8907817 .938649
|
_cons | 1.63e-06 3.60e-07 4.53 0.000 9.27e-07 2.34e-06
------------------------------------------------------------------------------
This model suggests the ARCH coefficient for a negative shock (=0.1453) is much larger than the
ARCH coefficient for a positive shock (=0.1453–0.1529=–0.0024, not significantly different from
zero). Therefore the asymmetry is large and highly significant.
