QUESTION 1

.15

.10

.05

.00

-.05

-.10

-.15

-.20

-.25
90 92 94 96 98 00 02 04 06 08 10 12

SPR BDR

This graph shows that the series exhibits volatility clustering meaning that the variance of the series is a function of past
shocks. This means that periods when large changes are followed by further large changes and periods when small
changes are followed by further small changes. In this specific case there is a low volatility in some periods such as from
92 to 96 and some high volatility in other periods such as 2008

QUESTION 2

To perform a significance test for each lag

Date: 11/18/21 Time: 13:36
Sample (adjusted): 1/06/1989 1/25/2013 separately we have to look the bar of the
Included observations: 1256 after adjustments autocorrelation and the partial correlation: If the
Autocorrelation Partial Correlation AC PAC Q-Stat Prob
bar exceeds the confidence interval means that the
1 0.283 0.283 100.92 0.000
correlation for that particular lag is statistically
2 0.149 0.075 128.84 0.000
3 0.213 0.167 185.84 0.000 significant because the value of the test-statistic
4 0.117 0.014 203.14 0.000
5 0.124 0.068 222.61 0.000
exceeds the critical one.
6 0.144 0.066 248.84 0.000
7 0.220 The autocorrelation from the first lag to the
0.158 309.79 0.000
8 0.064 -0.073 315.00 0.000
9 0.044
seventh lag are statistically significant because
-0.013 317.50 0.000
10 0.039 the bar exceeds the critical value. While the last
-0.042 319.47 0.000
three lags are not statistically significant
because both the autocorrelation and the partial correlation’s bar does not exceed the critical value.

We have also to say that p-value is zero meaning that it indicates serial correlation in squared residuals which means
evidence for conditional heteroscedasticity.
Heteroskedasticity Test: ARCH

F-statistic 58.31370 Prob. F(2,1251) 0.0000

Obs*R-squared 106.9376 Prob. Chi-Square(2) 0.0000 As we can see from the heteroscedastic ARCH effects test,
the p-value is zero so that’s mean that both lags are
Test Equation:
Dependent Variable: RESID^2 statistically significant
Method: Least Squares
Date: 11/22/21 Time: 16:45
Sample (adjusted): 1/20/1989 1/25/2013
Included observations: 1254 after adjustments

Variable Coefficient Std. Error t-Statistic Prob.

C 0.000369 4.78E-05 7.708720 0.0000

RESID^2(-1) 0.261941 0.028194 9.290692 0.0000
RESID^2(-2) 0.074719 0.028194 2.650165 0.0081
Date: 11/18/21 Time: 16:55
Sample (adjusted): 1/06/1989 1/25/2013
Included observations: 1256 after adjustments The autocorrelation and the partial correlation are
Autocorrelation Partial Correlation AC PAC Q-Stat Prob
statistically significant up the first until the sixth lag,
1 0.067 0.067 5.6671 0.017 including also the eighth and ninth lag. The partial
2 0.118 0.114 23.066 0.000
3 0.138 0.126 47.067 0.000 correlation of the lags from five to ten are not
4 0.163 0.140 80.572 0.000 statistically significant because it doesn’t’ exceed
5 0.075 0.036 87.681 0.000
6 0.098 0.050 99.923 0.000 the critical value.
7 0.043 -0.009 102.30 0.000
8 0.118 0.072 119.93 0.000 Also here we can see that the p-value is very close
9 0.080 0.042 127.99 0.000
10 0.018 -0.029 128.40 0.000 to zero meaning that there’s serial correlation and
so there’s conditional heteroscedasticity.

Heteroskedasticity Test: ARCH

F-statistic 11.03263 Prob. F(2,1251) 0.0000

Obs*R-squared 21.73481 Prob. Chi-Square(2) 0.0000

From the heteroscedastic ARCH effects test we can

Test Equation: say the p-value are close to zero so that’s mean that
Dependent Variable: RESID^2
Method: Least Squares both lags are statistically significant and so the null
Date: 11/22/21 Time: 17:05 hypothesis of no ARCH effect is rejected
Sample (adjusted): 1/20/1989 1/25/2013
Included observations: 1254 after adjustments

Variable Coefficient Std. Error t-Statistic Prob.

