Anda di halaman 1dari 4

EE 627: Term 2/2016

Homework 4
Due April 25, 2017
1. The ARMA(2,1) process given by
[ ]
1 + 0.2L 0.24L2 yt = [1 0.8L] t

where t is a white noise process and as such satises the properties of E(t ) = 0, E(2t ) = 2 and
E(t s ) = 0 for s = t.
(a) Is the process stationary? Why?
Solution: The autoregressive polynomial can be factorized as,

(1 + .6L)(1 .4L)yt = (1 .8L)t

and since the AR and MA polynomials do not share any common factors, it follows that the
process is identied. The roots of (L) are 5/3 and 5/2, which both lie outside the unit circle,
so the process is stationary. (or we can consider the ||, which are .6 and .4, are less than 1.)

(b) Find the variance of yt and also nd a general expression for the autocorrelation function k for
k > 2 and compute the numerical value of these autocorrelations for k = 0, 1, 2, 3 [Note that
E(t ytj ) = 0 for j > 0.]

Solution: Let wk = E(yt tk ), and on successively multiplying through by t , t1 , ... and on


taking expectations,

E(yt t ) = E(.2t yt1 + .24t yt2 + 2t .8t t1 ) = 2

w1 = E(yt t1 ) = .2w0 + .24w1 .8 2 = 2

The variance and autocovariance can be solved from following equations;

0 = E(yt yt ) = .21 + .242 + w0 .8w1


1 = E(yt yt1 ) = .20 + .241 .8w0 ,
2 = E(yt yt2 ) = .21 + .240 ,
k = E(yt ytk ) = .2k1 + .24k2 , for k > 2

Solving these equations gives; 0 = 2.34982 , 1 = 1.6712 , and 2 = .89812 .


Hence 0 = 1, 1 = .7111, 2 = .3822 and on using the recursion that

k = .2k1 + .24k2

and hence
k = 1.11(3/5)k .11(2/5)k .

1
(c) Find a general expression for the innite moving average representation coefcients


yt = i ti
i=0

and calculate the i for i = 0, 1, 2, 3.



Solution: Since yt = k tk = (L)t , pluging this into equation (L)yt = (L)t , then

(L)(L)t = (L)t

or (L)(L) = (L). We have

(1 + .2L .24L2 )(1 + 1 L + 2 L2 + 3 L3 + ...) = (1 .8L)

(1 + .2L .24L2 + 1 L + .21 L2 .241 L3 + 2 L2 + .22 L3 .242 L4 +

3 L3 + .23 L4 .243 L5 + ...) = (1 .8L)


We can following equations;

1 + .2 = .8 1 = 1

.24 + .2 + 2 = 0 2 = .44

In general case, k + .2k1 .24k2 = 0. Therefore, 3 = .33.

2. Answer following questions using the Thailand export monthly growth rate (x) between February 1992
and September 2014 in le growth_gr.dta. Data is available at http://www.gg.gg/ee627data
(a) Is x a white noise series?

Solution:
. wntestq x

Portmanteau test for white noise


---------------------------------------
Portmanteau (Q) statistic = 485.6789
Prob > chi2(40) = 0.0000

From the p-value of L-B Q test, we can reject the Null hypothesis that x is White noise. Therefore,
we can construct the time series model for x.

(b) Find the suitable ARMA(p,q) model for x using Schwarz Information Criteria (SIC) with maximum p and
q equal 6.

Solution: The model with minimum SIC(BIC) is ARM A(4, 3). I will estimate this model
and used it for answering following question. The result from Stata are as follow.
. arima x, ar(1/4) ma(1/3)

2
ARIMA regression

Sample: 1 - 272 Number of obs = 272


Wald chi2(6) = 18900.06
Log likelihood = 334.8141 Prob > chi2 = 0.0000

------------------------------------------------------------------------------
| OPG
x | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
x |
_cons | .0076024 .0005146 14.77 0.000 .0065938 .008611
-------------+----------------------------------------------------------------
ARMA |
ar |
L1. | -.4118614 .0617872 -6.67 0.000 -.5329621 -.2907608
L2. | -.0832095 .0335921 -2.48 0.013 -.1490488 -.0173702
L3. | .904379 .033125 27.30 0.000 .8394552 .9693027
L4. | .3146417 .0626233 5.02 0.000 .1919023 .4373812
|
ma |
L1. | -.0680051 .0426499 -1.59 0.111 -.1515974 .0155872
L2. | .0318831 .0466553 0.68 0.494 -.0595595 .1233258
L3. | -.9638783 . . . . .
-------------+----------------------------------------------------------------
/sigma | .0698746 .0029405 23.76 0.000 .0641112 .0756379
------------------------------------------------------------------------------
Note: The test of the variance against zero is one sided, and the two-sided confidence
interval is truncated at zero.

(c) Is the model(b) adequate?

Solution: We generate the residuals by using the `predict' command and test a model adequacy
with the white noise test.
. predict e, residuals

. wntestq e, lags(10)

Portmanteau test for white noise


---------------------------------------
Portmanteau (Q) statistic = 9.1341
Prob > chi2(10) = 0.5194

We compare L-B Q statistics =9.13 with critical chi-sq (df=10-3-4=3)=7.8. We found that Q statistic
> critical value, we can reject the null hypothesis that residuals is white noise and conclude that
the model is not adequate to explain the data.

(d) Using the model from (b) to forecast an export growth 1 month-ahead.

Solution: We add more observations by using tsappend. Then we can forecast x from following
command.
. tsappend, add(10)

. predict xf if t>=272
(option xb assumed; predicted values)
(271 missing values generated)

3
. list t xf if t>=273

+-----------------+
| t xf |
|-----------------|
273. | 273 -.0113607 |
274. | 274 .006384 |
275. | 275 .018527 |
276. | 276 -.0006369 |
277. | 277 .0030183 |
|-----------------|
278. | 278 .0196726 |
279. | 279 -.0010016 |
280. | 280 .0034035 |
281. | 281 .0195214 |
282. | 282 -.0009406 |
+-----------------+

Export growth in October is .01136.

Anda mungkin juga menyukai