Dr. Ceylan Yozgatlıgil 22.4.

2009
METU-SPRING 2009
STAT 497
MIDTERM EXAM I

QUESTIONS

1. Eliminate parameter redundancy from the following series and determine if they are
stationary and/or invertible.
a) X t  0.8 X t 1  0.15 X t  2  at  0.3at 1
b) X t  X t 1  0.5 X t  2  at  at 1
c) Xt = −0.7Xt−2 + 1.5Xt−1 +at 0.2at-2

2. Suppose Cov  Zt , Zt k    k is free of t but E  Zt    t where   0 .

a) Is  Z t  stationary?
b) Let Yt     t  Zt . Is  Yt  stationary?

3. Consider the following ARMA(p, q) process

Yt  0.8Yt 1  at  0.5at 1


where a ~ WN 0,  2 .
t a 
a) Find the values of p and q.

b) Write the equation of the process using the backward shift operator B.

c) Write Yt in a random shock form, if possible.

d) Write Yt in an inverted form, if possible.

e) Find the ACVF and the ACF of this process.

 2
4. Let Yt  15  0.5t  Yt 1  at  0.6at 1 where at ~ WN 0,  a . 
a) Is Yt level stationary and/or trend stationary and/or difference stationary? Explain?

b) How would you eliminate the trend in this series? Explain.

c) After eliminating the trend, which ARMA process fits well to this series?

5. Following plots belongs to several time series data. Each row gives the time series plot, sample
autocorrelation function and sample partial autocorrelation function of one series. Describe the series

1
and explain whether series is stationary or not. If it is stationary, what is the order of ARMA model
for that series? If it is not stationary, explain whether the series is trend stationary and/or difference
stationary by just looking at the plots.

E
6. For three different time series data, the KPSS test is applied and following results are obtained.
Interpret these outputs, assuming that =0.05.

2
 a) > kpss.test(x)

KPSS Level = 0.1651, Truncation lag parameter = 7, p-value = 0.1

b) > kpss.test(y)

KPSS Level = 1.3194, Truncation lag parameter = 7, p-value = 0.01

c) > kpss.test(z, null = "Trend")

KPSS Trend = 0.0487, Truncation lag parameter = 7, p-value = 0.1

7. The hourly chemical viscosity readings are given and before fitting any model to the data, we want
to identify a proper model for this series. The time series plot is given in Table 1.

a) Interpret the plot.

Table 1. Hourly chemical viscosity readings

b) The sACF and sPACF plots are given in Table 2. Do you think that the series is stationary?
Explain.

3
c) Explain the following results.
Minimum Information Criterion

Lags MA 0 MA 1 MA 2 MA 3 MA 4 MA 5

AR 0 -1.12269 -1.33576 -1.52847 -1.70328 -1.81297 -1.91385

AR 1 -2.45114 -2.43443 -2.42587 -2.40751 -2.39275 -2.37503
AR 2 -2.44452 -2.42653 -2.40803 -2.38971 -2.37428 -2.35653
AR 3 -2.4261 -2.40823 -2.39096 -2.37256 -2.35645 -2.33911
AR 4 -2.40853 -2.3906 -2.37387 -2.3554 -2.338 -2.32077
AR 5 -2.39355 -2.37504 -2.35727 -2.33877 -2.32027 -2.30227

Error series model: AR(9)

Minimum Table Value: BIC(1,0) = -2.45114

d) We decided to take a regular difference and, the time series plot, sACF/sPACF plots and
inverse autocorrelations are given in Table 3. Interpret the plots and explain whether we did a
good job by differencing or not?

4
 Inverse Autocorrelations

Lag Correlation -1 9 8 7 6 5 4 3 2 1 0 1 2 3 4 5 6 7 8 9 1

1 -0.91429 | ******************| . |
2 0.88358 | . |****************** |
3 -0.79922 | ****************| . |
4 0.76953 | . |*************** |
5 -0.68733 | **************| . |
6 0.65809 | . |************* |
7 -0.58482 | ************| . |
8 0.55704 | . |*********** |
9 -0.49020 | **********| . |
10 0.46514 | . |********* |
Table 3. The time series plot, sACF/sPACF plots and inverse autocorrelations of the
differenced series.

Dr. Ceylan Yozgatlıgil 14.5.2009

METU-SPRING 2009
STAT 497
MIDTERM EXAM II
QUESTIONS

1. Consider the 100 yearly unit sales of X brand pencil. Based on the ACFs and PACFs of the
original series and its differencing, a reasonable model for the series is chosen as an AR(2)
after differencing. Therefore, we can assume an AR(3) model for the original series and test
for a unit root. Without using assumed initial values, we fit the ADF equation with t = 4,
5, ...,100. The OLS regression equation becomes
Yˆt  15.95 0.75 Yt 1  0.06 Yt 1  0.58 Yt  2
 3.10   0.05  0.08   0.08 

The corresponding standard errors for the estimators are given in the parenthesis. Calculate
the test statistics to test for a unit root and decide whether to reject the null hypothesis of
existence of a unit root using the DF critical values table.

5
2. For the multiplicative seasonal ARIMA model of order  0,1,1   0,1,2  6 ,
a) Write the model in operator and difference equation form.
b) Can we directly calculate the autocovariances for this model? Why?
c) After taking the regular and seasonal differences obtain the autocovariance function by
using the autocovariance generating function.

 
3. Consider the model 1  0.6 B  0.8B 2 Yt  10  1  0.2 B  at where at ’s are WN with mean
0 and  a2  1 .

a) Given Y58  15, Y59  13, Y60  10 and a59  0 , obtain a60 and then first three-step
ahead minimum MSE forecasts for Y61 , Y62 and Y63 , Yˆ60   ,   1, 2,3.
b) Find the  weights  j , j  1, 2,3 . State the recursive formula for  j when j  3 .
c) Calculate the forecast error variances for e60   ,   1, 2,3 and construct 95% prediction
intervals for Y61 , Y62 and Y63 .
d) Suppose that the observation at t = 61 turns out to be Y61  16 . Find the updated
forecast for Y62 .
4. Consider the following test results

a) For a series with nonzero mean, interpret the test results and discuss what should be done
next based on the test results.
Augmented Dickey-Fuller Unit Root Tests

Type Lags Rho Pr < Rho Tau Pr < Tau

Zero Mean 0 1.3889 0.9567 5.80 0.9999

1 1.3935 0.9571 7.63 0.9999
Single Mean 0 -0.0332 0.9542 -0.07 0.9493
1 0.0479 0.9583 0.17 0.9694
Trend 0 -106.956 0.0001 -10.72 <.0001
1 -113.606 0.0001 -7.47 <.0001

b) For a series with mean 48.13, interpret the test results and discuss what should be done
next based on the test results.

Augmented Dickey-Fuller Unit Root Tests

Type Lags Rho Pr < Rho Tau Pr < Tau

Zero Mean 1 -0.6838 0.5295 -0.56 0.4702

4 0.1941 0.7261 2.16 0.9924
Single Mean 1 -187.410 0.0001 -9.70 <.0001
4 -5.7476 0.3576 -4.00 0.0022
Trend 1 -515.772 0.0001 -15.72 <.0001
4 -6.5794 0.6853 -2.69 0.2421

6
5. The time series plot of quarterly US beer production between 1975 and 1997 is given in the
following figure

a) Interpret the figure.

b) The sample ACF and PACF of the seasonally differenced series are given below. Interpret the
results and suggest an approprite SARIMA model for this series.

1.0 1.0

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0.0 0.0
5 10 15 20 5 10 15 20

-0.2 -0.2

-0.4 -0.4

-0.6 -0.6

-0.8 -0.8

-1.0 -1.0

7
Dr. Ceylan Yozgatlıgil 4.6.2009
METU-SPRING 2009
STAT 497
FINAL EXAM

QUESTIONS

1. A certain quarterly time series of economic data consisting of n = 92 observations exhibited

seasonality (of period s = 4) as well as upward trend. The following table shows the first 12
autocorrelations of certain differences of the original series:

Wt 1 2 3 4 5 6 7 8 9 10 11 12 ˆ0 W

1  B 4 Yt .26 .06 −.14 −.38 -.12 .03 −.10 −.12 .03 −.04 .10 .05 25.8 1.8

1  B  1  B 4 Y−.39
t .04 .03 -.31 .10 .18 .01 −.11 .09 −.05 .06 −.10 38.2 0.2

Suggest an appropriate seasonal model (or models) for the series, to the extent that this is possible
based on the limited information, and calculate preliminary estimates for the parameters of the model,
including  a2 , using a method of moments technique.

2. Consider the following multiplicative seasonal ARIMA model

1  0.7 B  1  0.5 B 4 Yt  20  1  0.3B  at

where at ’s are WN with mean 0 and  a2  1 .

a) Give the order of this ARIMA model, i.e., what are p, d, q, P, D and Q?
b) Obtain first four-step ahead minimum MSE forecasts for Y101, Y102,Y103 and Y104 ,
Ŷ100   ,   1,2,3,4 using the following values:

Y95=17 Y96=24 Y97=25 Y98=18 Y99=16 Y100=22

a95=3.03 a96=−2.05 a97=0.1 a98=−1.01 a99= 0.01 a100= −3.147

c) Find the  weights  j , j  1, 2, 3 and 4 in the random shock form of the model.
d) Calculate the forecast error variances for e100   ,   1,2,3,4 and construct 95%
prediction intervals for Y101 , Y102 , Y103 and Y104 .

8
 3. What type of outliers can be seen in time series analysis? Explain each of them briefly.

4. U.S. monthly unemployment rate for ages 16 and over from January 1948 to April 2009 is analyzed.
a) Interpret the following time series plot.

VALUE
11

10

JAN45 JAN50 JAN55 JAN60 JAN65 JAN70 JAN75 JAN80 JAN85 JAN90 JAN95 JAN00 JAN05 JAN10

date

b) Explain the following results. What should we do next?