C 8.46E-05 6.26E-06 13.50625 0.0000

RESID^2(-1) 0.059451 0.028090 2.116446 0.0345
RESID^2(-2) 0.113550 0.028091 4.042157 0.0001

QUESTION 3
ARCH (1)
Dependent Variable: SPR
Method: ML ARCH - Normal distribution (BFGS / Marquardt steps) As can be seen from the test output, the Engle test is based on
Date: 11/22/21 Time: 17:21
Sample (adjusted): 1/06/1989 1/25/2013 the null hypothesis that there are no ARCH effects against the
Included observations: 1256 after adjustments
Convergence achieved after 9 iterations alternative hypothesis that the data is characterized by ARCH
Coefficient covariance computed using outer product of gradients
Presample variance: backcast (parameter = 0.7) disturbances.
GARCH = C(2) + C(3)*RESID(-1)^2

Variable Coefficient Std. Error z-Statistic Prob. The test shows a p-value of 0.0000, which is well below 0.05,
C 0.002182 0.000557 3.917419 0.0001 suggesting the presence of ARCH effects
Variance Equation
Included observations: 1256 after adjustments
C 0.000354 1.15E-05 30.87644 0.0000 Autocorrelation Partial Correlation AC PAC Q-Stat Prob*
RESID(-1)^2 0.336502 0.038020 8.850640 0.0000
1 0.004 0.004 0.0155 0.901
2 0.047 0.047 2.8474 0.241
3 0.101 0.101 15.747 0.001
4 0.062 0.060 20.614 0.000
Heteroskedasticity Test: ARCH 5 0.057 0.048 24.670 0.000
6 0.071 0.057 31.041 0.000
F-statistic 1.416253 Prob. F(2,1251) 0.2430 7 0.085 0.071 40.200 0.000
Obs*R-squared 2.832884 Prob. Chi-Square(2) 0.2426
8 0.019 0.003 40.677 0.000
9 0.021 -0.003 41.248 0.000
Test Equation: 10 0.036 0.011 42.882 0.000
Dependent Variable: WGT_RESID^2
Method: Least Squares
Date: 11/22/21 Time: 23:34
Sample (adjusted): 1/20/1989 1/25/2013
Included observations: 1254 after adjustments

Variable Coefficient Std. Error t-Statistic Prob.

C 0.950529 0.074230 12.80521 0.0000

WGT_RESID^2(-1) 0.003271 0.028241 0.115807 0.9078
WGT_RESID^2(-2) 0.047407 0.028242 1.678613 0.0935
GARCH (1, 1)
Dependent Variable: SPR
Method: ML ARCH - Normal distribution (BFGS / Marquardt steps) For the second model we can say that since the p-value is
Date: 11/22/21 Time: 17:25
Sample (adjusted): 1/06/1989 1/25/2013 less than 0.05 we reject the null hypothesis and confirm the
Included observations: 1256 after adjustments existence of ARCH effects. This means that the terms are
Convergence achieved after 16 iterations
Coefficient covariance computed using outer product of gradients statistically significant
Presample variance: backcast (parameter = 0.7)
GARCH = C(2) + C(3)*RESID(-1)^2 + C(4)*GARCH(-1)
As we can see from the heteroscedasticity ARCH test, the
Variable Coefficient Std. Error z-Statistic Prob.
constant is 0 meaning that is statistically significant. The
C 0.002323 0.000497 4.678135 0.0000 variance equation gives all statistically significant
Variance Equation parameters which means that model of GARCH (1,1)
C 2.30E-05 5.26E-06 4.368162 0.0000 appropriate here.
RESID(-1)^2 0.186014 0.017823 10.43675 0.0000
GARCH(-1) 0.782451 0.021514 36.36969 0.0000

Heteroskedasticity Test: ARCH

F-statistic 3.845372 Prob. F(2,1251) 0.0216

Obs*R-squared 7.662084 Prob. Chi-Square(2) 0.0217

Test Equation:
Dependent Variable: WGT_RESID^2
Method: Least Squares
Date: 11/22/21 Time: 23:36
Sample (adjusted): 1/20/1989 1/25/2013
Included observations: 1254 after adjustments

Variable Coefficient Std. Error t-Statistic Prob.