Augmented Dickey-Fuller Unit Root Tests

Type Lags Rho Pr < Rho

Zero Mean 1 0.3307 0.7636

12 -0.2512 0.6257
Single Mean 1 -7.8805 0.2230
12 -20.5753 0.0099
Trend 1 -8.8093 0.5207
12 -23.2649 0.0344

c) After obtaining a stationary series, following sACF and sPACF plots are observed.
Identify an appropriate ARIMA model using these plots.
Autocorrelations

Lag Covariance Correlation -1 9 8 7 6 5 4 3 2 1 0 1 2 3 4 5 6 7 8 9 1 Std Error

0 0.046679 1.00000 | |********************| 0

1 0.0052854 0.11323 | .|** | 0.036886
2 0.013945 0.29873 | .|****** | 0.037355
3 0.010261 0.21983 | . |**** | 0.040475
4 0.0079756 0.17086 | . |*** | 0.042069
5 0.0092051 0.19720 | . |**** | 0.043002
6 0.0039554 0.08474 | . |** | 0.044216
7 0.0024250 0.05195 | . |*. | 0.044436
8 0.0026688 0.05717 | . |*. | 0.044519
9 0.0012296 0.02634 | . |*. | 0.044618
10 -0.0042096 -.09018 | **| . | 0.044640
11 0.0013172 0.02822 | . |*. | 0.044887
12 -0.0073958 -.15844 | ***| . | 0.044911
13 -0.0020731 -.04441 | .*| . | 0.045665
14 -0.0034140 -.07314 | .*| . | 0.045724
15 -0.0025685 -.05503 | .*| . | 0.045883

Partial Autocorrelations

9
 Lag Correlation -1 9 8 7 6 5 4 3 2 1 0 1 2 3 4 5 6 7 8 9 1

1 0.11323 | .|** |
2 0.28963 | .|****** |
3 0.18024 | .|**** |
4 0.07092 | .|* |
5 0.08855 | .|** |
6 -0.02670 | *|. |
7 -0.07436 | *|. |
8 -0.02251 | .|. |
9 -0.01007 | .|. |
10 -0.13536 | ***|. |
11 0.02758 | .|* |
12 -0.12111 | **|. |
13 -0.01923 | .|. |
14 0.01524 | .|. |
15 0.04384 | .|* |

d) After deciding on two models, following estimation results are obtained for the identified
models. Identify the insignificant parameters in these models, if there are any. Which
estimation result is better? Discuss the reasons.
Maximum Likelihood Estimation I

Standard Approx
Parameter Estimate Error t Value Pr > |t| Lag

MA1,1 -0.03265 0.25806 -0.13 0.8993 1

MA1,2 0.38848 0.21695 1.79 0.0733 2
MA1,3 0.19670 0.18820 1.05 0.2960 3
MA1,4 -0.0028007 0.05721 -0.05 0.9610 4
MA1,5 -0.49940 0.12042 -4.15 <.0001 5
MA2,1 0.86632 0.04094 21.16 <.0001 12
AR1,1 -0.04123 0.26449 -0.16 0.8761 1
AR1,2 0.59787 0.22143 2.70 0.0069 2
AR1,3 0.36654 0.23543 1.56 0.1195 3
AR1,4 -0.42033 0.16175 -2.60 0.0094 5
AR2,1 0.64221 0.06508 9.87 <.0001 12

Variance Estimate 0.036835

Std Error Estimate 0.191923
AIC -325.291
SBC -274.692
Number of Residuals 735

Maximum Likelihood Estimation II

Standard Approx
Parameter Estimate Error t Value Pr > |t| Lag

MA1,1 0.84354 0.13364 6.31 <.0001 1

MA1,2 -0.54942 0.15698 -3.50 0.0005 2
MA1,3 0.76941 0.12211 6.30 <.0001 3
MA1,4 -0.60349 0.07979 -7.56 <.0001 4
MA2,1 0.85007 0.04253 19.99 <.0001 12
AR1,1 0.83573 0.14076 5.94 <.0001 1
AR1,2 -0.33166 0.16480 -2.01 0.0442 2
AR1,3 0.73561 0.13725 5.36 <.0001 3
AR1,4 -0.51011 0.10166 -5.02 <.0001 4
AR2,1 0.60088 0.06665 9.02 <.0001 12

Variance Estimate 0.036807

Std Error Estimate 0.191851
AIC -326.999
SBC -281
Number of Residuals 735

10
 e) You have decided on a best model fit but now you want to check the diagnostics. Several
diagnostic analyses are performed and following results are obtained. Explain each
analysis and discuss whether assumptions are violated or not. If violated, discuss what
should be done to correct this.
Autocorrelation Check of Residuals

To Chi- Pr >
Lag Square DF ChiSq --------------------Autocorrelations--------------------

6 . 0 . -0.011 0.009 0.014 0.006 -0.010 -0.052

12 5.64 2 0.0597 0.032 0.011 0.016 -0.032 0.030 0.032
18 7.52 8 0.4816 -0.020 -0.041 -0.014 0.003 -0.015 -0.007
24 17.64 14 0.2235 0.075 0.052 -0.042 -0.016 -0.049 -0.026
30 22.90 20 0.2937 -0.054 -0.026 0.045 -0.007 0.021 0.028
36 26.88 26 0.4155 0.011 -0.032 -0.024 0.034 -0.047 0.000
42 29.32 32 0.6027 -0.045 0.016 0.001 -0.027 0.008 0.011
48 38.09 38 0.4656 -0.024 -0.053 -0.072 0.031 0.010 0.038

Residual: Actual-Forecast
2

-1

-2

JAN45 JAN50 JAN55 JAN60 JAN65 JAN70 JAN75 JAN80 JAN85 JAN90 JAN95 JAN00 JAN05 JAN10

date

45

40

35

30
P
e
r 25
c
e
n 20
t

15

10

-1.5 -1.3 -1.1 -0.9 -0.7 -0.5 -0.3 -0.1 0.1 0.3 0.5 0.7 0.9 1.1

Residual: Actual-Forecast

11
 Goodness-of-Fit Tests for Normal Distribution

Test ---Statistic---- -----p Value-----

Kolmogorov-Smirnov D 0.04645115 Pr > D <0.010

Cramer-von Mises W-Sq 0.52814031 Pr > W-Sq <0.005
Anderson-Darling A-Sq 3.76809269 Pr > A-Sq <0.005

1.5
R
e
s
i 1.0
d
u
a
0.5
l
:

A 0
c
t
u
a -0.5
l
-
F
-1.0
o
r
e
c -1.5
a
s
t
-2.0

-4 -3 -2 -1 0 1 2 3 4

Normal Quantiles

Godfrey's Serial Correlation Test

Alternative LM Pr > LM

AR(1) 0.1582 0.6908

AR(2) 0.2055 0.9024
AR(3) 0.3328 0.9537

Autocorrelations of the squared residuals

Lag Covariance Correlation -1 9 8 7 6 5 4 3 2 1 0 1 2 3 4 5 6 7 8 9 1 Std Error

0 0.011999 1.00000 | |********************| 0

1 0.0040506 0.33758 | .|******* | 0.036886
2 0.00066280 0.05524 | . |*. | 0.040873
3 0.00023046 0.01921 | . | . | 0.040975
4 0.00001573 0.00131 | . | . | 0.040987
5 0.00071475 0.05957 | . |*. | 0.040987
6 0.0010449 0.08708 | . |** | 0.041105
7 0.00007174 0.00598 | . | . | 0.041355
8 0.00041095 0.03425 | . |*. | 0.041356
9 0.00081034 0.06753 | . |*. | 0.041395
10 0.00035645 0.02971 | . |*. | 0.041544

Partial Autocorrelations of the squared residuals

Lag Correlation -1 9 8 7 6 5 4 3 2 1 0 1 2 3 4 5 6 7 8 9 1

1 0.33758 | .|******* |
2 -0.06627 | *|. |
3 0.02459 | .|. |
4 -0.01110 | .|. |
5 0.07115 | .|* |
6 0.04897 | .|* |
7 -0.04531 | *|. |
8 0.05434 | .|* |
9 0.04447 | .|* |
10 -0.00978 | .|. |

Heteroscedasticity Test
Equation Test Statistic DF Pr > ChiSq Variables

RESIDUAL White's Test 0.00 0 <.0001 Cross of all vars

12
 f) The following forecasts and 95% confidence limits are obtained at the end of the model
fit. Interpret the plot and discuss the reasonability of the forecasts.

VALUE
12

11

10

JAN72 JAN74 JAN76 JAN78 JAN80 JAN82 JAN84 JAN86 JAN88 JAN90 JAN92 JAN94 JAN96 JAN98 JAN00 JAN02 JAN04 JAN06 JAN08 JAN10

date

Dr. Ceylan Yozgatlıgil 19.4.2009

METU-SPRING 2010
STAT 497
MIDTERM EXAM I
QUESTIONS

1. (20 pts.) Consider the following processes where it is assumed that {at}’s are White Noise
with zero mean and  a2 variance.
a. Yt  0.4Yt 1  at  0.6at  2
b. Yt  0.8Yt 1  0.3Yt  2  at
c. Yt  0.2Yt 1  at  1.2at 1  0.8at  2
d. Yt  1.5Yt 1  0.56Yt  2  at  1.3at 1  0.4at  2
i) For each of the above process, identify the process as ARMA(p,q).
ii) Express each of the above process in backshift operator form.
iii) For each of the above process, determine if they are stationary and/or invertible (p.s.
please eliminate any parameter redundancy, if exists).

13
2. (15 pts.) Consider the process

Yt  10  0.5Yt 1  0.5Yt  2  at  0.3at 1

 
where at ~ WN 0,  a2 . .

a. Is the process stationary?

b. Write the model in Random Shock Form and Inverted Form. If you cannot write them in
desired forms, explain the reason and take necessary actions to write the process in these
forms.
c. Find the mean and the autocovariance function of the model. If you cannot find them,
explain the reason and take necessary actions to find the mean and autocovarince function
of the process.

3. (15 pts.) To be able to explain the variations in quarterly profit of U.S. manufacturing
corporations, the sales for U.S. manufacturing corporations is used. Besides of sales, linear
time trend and sesonal dummies (S1, S2, S3 and S4) are used to explain the seasonal variation
in the series. Time series plot of the profit series is given below:

The following two equations are obtained:

The first equation is:
Profit = 0.2784*Sales-2755*Trend-26510*S1-27000*S2-27030*S3-28160*S4
(0.04) (457) (5605) (5913) (5662) (5924)

Residual standard error: 642.7 on 18 degrees of freedom

Multiple R-squared: 0.9981, Adjusted R-squared: 0.9975
F-statistic: 1612 on 6 and 18 DF, p-value: < 2.2e-16

The second equation is:

14
 Profit = -26510+0.2784*Sales-2755*Trend-491.4*S2-515.7*S3-1650*S4
(5605) (0.04) (457) (483) (377) (492)

Residual standard error: 642.7 on 18 degrees of freedom

Multiple R-squared: 0.8426, Adjusted R-squared: 0.7989
F-statistic: 19.28 on 5 and 18 DF, p-value: 1.161e-06

Interpret the coefficients of these equations and explain the difference of them in terms of
explaining the variation of profit series.

4. (10 pts.) For one time series, the KPSS test is applied and following results are obtained.
Interpret these outputs, assuming that =0.05.

KPSS Test for Trend Stationarity

KPSS Trend = 0.0523, Truncation lag parameter = 1, p-value = 0.1
Warning message:
In kpss.test(x, null = c("Trend")) : p-value greater than printed p-value

KPSS Test for Level Stationarity

KPSS Level = 2.561, Truncation lag parameter = 1, p-value = 0.01
Warning message:
In kpss.test(x, null = c("Level")) : p-value smaller than printed p-value

5. (10 pts.) Monthly values of the The Southern Oscillation Index (SOI) during 1950-1995. This
series consists of 540 observations on the SOI, computed as the ``difference of the departure from
the long-term monthly mean sea level pressures'' at Tahiti in the South Pacific and Darwin in
Northern Australia. The index is one measure of the so-called "El Nino-Southern Oscillation" -- an
event of critical importance and interest in climatological studies in recent decades. The fact that
most of the observations in the last part of the series take negative values is related to a recent
warming in the tropical Pacific, and this pattern has been of key interest in recent studies. A key
question of interest is to determine just how unusual this event is, and if it can reasonably be
explained by standard "stationary" time series models, or requires models that include drifts/trends
that may be related to global climatic change.

The following analysis are done.

15
 KPSS Test for Level Stationarity
KPSS Level = 0.6675, Truncation lag parameter = 5, p-value = 0.01650
KPSS Test for Trend Stationarity
KPSS Trend = 0.1412, Truncation lag parameter = 5, p-value = 0.05897
Phillips-Perron Unit Root Test
Dickey-Fuller Z(alpha) = -230.5145, Truncation lag parameter = 6,
p-value = 0.01
Warning message: In pp.test(soi): p-value smaller than printed p-value

a. Interpret the time series plot for this series.

b. Interpret the ACF nad PACF plots. Can you identify any ARMA(p,q) process using these
plots.
c. Interpret the results of KPSS and PP test results. What should we do next?

6. (20 pts.) Consider the following stochastic processes:

i) Yt  at  0.5at 1
ii) Yt  1.2Yt 1  0.2Yt  2  at
iii) Yt  0.1Yt 1  0.7Yt  2  at
iv) Yt  20  0.7t  at

16
 B

Match the presented time series plots A to D with the generating stochastic processes given by i) to
vi). Explain the reasons of your identification.
7. (10 pts.) Following time series plot and ACF/PACF plots belongs to Ice Cream and Frozen
Desserts PPI from January 2000 to December 2009

a) Interpret these plots. What are the necessary steps that we need to consider next?

17
 b) We decided to take a regular and seasonal difference to work with a statioanry series. The
time series plots and ACF/PACF plots of the differenced series given below. Interpret these
plots. Identify an appropriate SARIMA(p,d,q)(P,D,Q)s model for this series.

Dr. Ceylan Yozgatlıgil 3.6.2010

METU-SPRING 2010
STAT 497
FINAL EXAM
QUESTIONS

1. A time series of n = 100 observations gave the following sample ACF and sample PACF,
Lag k 1 2 3 4 5 6 7 8
rk 0.22 −0.68 −0.47 −0.34 0.56 −0.01 −0.51 −0.26
kk 0.22 −0.76 −0.09 0.05 0.09 0.04 −0.11 −0.11

with the sample mean Z  10.04 and sample variance ˆ0  2.77 . An AR(2) model,

 1 B  B  Z
1 2
2
t    at , appears to be appropriate for the data.

a) Obtain the Yule-Walker estimates ˆ1 and ˆ2 . Also, obtain estimates for  and  a2 .

b) Using ˆ1 and ˆ2 as the true values of 1 and 2 , calculate  k , k  1, 2,3, 4.
2. Consider the following ARIMA model

1  1.8B  0.81B 2 Yt  0.2  at

where at ’s are WN with mean 0 and  a2  4 .
e) Find the mean of the process.
f) Find the variance of the process.
g) Obtain the first three-step ahead minimum MSE forecasts for Y51 ,Y52 , and Y53 ,
Ŷ50   ,   1,2 ,3 using the following values:

18
 Y47=19 Y48=22 Y49=17 Y50=21

h) Find the  weights  j , j  1, 2,3 in the random shock form of the model.
i) Calculate the forecast error variances for e50   ,   1,2 ,3 and construct 95% prediction
intervals for Y51 ,Y52 and Y53 .
3. Consider the ARCH(1) model nt   t et , where et are i.i.d. random variable with mean 0 and
variance 1, and  t2   0  1nt21 . Show that the unconditional variance of nt is
0
Var  nt   .
1  1
4. Consider the two-dimensional vector AR(1) process

Y1t  0   Y1,t 1  a1t 

Yt= Y   0     
 2t   1 Y2,t 1  a2t 

a) Is the series I(1) or I(0)? Show.

b) Is Yt cointegrated? If so, find the cointegrating vector.
c) Write the process in error correction representation form.
5. Lydia Pinkham’s Vegetable Compound was first brewed and prepared by the lady of the same
name as a home remedy of her friends. The compound, an herbal extract in alcoholic solution,
was considered to be effective against “women’s weakness”. The first commercial sale of the
medicine was in 1873, when sales reached $500,000. A part of the profits was reinvested
immediately into high intensity advertising campaigns. Records of a court litigation involving
a family argument produced a complete listing of the firm’s annual advertising expenditures
and sales over the period 1908-1935. Palda (1964) was later able to obtain the data from
1907-1960.

a) Interpret the below time series plot.

Time series plot of Lydia Pinkham’s Advertising and Sales Data from 1907-1960

19
 4000

3000

Sales
2000

Advertisement
1000

1900 1910 1920 1930 1940 1950 1960

b) What is the purpose of the Granger-Causality test? Explain. Interpret the below output.
Are the results logical? Explain.
Granger-Causality Wald Test

Test DF Chi-Square Pr > ChiSq

1 1 16.91 <.0001
2 1 8.73 0.0031

Test 1: Group 1 Variables: Advertising

Group 2 Variables: Sales

Test 2: Group 1 Variables: Sales

Group 2 Variables: Advertising

6. Consider the number of visitors (in thousands) to a Colorado ski resort data between the first
quarter of 1981 to last quarter of 1986.
a) Interpret the below time series plot.

20
 b) Considering the below sample ACF, sample PACF plots and ADF test results what
should we do next? Why? Explain.

Augmented Dickey-Fuller Unit Root Tests

Type Lags Tau Pr < Tau

Zero Mean 1 5.02 0.9999

4 0.67 0.8508
Single Mean 1 0.01 0.9500
4 -0.44 0.8839
Trend 1 -2.96 0.1638
4 -2.14 0.4955
c) After regular and seasonal differencing, we fit a certain model to the data. The results are given in
the below table:

R output

21
Coefficients:
ar1 sar1
-0.7166 -0.6931
s.e. 0.1526 0.1645

sigma^2 estimated as 2.298: log likelihood = -36.36, aic = 78.73

Which model is fitted to the data? Are all parameters in the model significant? Write the model.

d) The diagnostic analysis on the residuals are given below. Explain each graph and test result. Do
you think that the model that you found in part c) is a plausible model for the data? If not, what do
you suggest?

Autocorrelation Check of Residuals

To Chi- Pr >
Lag Square DF ChiSq -------------Autocorrelations------------

6 2.72 4 0.6065 -0.150 0.014 0.115 -0.097 0.090 -0.214

12 5.59 10 0.8486 -0.038 -0.138 -0.050 -0.009 -0.200 0.028
18 10.37 16 0.8463 0.214 -0.026 0.108 0.061 0.010 0.003

Breusch-Godfrey test for serial correlation of order 1

data: resid(ski.fit) ~ ski

LM test = 1.6141, df = 1, p-value = 0.2039

22
 Jarque Bera Test
data: resid(ski.fit)
X-squared = 10.897, df = 2, p-value = 0.004303

The Sample ACF, Sample PACF plots and “Autocorrelation Check for White Noise” of
Squared Residuals
Autocorrelations

Lag Covariance Correlation -1 9 8 7 6 5 4 3 2 1 0 1 2 3 4 5 6 7 8 9 1

0 17.032300 1.00000 | |********************|

1 5.793277 0.34013 | . |******* . |
2 -2.034588 -.11945 | . **| . |
3 -4.166656 -.24463 | . *****| . |
4 -3.092140 -.18155 | . ****| . |
5 1.312139 0.07704 | . |** . |
6 0.451033 0.02648 | . |* . |
7 -0.604024 -.03546 | . *| . |
8 -1.900935 -.11161 | . **| . |
9 -2.330850 -.13685 | . ***| . |
10 -2.459929 -.14443 | . ***| . |
11 1.310643 0.07695 | . |** . |
12 1.717580 0.10084 | . |** . |
13 -0.937858 -.05506 | . *| . |
14 -1.515496 -.08898 | . **| . |
15 -0.824299 -.04840 | . *| . |
16 0.249750 0.01466 | . | . |

"." marks two standard errors

Partial Autocorrelations

Lag Correlation -1 9 8 7 6 5 4 3 2 1 0 1 2 3 4 5 6 7 8 9 1

1 0.34013 | . |******* . |
2 -0.26591 | . *****| . |

23
 3 -0.12504 | . ***| . |
4 -0.08301 | . **| . |
5 0.13486 | . |*** . |
6 -0.15290 | . ***| . |
7 -0.00885 | . | . |
8 -0.10794 | . **| . |
9 -0.06764 | . *| . |
10 -0.18666 | . ****| . |
11 0.17740 | . |**** . |
12 -0.12110 | . **| . |
13 -0.11265 | . **| . |
14 -0.05091 | . *| . |
15 0.05536 | . |* . |
16 -0.14527 | . ***| . |

Autocorrelation Check for White Noise

To Chi- Pr >
Lag Square DF ChiSq -------------Autocorrelations------------

6 5.46 6 0.4864 0.340 -0.119 -0.245 -0.182 0.077 0.02

12 8.50 12 0.7450 -0.035 -0.112 -0.137 -0.144 0.077 0.10

Studentized Breusch-Pagan test

data: resid(ski.fit) ~ ski
BP = 2.8486, df = 1, p-value = 0.09145

Dr. Ceylan Yozgatlıgil 8.4.2011

METU-SPRING 2011
STAT 497
MIDTERM EXAM I
(Time Duration: 90 minutes)

Name:__________________________________________________________________

Please read the Academic Integrity Pledge:

I pledge to uphold the high standards of academic ethics and integrity expressed by the Middle East
Technical University Student Honor Code by which I am bound. In particular, “I will not
misrepresent the work of others as my own, nor will I give or receive unauthorized assistance in the
performance of academic coursework.” I understand that my instructor will report any infraction of
academic integrity to the Department Head and that any such matter will be investigated and
prosecuted fully.

Signature: ______________

QUESTIONS

1. (20 pts.) Consider the following processes where it is assumed that {at}’s are White Noise
with zero mean and  a2 variance.