C 0.953671 0.064186 14.85790 0.0000

WGT_RESID^2(-1) 0.075098 0.028263 2.657096 0.0080
WGT_RESID^2(-2) -0.027865 0.028265 -0.985853 0.3244

GARCH (2,1)
Dependent Variable: SPR
Method: ML ARCH - Normal distribution (BFGS / Marquardt steps)
Date: 11/22/21 Time: 17:26
Sample (adjusted): 1/06/1989 1/25/2013
Included observations: 1256 after adjustments For the last model GARCH (2,1) as we can see the p value is less
Convergence achieved after 29 iterations
Coefficient covariance computed using outer product of gradients than 5% significance level meaning that exists ARCH effects
Presample variance: backcast (parameter = 0.7)
GARCH = C(2) + C(3)*RESID(-1)^2 + C(4)*GARCH(-1) + C(5)*GARCH(
-2)

Variable Coefficient Std. Error z-Statistic Prob.

C 0.002238 0.000493 4.535148 0.0000

Variance Equation

C 2.21E-05 5.65E-06 3.907227 0.0001

RESID(-1)^2 0.205289 0.022026 9.320431 0.0000
GARCH(-1) 0.323466 0.131001 2.469196 0.0135
GARCH(-2) 0.439907 0.124547 3.532065 0.0004

Heteroskedasticity Test: ARCH

F-statistic 0.829884 Prob. F(2,1251) 0.4363

Obs*R-squared 1.661544 Prob. Chi-Square(2) 0.4357

Test Equation:
Dependent Variable: WGT_RESID^2
Method: Least Squares
Date: 11/22/21 Time: 23:49
Sample (adjusted): 1/20/1989 1/25/2013
Included observations: 1254 after adjustments

Variable Coefficient Std. Error t-Statistic Prob.

C 0.963713 0.065300 14.75824 0.0000

WGT_RESID^2(-1) 0.036317 0.028274 1.284456 0.1992
WGT_RESID^2(-2) 0.001499 0.028275 0.053025 0.9577
QUESTION 4
Dependent Variable: SPR
Method: ML ARCH - Normal distribution (BFGS / Marquardt steps)
Date: 11/23/21 Time: 11:36
Sample (adjusted): 1/06/1989 1/25/2013
Included observations: 1256 after adjustments
Convergence achieved after 13 iterations
Coefficient covariance computed using outer product of gradients
Presample variance: backcast (parameter = 0.7)
LOG(GARCH) = C(2) + C(3)*ABS(RESID(-1)/@SQRT(GARCH(-1))) +
C(4)*RESID(-1)/@SQRT(GARCH(-1))

Variable Coefficient Std. Error z-Statistic Prob.

C 0.001154 0.000595 1.937279 0.0527

Variance Equation

C(2) -7.955479 0.036568 -217.5548 0.0000

C(3) 0.427576 0.045222 9.455121 0.0000
C(4) -0.174079 0.032448 -5.364779 0.0000
QUESTION 6

MARCH_1_SPR_BDR

.0004

.0002

.0000

-.0002

-.0004

-.0006
90 92 94 96 98 00 02 04 06 08 10 12

As we can see from the graph, the correlation depends on time

.016

.014

.012

.010

.008

.006

.004

.002

.000
90 92 94 96 98 00 02 04 06 08 10 12

ARCH_1_SPR MARCH_1_SPR

Comparing the estimated conditional variances with those obtained using corresponding univariate models we can see that
there are some differences, but they are so small and not visible on the graph