24
 a. Yt  0.2Yt 1  at  0.1at 1
b. Yt  0.1Yt 1  0.12Yt  2  at  .9at 1  0.2at  2
c. Yt  0.2Yt 1  0.8Yt  2  at  0.3at 1  0.4at  2
iv) For each of the above process, identify the process as ARMA(p,q).
v) Express each of the above process in backshift operator form.
vi) For each of the above process, determine if they are stationary and/or invertible (p.s.
please eliminate any parameter redundancy, if exists).
i .i .d
2. (15 pts.) Let Xt be a MA(1) process: X t  at  at 1 where  at  ~ N  0,1 . Let
Yt  X t  X t 1 ,t  1,2 ,
a) Define the name of the process for Yt  . Write the form of the process.
b) Derive the autocorrelation function of Yt  .

3. (20 pts.) Consider the following process

Yt  Yt 1  0.6Yt  2  1  at  0.5at 1


where at ~ WN 0,  a2 . . 
d. Is the process stationary?
e. Write the model in Random Shock Form and Inverted Form. If you cannot write them in
desired forms, explain the reason and take necessary actions to write the process in these
forms.
f. Find the mean and the autocovariance function of the model. If you cannot find them,
explain the reason and take necessary actions to find the mean and autocovarince function
of the process.

4. (10 pts.) Below output is obtained after applying a first order difference to the series. What
can you conclude about the necessity of differencing and the order of the process?
Autocorrelations

Lag Covariance Correlation -1 9 8 7 6 5 4 3 2 1 0 1 2 3 4 5 6 7 8 9 1 Std Error

0 1.302091 1.00000 | |********************| 0

1 -0.475096 -.36487 | *******| . | 0.100504
2 0.048474 0.03723 | . |* . | 0.113095
3 -0.201388 -.15467 | . ***| . | 0.113219
4 -0.112572 -.08645 | . **| . | 0.115333
5 0.250546 0.19242 | . |****. | 0.115986
6 -0.026669 -.02048 | . | . | 0.119167
7 -0.187242 -.14380 | . ***| . | 0.119203
8 0.021515 0.01652 | . | . | 0.120942
9 -0.044344 -.03406 | . *| . | 0.120965
10 0.089578 0.06880 | . |* . | 0.121062
11 0.033087 0.02541 | . |* . | 0.121456
12 -0.019063 -.01464 | . | . | 0.121510
13 -0.130010 -.09985 | . **| . | 0.121527
14 0.221537 0.17014 | . |*** . | 0.122353
15 -0.109772 -.08430 | . **| . | 0.124720

Inverse Autocorrelations

Lag Correlation -1 9 8 7 6 5 4 3 2 1 0 1 2 3 4 5 6 7 8 9 1

25
 1 0.69470 | . |************** |
2 0.59781 | . |************ |
3 0.63658 | . |************* |
4 0.57689 | . |************ |
5 0.44585 | . |********* |
6 0.52649 | . |*********** |
7 0.49690 | . |********** |
8 0.40663 | . |******** |
9 0.40134 | . |******** |
10 0.35074 | . |******* |
11 0.26079 | . |***** |
12 0.27159 | . |***** |
13 0.27067 | . |***** |
14 0.17777 | . |**** |
15 0.21052 | . |**** |

Partial Autocorrelations

Lag Correlation -1 9 8 7 6 5 4 3 2 1 0 1 2 3 4 5 6 7 8 9 1

1 -0.36487 | *******| . |
2 -0.11063 | . **| . |
3 -0.21015 | ****| . |
4 -0.27139 | *****| . |
5 0.03307 | . |* . |
6 0.03035 | . |* . |
7 -0.21807 | ****| . |
8 -0.12548 | .***| . |
9 -0.07506 | . **| . |
10 -0.09722 | . **| . |
11 -0.07667 | . **| . |
12 -0.00303 | . | . |
13 -0.15226 | .***| . |
14 0.06496 | . |* . |
15 -0.00928 | . | . |

26
 5. (20 pts.) Consider the following stochastic processes:
v) Yt  0.1Yt 1  0.9Yt  2  at
vi) Yt  at  0.5at 1  0.8at  2
vii) Yt  0.6Yt 1  0.4Yt  2  at  0.5at 1  0.8at  2  0.3at  3
viii) Yt  50  0.5t  0.3t 2  at

Match the presented time series plots A to D with the generating stochastic processes given by i) to
vi). Explain the reasons of your identification.
6. (15 pts.) Consider the analysis of quarterly U.S. GNP from 1947(1) to 2002(3), n = 223
observations. The data are Real U.S. Gross National Product in billions of chained 1996

27
 dollars and they have been seasonally adjusted. The data were obtained from the Federal
Reserve Bank of St. Louis (http://research.stlouisfed.org/).

c) Interpret these plots.

d) By considering the above ACF plot and below output, what can you say about the stationarity
of the series? What are the necessary steps that we need to consider next?

KPSS Test for Trend Stationarity

data: gnp
KPSS Trend = 1.1734, Truncation lag parameter = 3, p-value = 0.01

KPSS Test for Level Stationarity

data: gnp
KPSS Level = 5.49, Truncation lag parameter = 3, p-value = 0.01

e) After having a stationary process, the time series plots and ACF/PACF plots of the
differenced series given below. Interpret these plots. Identify an appropriate ARIMA(p,d,q)
model for this series.

EACF TABLES

28
 AR/MA (gnp) AR/MA (diff(gnp)
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
0 xxxxxxxxxxx 0 xxxoooooooo
1 xxooooooooo 1 xoooooooooo
2 xoooooooooo 2 ooooooooooo
3 xxooooooooo 3 xoxoooooooo
4 xxooooooooo 4 oxxoooooooo
5 oxooooooooo 5 xxooooooooo
6 oxxoooooooo 6 xxoxoooxooo
7 xoxooooxooo 7 xooxooxoooo
8 xoxoooxxooo 8 xoxxxoxoooo
9 xxxoooxoooo 9 xoxoxxxoooo
10 x o x x x x x x o o o 10 x o x o o o x x o o o

Minimum Information Criterion

Lags MA 0 MA 1 MA 2 MA 3 MA 4 MA 5

AR 0 15.43702 14.80751 14.69275 14.63935 14.59999 14.56535

AR 1 7.978374 7.529241 7.449925 7.461506 7.477057 7.501157
AR 2 7.520222 7.467011 7.456928 7.479061 7.500797 7.500118
AR 3 7.445317 7.456324 7.475954 7.497819 7.511047 7.523381
AR 4 7.457902 7.477145 7.495089 7.506347 7.527725 7.526675
AR 5 7.47609 7.500288 7.511022 7.527523 7.55025 7.550153

Error series model: AR(5)

Minimum Table Value: BIC(3,0) = 7.445317
Dr. Ceylan Yozgatlıgil 6.5.2011
METU-SPRING 2011
STAT 497
MIDTERM EXAM II

QUESTIONS

1. Suppose that Yt  satisfies

Yt  a  bt  S t  X t
where St is deterministic and periodic with period s and Xt is a seasonal
ARIMA(p,0,q)x(P,1,Q)s series. What is the model for Wt  Yt  Yt  s ?

2. Based on the quarterly data, a seasonal model of the form

Yt  at  1at 1   2 at  4
has been fit to a certain time series.
a) Find the first four  weights for this model.
b) Suppose that 1  0.8, 2  0.5, and  a2  2. Find forecasts for the next four quarters, if
data for the last four quarters are

Quarters I II III IV

Series 15 10 15 30

29
 Residuals 1 2 2 3

3. Suppose that annual sales (in millions of dollars) of the Acne Corporation follow the AR(2)
model

Yt  10  0.1Yt 1  0.6Yt  2  at

with  a2  1.

a) If sales for 2006, 2007, and 2008 were $5 million, $8 million, and $9 million, respectively,
forecast for 2009 and 2010.
b) Show that 1  0.1 for this model.
c) Calculate 95% prediction limits for your forecasts in part a) for 2009 and 2010.
d) If sales in 2009 turn out to be $10 million, update your forecast for 2010.

4. Consider the model Yt   0  1t  X t , where X t  X t 1  at . We assume that

 0 , 1 , and  are known. Show that the minimum mean square error forecast  steps ahead
can be written as Ŷt     0  1  t       Yt   0  1t  .

5. Levels of carbon dioxide (CO2) are monitored at several sites around the world to investigate
atmospheric changes. One of the sites is at Alert, Northwest Territories, Canada, near the
Arctic Circle. The time series plot of monthly CO2 levels from January 1994 through
December 2004 is given below.

a) Interpret the above time series plot.

b) The Least Square fit for the monthly time trend with seasonal dummies is given below.
Interpret the regression output.
Call:
lm(formula = co2 ~ 0 + trend + Q, na.action = NULL)

Residuals:
Min 1Q Median 3Q Max
-1.73874 -0.59689 -0.06947 0.54086 2.15539

30
Coefficients:
Estimate Std. Error t value Pr(>|t|)
trend 1.8321 0.0221 82.90 <2e-16 ***
Q1 -3290.5412 44.1790 -74.48 <2e-16 ***
Q2 -3289.8730 44.1809 -74.46 <2e-16 ***
Q3 -3289.5775 44.1827 -74.45 <2e-16 ***
Q4 -3289.3102 44.1846 -74.44 <2e-16 ***
Q5 -3289.0138 44.1864 -74.44 <2e-16 ***
Q6 -3291.2173 44.1883 -74.48 <2e-16 ***
Q7 -3297.8264 44.1901 -74.63 <2e-16 ***
Q8 -3303.9827 44.1919 -74.76 <2e-16 ***
Q9 -3303.3617 44.1938 -74.75 <2e-16 ***
Q10 -3298.8017 44.1956 -74.64 <2e-16 ***
Q11 -3294.4689 44.1975 -74.54 <2e-16 ***
Q12 -3291.8779 44.1993 -74.48 <2e-16 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.8029 on 119 degrees of freedom

Multiple R-squared: 1, Adjusted R-squared: 1
F-statistic: 2.146e+06 on 13 and 119 DF, p-value: < 2.2e-16
c) The below graph is the plot of actual values (line with dots) versus fitted values (straight line).
Interpret the output and discuss the validty of the fit. Moreover, discuss the assumptions that
can be made during the estimation.

6. Consider the monthly U.S. electricity generation (in millions of kilowatt hours) of all types:
coal, natural gas, nuclear, petroleum, and wind, 01/1973 - 12/2005.
a) The time series plot, ACF and PACF plots are given below. Please interpret the plots.

31
 b) ADF and HEGY test results can be seen in the below output. Please interpret the outputs
and explain what should be the next step.
## ADF test with constant, no trend and seasonal dummies.
> ADF.test(wts=electricity, itsd=c(1,0,c(1:11)),regvar=0, selectlags=list(mode="bic",
Pmax=12))
--------- ------ - ------ ----
Augmented Dickey & Fuller test
--------- ------ - ------ ----

ADF statistic:

Estimate Std. Error t value Pr(>|t|)

adf.reg 0.002 0.007 0.334 0.1

## ADF test with constant, trend and seasonal dummies.

> ADF.test(wts=electricity, itsd=c(1,1,c(1:11)),regvar=0, selectlags=list(mode="bic",
Pmax=12))

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

Augmented Dickey & Fuller test
--------- ------ - ------ ----

ADF statistic:

Estimate Std. Error t value Pr(>|t|)

adf.reg -0.094 0.051 -1.854 0.1

> HEGY.test(wts=electricity,itsd=c(1,1,c(1:11)),regvar=0, selectlags=list(mode="bic",

Pmax=12))

---- ----
HEGY test
---- ----
HEGY statistics:

Stat. p-value
tpi_1 -1.494 0.10
tpi_2 -3.856 0.01
Fpi_3:4 33.858 0.10
Fpi_5:6 28.442 0.10
Fpi_7:8 3.747 0.01
Fpi_9:10 12.883 0.10
Fpi_11:12 2.303 0.01
Fpi_2:12 20.113 NA
Fpi_1:12 18.453 NA

Lag orders: 3 12
Number of available observations: 372

c) After a regular difference, we obtained the below figures.

32
After a seasonal difference, we have

and after a regular and seasonal difference we have

What is your preliminary guess about the order of the seasonal multiplicative model? Please write the
model that you suggest?

33
Dr. Ceylan Yozgatlıgil 10.6.2011
METU-SPRING 2011
STAT 497
FINAL EXAM
QUESTIONS

1. A non-seasonal time series of n = 100 observations gave the following sample ACF and
sample PACF for the original data,
Lag k 1 2 3 4 5 6 7 8 9 10
rk 0.73 0.11 −0.50 −0.79 −0.63 −0.15 0.36 0.64 0.55 0.17
kk 0.73 −0.91 0.09 0.03 0.08 −0.05 −0.03 0.12 −0.14 −0.05

with the sample mean Y  0.07 and sample variance ˆ 0  5.29 and for the first differenced data
we have
Lag k 1 2 3 4 5 6 7 8 9 10
rk 0.65 −0.01 −0.61 −0.84 −0.59 −0.06 0.44 0.68 0.54 0.11
kk 0.65 −0.75 −0.42 −0.33 −0.15 −0.16 −0.24 0.00 −0.09 −0.13

with mean 0 and variance 2.89.

a) Identify the model from the given information.
b) Obtain the Yule-Walker estimates of the parameters. Also, obtain estimates for the error
variance.
c) Using the estimates as the true parameter values, calculate  k , k  1, 2,3, 4.
2. Consider the following multiplicative seasonal ARIMA model

1  0.2 B  1  0.8B 4 Yt 

 50  1  0.6 B  1  0.3B 4 at
where at ’s are WN with mean 0 and  a2  1 .
a) Find the mean of the process.
b) If we have 200 data points, obtain first three-step ahead minimum MSE forecasts for
Y201 ,Y202 and Y203 , i.e., Ŷ200   ,   1,2 ,3 using the following values:

Y194=52 Y195=56 Y196=55 Y197=53 Y198=51 Y199=54

a194=0.05 a195=0.55 a196=0.15 a197=1.00 a198= -0.50 a199= 1.98

c) Find the  weights  j , j  1, 2,3 in the random shock form of the model.
d) Calculate the forecast error variances for e200   ,   1,2 ,3 and construct 95% prediction
intervals for Y201 ,Y202 and Y203 . .

34
3. Consider the monthly seasonally adjusted U.S. Disposable Personal Income Per Capita and M2
money stock data from 1959:1 to 2011:1.

a) What features of the data are evident in the time series plots and related statistics that should
be captured by a model? Is there other evidence that would be useful in formulating a
modeling strategy?

b) Interpret the below output. Explain the next step.

Dickey-Fuller Unit Root Tests

Variable Type Rho Pr < Rho Tau Pr < Tau

income Zero Mean 2.06 0.9902 11.76 0.9999

Single Mean 1.42 0.9943 5.30 0.9999
Trend -2.29 0.9623 -1.94 0.6346
m2 Zero Mean 2.84 0.9982 14.29 0.9999
Single Mean 2.58 0.9993 10.69 0.9999
Trend 2.11 0.9998 3.22 0.9999

c) We perform a Johansen cointegration test to these series and received the following results.
Explain what this output tells you.
Cointegration Rank Test Using Trace

5%
H0: H1: Critical Drift Drift in
Rank=r Rank>r Eigenvalue Trace Value in ECM Process

0 0 0.2086 148.6349 12.21 NOINT Constant

1 1 0.0042 2.6478 4.14

35
d) We decided to fit a VECM(4) to the series and obtain the following output on the residuals.
Interpret each output and if the assumptions are not satisfied, suggest a method to fix it.

Schematic Representation of Cross Correlations of Residuals

Variable/
Lag 0 1 2 3 4 5 6

income +- .. .. .. .. .+ ..
m2 -+ .. .. .. .- .. ..

+ is > 2*std error, - is <

-2*std error, . is between

Portmanteau Test for Cross Correlations of Residuals

Up To
Lag DF Chi-Square Pr > ChiSq

5 4 31.48 <.0001
6 8 35.03 <.0001

Univariate Model ANOVA Diagnostics

Standard
Variable R-Square Deviation F Value Pr > F

income 0.0969 138.52261 9.45 <.0001

m2 0.4582 12.84901 74.42 <.0001

Univariate Model White Noise Diagnostics

Durbin Normality ARCH

Variable Watson Chi-Square Pr > ChiSq F Value Pr > F

income 2.00646 9999.99 <.0001 15.04 0.0001

m2 1.94936 9999.99 <.0001 12.23 0.0005

Univariate Model AR Diagnostics

AR1 AR2 AR3 AR4

Variable F Value Pr > F F Value Pr > F F Value Pr > F F Value Pr > F

income 0.03 0.8728 0.01 0.9872 0.01 0.9978 0.01 0.9996

m2 0.39 0.5307 0.70 0.4978 0.54 0.6547 4.18 0.0024

e) What is weak exogeneity? Explain the below output.

Testing Weak Exogeneity of Each Variables

Variable DF Chi-Square Pr > ChiSq

income 1 59.82 <.0001
m2 1 76.38 <.0001

36
 4. U.S. quarterly Real Gross Domestic Product data are obtained from U.S. Department of
Commerce. Data start from first quarter of 1947 to first quarter of 2011.

a) What features of the data are evident in the time series plots and related statistics that should
be captured by a model? Is there other evidence that would be useful in formulating a
modeling strategy?

b) Interpret the below outputs and discuss what should be the next step.
HEGY.test (gdp, c(1,0,c(0)), regvar=0, selectlags=list(mode="signf", Pmax=NULL))

HEGY statistics:

Stat. p-value
tpi_1 -1.527 0.10
tpi_2 -8.078 0.01
Fpi_3:4 180.727 0.01
Fpi_2:4 734.935 NA
Fpi_1:4 558.200 NA

HEGY.test (diff(gdp), c(0,0,c(0)), regvar=0, selectlags=list(mode="signf", Pmax=NULL))

----
HEGY statistics:

Stat. p-value
tpi_1 -3.500 0.01
tpi_2 -6.759 0.01
Fpi_3:4 82.517 0.01
Fpi_2:4 85.489 NA
Fpi_1:4 66.266 NA

c) After a regular order of difference, we obtain the below time series plot, sample ACF and
sample PACF plots. Interpret the plots and suggest a tentative model.

37
d) You can see six different fitted models for the series below. Interpret the results and choose one
model as the best one for this time series. Explain the reasoning.
> fit1=arima(gdp,order=c(2,1,0),seasonal=list(order=c(2,0,0), period=4))
> fit1

Call:
arima(x = gdp, order = c(2, 1, 0), seasonal = list(order = c(2, 0, 0), period = 4))

Coefficients:
ar1 ar2 sar1 sar2
0.4842 0.3162 -0.0198 -0.1198
s.e. 0.0602 0.0677 0.0727 0.0679

sigma^2 estimated as 2669: log likelihood = -1373.57, aic = 2757.14

> fit2=arima(gdp,order=c(2,1,2),seasonal=list(order=c(2,0,0), period=4))

> fit2

Call:
arima(x = gdp, order = c(2, 1, 2), seasonal = list(order = c(2, 0, 0), period = 4))

Coefficients:
ar1 ar2 ma1 ma2 sar1 sar2
0.0605 0.7901 0.3714 -0.2949 -0.1256 -0.1659
s.e. 0.0955 0.1053 0.1310 0.1304 0.0780 0.0695

sigma^2 estimated as 2615: log likelihood = -1371.02, aic = 2756.04

> fit3=arima(gdp,order=c(2,1,2),seasonal=list(order=c(2,0,1), period=4))

> fit3

Call:
arima(x = gdp, order = c(2, 1, 2), seasonal = list(order = c(2, 0, 1), period = 4))

Coefficients:
ar1 ar2 ma1 ma2 sar1 sar2 sma1
0.0255 0.9043 0.3589 -0.4158 0.2824 -0.1659 -0.4933

38
s.e. 0.0518 0.0620 0.1061 0.1203 0.1340 0.0712 0.1318

sigma^2 estimated as 2549: log likelihood = -1367.92, aic = 2751.84

> fit4=arima(gdp,order=c(2,1,3),seasonal=list(order=c(2,0,1), period=4))

> fit4

Call:
arima(x = diff(gdp), order = c(2, 1, 3), seasonal = list(order = c(2, 0, 1),
period = 4))

Coefficients:
ar1 ar2 ma1 ma2 ma3 sar1 sar2 sma1
-0.3477 0.4668 -0.2431 -0.6768 0.0012 0.3553 -0.1937 -0.4686
s.e. 0.1543 0.1392 0.1687 0.1302 0.1335 0.1462 0.0701 0.1445

sigma^2 estimated as 2346: log likelihood = -1352.56, aic = 2721.11

> fit5=arima(gdp,order=c(2,1,2),seasonal=list(order=c(2,0,2), period=4))

> fit5

Call:
arima(x = gdp, order = c(2, 1, 2), seasonal = list(order = c(2, 0, 2), period = 4))

Coefficients:
ar1 ar2 ma1 ma2 sar1 sar2 sma1 sma2
0.0419 0.8538 0.3509 -0.3371 1.0561 -0.6825 -1.2874 0.7230
s.e. 0.0577 0.0663 0.1030 0.1070 0.1447 0.0990 0.1501 0.1153

sigma^2 estimated as 2441: log likelihood = -1362.87, aic = 2743.75

> fit6=arima(gdp,order=c(3,1,2),seasonal=list(order=c(2,0,2), period=4))

> fit6

Call:
arima(x = gdp, order = c(3, 1, 2), seasonal = list(order = c(2, 0, 2), period = 4))

Coefficients:
ar1 ar2 ar3 ma1 ma2 sar1 sar2 sma1 sma2
0.6857 0.7802 -0.4686 -0.2835 -0.6324 1.0094 -0.7214 -1.1548 0.6706
s.e. 0.1338 0.1304 0.0701 0.1374 0.1104 0.1388 0.1263 0.1455 0.1457

sigma^2 estimated as 2277: log likelihood = -1354.26, aic = 2726.53

e) After deciding on the best model, we checked the diagnostics of the model. Please interpret each
output. If assumptions are not satisfied, suggest a method to solve the problem.

39
> shapiro.test(window(rstandard(fit)))

Shapiro-Wilk normality test

data: window(rstandard(fit))
W = 0.975, p-value = 0.0001719

40
`
> Box.test(resid(fit),lag=15,type = c("Ljung-Box"))

Box-Ljung test

data: resid(fit)
X-squared = 10.0169, df = 15, p-value = 0.8187

Squared residuals are obtained and ACF and PACF plots are obtained as follows. Why are looking at
the squared residuals? What these graphs tell us?

41
f) We obtained the following forecasts and prediction limits. What can you say about these forecasts
and prediction limits? Are they plausible?

5. Consider two-dimensional vector AR(1) processes

Y1t   0.3  0.2 Y1,t 1  a1t 

Yt= Y    0.3     
 2t   0.9  Y2,t 1 a2t 

 X 1t   1 0  X 1,t 1  e1t 
 X   0.6  
0  X 2,t 1  e2t 
 2t  

a) Is the series I(1) or I(0)? Show.

b) Is Yt and/or Xt cointegrated? If so, find the cointegrating vector for the series.
c) Write the processes in error correction representation form if they are cointegrated.

6. Consider a quarterly data for U.S. from 1970:1 to 2008:4 (n=156) with variables

m1: log of real M1 money balance

42
 yt: log of real GDP (in billion of 2000 dollars)

tbrt: treasury bill rate

a) What is the problem when nonstationary series are used in regression analysis?
b) Why do we use augmented Dickey-Fuller test?
c) We perform a Granger-causality test. Analyze the results of Granger-causality Wald test at 5%
significance level.

Chi-square df p-value

m y 18.222 6 0.006

m tbr 91.873 6 0.000

y m 2.205 6 0.900

y tbr 26.29 6 0.000

tbr m 18.531 6 0.005

tbr y 16.902 6 0.010

FINAL EXAM 2011

QUESTIONS

1. A time series of n = 100 observations gave the following sample ACF and sample PACF
for the original data,
Lag k 1 2 3 4 5 6 7 8 9 10
rk 0.73 0.11 −0.50 −0.79 −0.63 −0.15 0.36 0.64 0.55 0.17
kk 0.73 −0.91 0.09 0.03 0.08 −0.05 −0.03 0.12 −0.14 −0.05

with the sample mean Y  0.07 and sample variance ˆ 0  5.29 and for the first differenced data
we have
Lag k 1 2 3 4 5 6 7 8 9 10
rk 0.65 −0.01 −0.61 −0.84 −0.59 −0.06 0.44 0.68 0.54 0.11
kk 0.65 −0.75 −0.42 −0.33 −0.15 −0.16 −0.24 0.00 −0.09 −0.13

with mean 0 and variance 2.89.

d) Identify the model from the given information.
e) Obtain the Yule-Walker estimates of the parameters. Also, obtain estimates for the error
variance.
f) Using the estimates as the true parameter values, calculate  k , k  1, 2,3, 4.

43
 2. Consider the following ARIMA model

1  1.8B  0.81B 2 Yt  0.2  at

where at ’s are WN with mean 0 and  a2  4 .
a) Find the mean of the process.
b) ind the variance of the process.
c) Obtain the first three-step ahead minimum MSE forecasts for Y51 ,Y52 , and Y53 ,
Ŷ50   ,   1,2 ,3 using the following values:

Y47=19 Y48=22 Y49=17 Y50=21

d) Find the  weights  j , j  1, 2,3 in the random shock form of the model.
e) Calculate the forecast error variances for e50   ,   1,2 ,3 and construct 95%
prediction intervals for Y51 ,Y52 and Y53 .
3. Consider the two-dimensional vector AR(1) process

Y1t  0   Y1,t 1  a1t 

Yt= Y   0     
 2t   1 Y2,t 1  a2t 

i. Is the series I(1) or I(0)? Show.

ii. Is Yt cointegrated? If so, find the cointegrating vector.
iii. Write the process in error correction representation form.

44
4. Lydia Pinkham’s Vegetable Compound was first brewed and prepared by the lady of the
same name as a home remedy of her friends. The compound, an herbal extract in alcoholic
solution, was considered to be effective against “women’s weakness”. The first
commercial sale of the medicine was in 1873, when sales reached $500,000. A part of the
profits was reinvested immediately into high intensity advertising campaigns. Records of
a court litigation involving a family argument produced a complete listing of the firm’s
annual advertising expenditures and sales over the period 1908-1935. Palda (1964) was
later able to obtain the data from 1907-1960.

i. Interpret the below time series plot.

Time series plot of Lydia Pinkham’s Advertising and Sales Data from 1907-1960
4000

3000

Sales
2000

Advertisement
1000

1900 1910 1920 1930 1940 1950 1960

ii. What is the purpose of the Granger-Causality test? Explain. Interpret the below output.
Are the results logical? Explain.
Granger-Causality Wald Test

Test DF Chi-Square Pr > ChiSq

1 1 16.91 <.0001
2 1 8.73 0.0031

Test 1: Group 1 Variables: Advertising

Group 2 Variables: Sales

Test 2: Group 1 Variables: Sales

Group 2 Variables: Advertising

45
5. Consider the number of visitors (in thousands) to a Colorado ski resort data between the
first quarter of 1981 to last quarter of 1986.
i. Interpret the below time series plot.

ii. Considering the below sample ACF, sample PACF plots and ADF test results what
should we do next? Why? Explain.

Augmented Dickey-Fuller Unit Root Tests

Type Lags Tau Pr < Tau

Zero Mean 1 5.02 0.9999

4 0.67 0.8508
Single Mean 1 0.01 0.9500
4 -0.44 0.8839
Trend 1 -2.96 0.1638
4 -2.14 0.4955

46
c) After regular and seasonal differencing, we fit a certain model to the data. The results are given in
the below table:

R output

Coefficients:
ar1 sar1
-0.7166 -0.6931
s.e. 0.1526 0.1645

sigma^2 estimated as 2.298: log likelihood = -36.36, aic = 78.73

Which model is fitted to the data? Are all parameters in the model significant? Write the model.

d) The diagnostic analysis on the residuals are given below. Explain each graph and test result. Do
you think that the model that you found in part c) is a plausible model for the data? If not, what do
you suggest?

Autocorrelation Check of Residuals

To Chi- Pr >
Lag Square DF ChiSq -------------Autocorrelations------------

6 2.72 4 0.6065 -0.150 0.014 0.115 -0.097 0.090 -0.214

12 5.59 10 0.8486 -0.038 -0.138 -0.050 -0.009 -0.200 0.028
18 10.37 16 0.8463 0.214 -0.026 0.108 0.061 0.010 0.003

Breusch-Godfrey test for serial correlation of order 1

data: resid(ski.fit) ~ ski

LM test = 1.6141, df = 1, p-value = 0.2039

47
 Jarque Bera Test
data: resid(ski.fit)
X-squared = 10.897, df = 2, p-value = 0.004303

The Sample ACF, Sample PACF plots and “Autocorrelation Check for White Noise” of
Squared Residuals
Autocorrelations

Lag Covariance Correlation -1 9 8 7 6 5 4 3 2 1 0 1 2 3 4 5 6 7 8 9 1

0 17.032300 1.00000 | |********************|

1 5.793277 0.34013 | . |******* . |
2 -2.034588 -.11945 | . **| . |
3 -4.166656 -.24463 | . *****| . |
4 -3.092140 -.18155 | . ****| . |
5 1.312139 0.07704 | . |** . |
6 0.451033 0.02648 | . |* . |
7 -0.604024 -.03546 | . *| . |
8 -1.900935 -.11161 | . **| . |
9 -2.330850 -.13685 | . ***| . |
10 -2.459929 -.14443 | . ***| . |
11 1.310643 0.07695 | . |** . |
12 1.717580 0.10084 | . |** . |
13 -0.937858 -.05506 | . *| . |
14 -1.515496 -.08898 | . **| . |
15 -0.824299 -.04840 | . *| . |
16 0.249750 0.01466 | . | . |

"." marks two standard errors

48
 Partial Autocorrelations

Lag Correlation -1 9 8 7 6 5 4 3 2 1 0 1 2 3 4 5 6 7 8 9 1

1 0.34013 | . |******* . |
2 -0.26591 | . *****| . |
3 -0.12504 | . ***| . |
4 -0.08301 | . **| . |
5 0.13486 | . |*** . |
6 -0.15290 | . ***| . |
7 -0.00885 | . | . |
8 -0.10794 | . **| . |
9 -0.06764 | . *| . |
10 -0.18666 | . ****| . |
11 0.17740 | . |**** . |
12 -0.12110 | . **| . |
13 -0.11265 | . **| . |
14 -0.05091 | . *| . |
15 0.05536 | . |* . |
16 -0.14527 | . ***| . |

Autocorrelation Check for White Noise

To Chi- Pr >
Lag Square DF ChiSq -------------Autocorrelations------------

6 5.46 6 0.4864 0.340 -0.119 -0.245 -0.182 0.077 0.02

12 8.50 12 0.7450 -0.035 -0.112 -0.137 -0.144 0.077 0.10

Studentized Breusch-Pagan test

data: resid(ski.fit) ~ ski
BP = 2.8486, df = 1, p-value = 0.09145

6. State whether the following statements are TRUE (T) or FALSE (F)? And if your answer
is FALSE, write the TRUE statement below it.

( ) is seasonal stationary process.

( ) If the seasonal difference (i.e., the season-to-season change) of a time series looks like
stationary noise, this suggests that the process is seasonal random walk.

( ) Cointegration refers to a case in which a vector process has individually nonstationary

components and there is some linear combination of them that is also nonstationary.

( ) The idea behind cointegration is to find a linear combination between two I(d)-variables that
yields a variable with a lower order of integration.

49
( ) White's (1980) test is a test of the null hypothesis of no heteroskedasticity against
heteroskedasticity of unknown, general form.

( ) The correlograms of the squared residuals can be used to check autoregressive conditional
heteroskedasticity. If there is ARCH in the residuals, the autocorrelations and partial autocorrelations
should be zero at all lags and the Ljung-Box Q-statistics should not be significant.

( ) The ARCH model does not provide any new insight for understanding the source of variations of
a financial time series. They only provide a mechanical way to describe the behavior of the
conditional variance. It gives no indication about what cause such behavior to occur.

( ) If we decide a spurious regression between any two variable, it means that there is cointegration
between these variables.

( ) Vector error correction model (VECM) is adding error correction feature to a multi-factor model
such as a vector autoregression model.

( ) A time series X is said to be Granger-Cause Y if the R-square of regression estimation is biger

than Durbin Watson test statistics.

FINAL EXAM 2012

QUESTIONS (Each question is 25 pts.)

1. From a series of length 100, we have computed the sample autocorrelations as r1 = 0.8, r2 = 0.5,
r3 = 0.4, Y = 2, and a sample variance of 5. If we assume that an AR(2) model with a constant
term is appropriate,
a) Obtain the Yule-Walker estimates of 1, 2, θ0, and 2, where 0 is the constant term and 2 is the
error variance?
b) Using the estimates of the parameters that you found in part a), obtain the two-step ahead forecasts
with the 95% prediction intervals assuming that Y99=5 and Y100=3.

2. Consider two-dimensional vector AR(1) processes

Y1t   0.7  0.4 Y1,t 1  a1t 

a)         
Y2t   0.3 0.5  Y2,t 1  a 2t 

50
  X1,t   0.4  0.6 X1,t 1  a1,t   0.8  0.3a1,t 1 
b)            
 X 2,t   0.2 0.3  X 2,t 1 a2,t   0.4 0.15 a2,t 1 

i. Is the series I(1) or I(0)? Show.

ii. Is Yt and/or Xt cointegrated? If so, find the cointegrating vector for the series.
iii. Write the processes in error correction representation form if they are cointegrated.

3. Quarterly oil prices of Turkey data are obtained from TURKSTAT from 1996 to 2011.

c) Interpret the below time series plot.

d) Interpret the following outputs separately and discuss about the next step to follow.
> HEGY.test(wts =oil, itsd = c(1, 0,c(0)), regvar = 0)
HEGY statistics:

Stat. p-value
tpi_1 0.304 0.10
tpi_2 -5.407 0.01
Fpi_3:4 7.807 0.01
Fpi_2:4 28.932 NA
Fpi_1:4 21.747 NA

> HEGY.test(wts =diff(oil), itsd = c(0, 0,c(0)), regvar = 0)

HEGY statistics:

Stat. p-value
tpi_1 0.001 0.10
tpi_2 -3.320 0.01
Fpi_3:4 1.561 0.10
Fpi_2:4 7.740 NA
Fpi_1:4 6.343 NA

> HEGY.test(wts =diff(diff(oil),4), itsd = c(0, 0,c(0)), regvar = 0)

HEGY statistics:

51
 Stat. p-value
tpi_1 -8.288 0.010
tpi_2 -7.219 0.011
Fpi_3:4 34.417 0.010
Fpi_2:4 36.234 NA
Fpi_1:4 30.594 NA

c) You can see five different fitted models for the series below. Interpret the results and choose one model
as the best one for this time series. Explain the reasoning.

> fit1=arima(oil,order=c(2,1,1),seasonal=list(order=c(2,1,0), period=4))

Coefficients:
ar1 ar2 ma1 sar1 sar2
1.1859 -0.3353 -0.9564 -0.6745 -0.3030
s.e. 0.1439 0.1410 0.1160 0.1554 0.1353

sigma^2 estimated as 0.0139: log likelihood = 41.41, aic = -72.81

> fit2=arima(oil,order=c(2,1,1),seasonal=list(order=c(2,1,1), period=4))

Coefficients:
ar1 ar2 ma1 sar1 sar2 sma1
0.8429 -0.3820 -0.5137 0.0668 0.0314 -0.8380
s.e. 0.3683 0.1572 0.3800 0.2213 0.1682 0.1647

sigma^2 estimated as 0.01266: log likelihood = 43.44, aic = -74.88

> fit3=arima(oil,order=c(2,1,1),seasonal=list(order=c(0,1,2), period=4))

Coefficients:
ar1 ar2 ma1 sma1 sma2

52
 0.8374 -0.3823 -0.5056 -0.7709 -0.0389
s.e. 0.3674 0.1589 0.3766 0.1758 0.1633

sigma^2 estimated as 0.01268: log likelihood = 43.42, aic = -76.84

> fit4=arima(oil,order=c(2,1,1),seasonal=list(order=c(0,1,1), period=4))

Coefficients:
ar1 ar2 ma1 sma1
0.7863 -0.3587 -0.4590 -0.8030
s.e. 0.3813 0.1362 0.4063 0.1186

sigma^2 estimated as 0.01272: log likelihood = 43.39, aic = -78.79

> fit5=arima(oil,order=c(2,1,0),seasonal=list(order=c(0,1,1), period=4))

ar1 ar2 sma1
0.3642 -0.2523 -0.8489
s.e. 0.1238 0.1262 0.1096

sigma^2 estimated as 0.0127: log likelihood = 43.21, aic = -80.41

d) After deciding on the best model, we checked the diagnostics of the model. Please interpret each
output one-by-one. If assumptions are not satisfied, suggest a method to solve the problem.

> shapiro.test(window(rstandard(fit),start=c(1996,1)))
Shapiro-Wilk normality test

53
W = 0.9374, p-value = 0.002623

> jarque.bera.test(rstandard(fit))
Jarque Bera Test

X-squared = 11.3494, df = 2, p-value = 0.003432

> Box.test(rstandard(fit),lag=15,type = c("Ljung-Box"))

Box-Ljung test

data: rstandard(fit)
X-squared = 9.9355, df = 15, p-value = 0.8238

> Box.test(rstandard(fit),lag=15,type = c("Box-Pierce"))

Box-Pierce test

data: rstandard(fit)
X-squared = 8.4557, df = 15, p-value = 0.9042

Squared residuals are obtained and ACF and PACF plots are obtained as follows. Why are looking at
the squared residuals? What these graphs tell us?
>rr=rstandard(fit)^2

54
d) We obtained the following forecasts and prediction limits. What can you say about these forecasts
and prediction limits? Are they plausible?

4. Consider the quarterly Turkish private consumption expenditure and current price gross domestic
product (GDP) series from 1990 to 2011.

a) Interpret the below plot. What could be the reason of the co-movement between these two series?
Explain.

55
b) What is the purpose of the Granger causality test? Explain the logic behind the test.

c) Analyze and interpret the results of Granger causality test applied to the consumption and GDP series
of Turkey. Do you think that the results are logical? Discuss.

> granger.test(cg, 3)
F-statistic p-value
GDP -> CONSUMPTION 2.331032 0.080678470
CONSUMPTION -> GDP 5.249620 0.002368946

MIDTERM EXAM SPRING 2013

QUESTIONS

8. Consider the following process

Yt  6  0.8Yt  2  at  0.027at  3

where it is assumed that {at}’s are White Noise with zero mean and  a2 variance.

vii) (5 pts.) Write the process in backshift operator form.

viii) (10 pts.) Is the process stationary and/or invertible? Show. If the process is not
stationary, obtain the stationary process using an appropriate method.
ix)(10 pts.) Find  , 0 , 1 and  2 .
9. Consider the multiplicative seasonal ARIMA model of order (1,0,0)x(2,0,1)4

c) (5 pts.) Write the stationarity and the invertibility condition of the process.
d) (5 pts.) Write the process in random shock form and find 1, 2, 3 and 4.
e) (5 pts.) Write the process in inverted form and find 1, 2, 3 and 4.

56
 10. Consider the following processes
a) (5 pts.) Yt  0.8Yt 1  0.9Yt  2  0.7Yt  3  at
b) (5 pts.) Yt  0.4Yt 12  0.6Yt  24  at  0.5at  2
c) (5 pts) Yt  2  t  Yt 1  at  at 1

where at ~ WN 0, 2 . 
Are the processes stationary? If not, obtain stationary processes applying an appropriate
technique.

11. (15 pts.) Examining and forecasting the global carbon emissions is very important to get
precautions before it is too late. Changing climate due to global warming and raise in carbon
emission affect all living creatures on earth. Source carbon data is published online
(November 21, 2010) in Nature Geoscience and released via GlobalCarbonProject.org. We
examine yearly global carbon emissions from 1959 to 2009.

a) What features of the data are evident in the time series plots and related statistics that should
be captured by a model? Is there other evidence that would be useful in formulating a
modeling strategy?

d) In the light of following test results, what are the necessary steps that we need to consider
next?

> pp.test(co2)

data: co2
Dickey-Fuller Z(alpha) = -10.6185, Truncation lag parameter = 3, p-value = 0.4778

> kpss.test(co2,c("Level"))

57
data: co2
KPSS Level = 2.5488, Truncation lag parameter = 1, p-value = 0.01

> kpss.test(co2,c("Trend"))

data: co2
KPSS Trend = 0.279, Truncation lag parameter = 1, p-value = 0.01

e) Interpret the below output.

> pp.test(diff(co2))
data: diff(co2)
Dickey-Fuller Z(alpha) = -38.2275, Truncation lag parameter = 3, p-value = 0.01

> kpss.test(diff(co2),c("Level"))
data: diff(co2)
KPSS Level = 0.0947, Truncation lag parameter = 1, p-value = 0.1

d) After having a stationary process, the time series plots and ACF/PACF plots of the series given
below. Interpret these plots. Identify an appropriate ARIMA(p,d,q) model for this series.

b) (15 pts.) Turkish monthly consumer prices index (CPI) is obtained from January 1995 to
April 2013.

a) Interpret the following plots.

58
 b) Interpret the below outputs and discuss what should be the next step. Explain each output
separately.

> HEGY.test(cpi, itsd=c(1,0,0),regvar=0, selectlags=list(mode="aic",

Pmax=12))

HEGY statistics:

Stat. p-value
tpi_1 -0.273 0.100
tpi_2 -4.226 0.010
Fpi_3:4 8.162 0.010
Fpi_5:6 4.491 0.012
Fpi_7:8 3.203 0.043
Fpi_9:10 14.468 0.010
Fpi_11:12 1.144 0.100
Fpi_2:12 7.660 NA
Fpi_1:12 7.029 NA

> HEGY.test(diff(cpi), itsd=c(0,0,0),regvar=0, selectlags=list(mode="aic",

Pmax=12))

HEGY statistics:
Stat. p-value
tpi_1 0.343 0.100
tpi_2 -3.004 0.010
Fpi_3:4 4.526 0.011
Fpi_5:6 2.787 0.068
Fpi_7:8 0.832 0.100
Fpi_9:10 6.168 0.010
Fpi_11:12 4.085 0.019
Fpi_2:12 3.129 NA
Fpi_1:12 2.906 NA

59
>HEGY.test(diff(cpi,12), itsd=c(0,0,0),regvar=0, selectlags=list(mode="aic",
Pmax=12))

HEGY statistics:
Stat. p-value
tpi_1 -0.413 0.10
tpi_2 -6.206 0.01
Fpi_3:4 17.810 0.01
Fpi_5:6 16.373 0.01
Fpi_7:8 24.747 0.01
Fpi_9:10 35.998 0.01
Fpi_11:12 21.145 0.01
Fpi_2:12 40.590 NA
Fpi_1:12 37.223 NA
> HEGY.test(diff(diff(cpi,12)), itsd=c(0,0,0),regvar=0,
selectlags=list(mode="aic", Pmax=12))

HEGY statistics:
Stat. p-value
tpi_1 -3.446 0.01
tpi_2 -10.117 0.01
Fpi_3:4 49.689 0.01
Fpi_5:6 35.239 0.01
Fpi_7:8 37.765 0.01
Fpi_9:10 43.328 0.01
Fpi_11:12 32.261 0.01
Fpi_2:12 30.757 NA
Fpi_1:12 28.998 NA

c) As a result of the operation that you suggest in part b), the series Zt is obtained and the time
series plot, correlogram and sample PACF plot are given below.
i) Interpret the time series plot and discuss the existence of any strange behavior.
ii) Which ARIMA(p,d,q) or SARIMA(p,d,q)(P,D,Q)s process(es) is(are) suitable for this
series? Why?

60
c) (15 pts.) Suggest appropriate processes for the below plots and explain the reasons.
i)

ii)

61
 iii)

FINAL EXAM SPRING 2013

QUESTIONS (Each question is 20 pts.)

3. From a series of length 200, the sample mean of the process is obtained as 25. The sample
autocorrelations are calculated as r1 = 0.76, r2 = 0.65, r3 = 0.14, and a sample variance of 4. If we
assume that an AR(2) model with a constant term is appropriate,
c) Obtain the Yule-Walker estimates of the constant term, parameters 1, 2 and the error variance
2?
d) Using the estimates of the parameters that you found in part a), obtain the two-step ahead forecasts
with the 95% prediction intervals assuming that Y199=27 and Y200=24.
 
4. Consider the process 1  0.7 B  1  0.3B 4 Yt  12  1  0.4 B  at where at ’s are WN with
mean 0 and  a2  1 .

62
 e) If the forecast origin 90, given is
Y85  9 ,Y86  20, Y87  8,Y88  15, Y89  10, a 90  0.5 and a89  0.3 , find the first four-
step ahead minimum MSE forecasts for Y91 ,Y92 , Y93 and Y94 , Ŷ90   ,   1,2 ,3,4 .
f) Calculate the forecast error variances for e90   ,   1,2 ,3,4 and construct 95%
prediction intervals for Y91 ,Y92 , Y93 and Y94 .
g) Suppose that the observation at t = 91 turns out to be Y91  16 . Find the updated
forecast for Y92 .
5. Consider two-dimensional vector AR(1) processes

Y1,t  1.0 0.8 Y1,t 1  a1,t   0.8 0.4a1,t 1 

a)         
Y2,t  0.3 0.24Y2,t 1 a2.t   0.,3 0.2a2,t 1

 X1,t  0.5  0.6 X1,t 1  a1,t 

b)       
 X 2,t  0.0 0.2  X 2,t 1  a2,t 

i. Is the series I(1) or I(0)? Show.

ii. Is Yt and/or Xt cointegrated? If so, find the cointegrating vector for the series.
iii. Write the processes in error correction representation form, if they are cointegrated.

63
3. Monthly median sales for new houses sold data are obtained from U.S. Department of Commerce:
Census Bureau between January 1963 and April 2013. The last four months’ data which will be used to
check the forecast accuracy were not used in the analysis.

e) Interpret the below time series plot.

f) Interpret the following outputs separately and discuss about the next step to follow.
> HEGY.test(wts =sales, itsd= c(1, 0,c(0)), regvar = 0)

HEGY statistics:

Stat. p-value
tpi_1 0.873 0.10
tpi_2 -5.535 0.01
Fpi_3:4 71.608 0.01
Fpi_5:6 52.256 0.01
Fpi_7:8 66.231 0.01
Fpi_9:10 65.960 0.01
Fpi_11:12 34.894 0.01
Fpi_2:12 71.147 NA
Fpi_1:12 65.527 NA

> HEGY.test(wts =diff(sales), itsd = c(0, 0,c(0)), regvar = 0)

HEGY statistics:

Stat. p-value
tpi_1 -2.542 0.01
tpi_2 -5.501 0.01
Fpi_3:4 52.080 0.01
Fpi_5:6 37.911 0.01
Fpi_7:8 37.033 0.01
Fpi_9:10 55.274 0.01
Fpi_11:12 25.054 0.01
Fpi_2:12 54.296 NA
Fpi_1:12 50.623 NA

64
> HEGY.test(wts =diff(diff(sales,12)), itsd = c(0, 0,c(0)), regvar = 0)

HEGY statistics:

Stat. p-value
tpi_1 -6.698 0.01
tpi_2 -6.404 0.01
Fpi_3:4 83.780 0.01
Fpi_5:6 61.907 0.01
Fpi_7:8 55.682 0.01
Fpi_9:10 82.737 0.01
Fpi_11:12 81.847 0.01
Fpi_2:12 50.644 NA
Fpi_1:12 49.703 NA

65
c) You can see six different fitted models for the series below. Interpret the results and choose one model
as the best one for this time series. Explain the reasoning.

> fit1=arima(sales,order=c(3,1,1),seasonal=list(order=c(2,0,0), period=12))

Coefficients:
ar1 ar2 ar3 ma1 sar1 sar2
-0.2212 -0.0190 -0.0658 -0.4019 -0.0471 0.1448
s.e. 0.1539 0.1015 0.0536 0.1509 0.0434 0.0435

sigma^2 estimated as 22384855: log likelihood = -5919.13, aic = 11850.27

> fit2=arima(sales,order=c(1,1,1),seasonal=list(order=c(0,1,1), period=12))

Coefficients:
ar1 ma1 sma1
-0.1243 -0.5415 -0.9691
s.e. 0.0631 0.0521 0.0271

sigma^2 estimated as 22490991: log likelihood = -5818.25, aic = 11642.49

> fit3=arima(sales,order=c(0,1,2),seasonal=list(order=c(2,1,1), period=12))

Coefficients:
ma1 ma2 sar1 sar2 sma1
-0.6519 0.0714 -0.0770 0.0957 -0.9710
s.e. 0.0412 0.0368 0.0459 0.0472 0.0309

sigma^2 estimated as 22168442: log likelihood = -5814.29, aic = 11638.58

> fit4=arima(sales,order=c(1,1,2),seasonal=list(order=c(2,0,1), period=12))

Coefficients:
ar1 ma1 ma2 sar1 sar2 sma1
-0.8483 0.2476 -0.4076 -0.0498 0.1402 -0.0118
s.e. 0.1043 0.1221 0.0791 0.3072 0.0467 0.3070

sigma^2 estimated as 22346837: log likelihood = -5918.62, aic = 11849.24

> fit5=arima(sales,order=c(5,1,1),seasonal=list(order=c(3,1,2), period=12))

Coefficients:
ar1 ar2 ar3 ar4 ar5 ma1 sar1 sar2
-1.1404 -0.6582 -0.3967 -0.2722 -0.1706 0.4854 0.6797 0.1488
s.e. 0.1915 0.1340 0.0861 0.0683 0.0411 0.1932 0.1358 0.0558
sar3 sma1 sma2
-0.1688 -1.7374 0.7629
s.e. 0.0533 0.1297 0.1236

sigma^2 estimated as 21599782: log likelihood = -5804.66, aic = 11631.31

> fit6=arima(sales,order=c(1,1,1),seasonal=list(order=c(2,0,0), period=12))

Coefficients:

66
 ar1 ma1 sar1 sar2
-0.1800 -0.4477 -0.0479 0.1430
s.e. 0.0696 0.0648 0.0434 0.0435

sigma^2 estimated as 22464452: log likelihood = -5920.19, aic = 11848.37

d) Below are the diagnostics of fit6. Please interpret each output one-by-one. If assumptions are not
satisfied, suggest a method to solve the problem.

> jarque.bera.test(rstandard(fit))

Jarque Bera Test

data: rstandard(fit)
X-squared = 402.8255, df = 2, p-value < 2.2e-16

> Box.test(rstandard(fit),lag=15,type = c("Ljung-Box"))

Box-Ljung test

data: rstandard(fit)
X-squared = 33.7756, df = 15, p-value = 0.003661

Squared residuals are obtained and then ACF and PACF plots are drawn as follows. Why are we
looking at the squared residuals? What do these graphs tell us?
>rr=rstandard(fit)^2

67
d) We obtained the following forecasts and prediction limits for fit6 and fit5, respectively. What can
you say about these forecasts and prediction limits? Are they plausible?

68
e) The original median new house sale prices, forecasts of fit5 and fit6 are given below. Use one of the
forecast accuracy measures and decide which fit is better than the other using this measure.

    FORECASTS
  Original Fit6 Fit5
Jan-13 251500 252853.9 254471.3
Feb-13 266300 250047.2 256708
Mar-13 250700 249942.7 253329
Apr-13 271600 250736.1 257136.6

5. Annual Carbon Dioxide Emissions (million tons carbon dioxide), coal consumption (million tons
oil equivalent) and primary energy consumption (million tons oil equivalent) of Turkey are
obtained by BP Statistical Review of World Energy June 2012 from 1965 to 2011.

a) Interpret the below time series plot.

69
 400

300

200

100

1960 1970 1980 1990 2000 2010 2020

Year

Note: Upper line is the graph of emission, middle one is for the general energy consumption and
the bottom one is for the coal consumption.

b) Explain the following test results. What is the next step to follow? Explain.
Cointegration Rank Test Using Trace

5%
H0: H1: Critical Drift Drift in
Rank=r Rank>r Eigenvalue Trace Value in ECM Process

0 0 0.5144 40.5992 24.08 NOINT Constant

1 1 0.1170 7.3657 12.21
2 2 0.0350 1.6407 4.14

70
c) Analyze and interpret the results of Granger causality test applied to the general energy consumption,
coal consumption and carbon dioxide emission series of Turkey. Do you think that the results are logical?
Discuss.
Granger-Causality Wald Test

Test DF Chi-Square Pr > ChiSq

1 1 0.15 0.7019
2 1 1.26 0.2610
3 1 5.36 0.0207

Test 1: Group 1 Variables: consumption

Group 2 Variables: emission

Test 2: Group 1 Variables: emission

Group 2 Variables: consumption

Test 3: Group 1 Variables: coal

Group 2 Variables: emission

71