Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Time Series Assignment 1

Saman Darougheh

1
1.a) R14.1
The GDP has positive but diminishing growth, hence it is not stationary. Since the growth rate is positive but also diminishing, it is not stationary neither.

1.b)

14.1

yt =
i=0

i (1 Li )(0 + ut )

yt1 =
i=0

i (1 Li1 )(0 + ut1 )

E[yt ] = E[
i=0

i 1 0 ] =

0 1 1 0 1 1

E[yt1 ] = E[
i=0

i 1 0 ] =

E[yt1 ] = E[yt ]

1.c)
i)

E14.1

The mean of the growth rate is 0.0084242, i.e. 0.8%. ii) 0.8 400 = 3.2, which is the annual growth rate of real gdp in percentage points.

iii) The std. deviation is 0.0099799 per month, that is 3.99%. iv) Correlations come along without units; it says how much of any growth-value of qgdp can be described by previous realizations of quarterly gdp growth.

1.d)
i)

E14.2

AR(1)

The estimated coecient is 0.332731. Somehow GRETL does not report the p-value for the autoregressive part, no matter which test I use. Anyhow, the ESS is 0.0202797. ii) AR(2)

The stimated coecients are 0.303122 (0.0000) and 0.081 (0.2183); p-values in parentheses. Hence, the AR(2) coecient is not signicantly dierent than 0 at any standard level of signicance. I would still prefer this to the AR(1) model, since the AR(2)-component takes over eects that belong to it and that we would have falsely attributed to the AR(1) component otherwise. iii) AR+

Uh, we didnt do BIC and AIC yet... I guess this exercise was not meant to be done by now.

2
For me, the most important part was to have previews on the results, instead of writing the paper as a story. Most of the other notes were known to me already, but the whole idea of write the results shortly, as soon as possible and dont repeat them (too often) was quite important, especially because it was not really what I was taught by my Bachelor Thesis advisor. Well, maybe these rules do not apply fully to thesis (that are not written in paper-form).

3
3.a)
The must be smaller than one, otherwise the polynomial expression of the inverse will explode as we sum up to innite.

3.b)
(L) = (1 (L)) (1 (L))(1 + 1 L + 2 L2 ...) = 1 i = Li C(L) = (1 L)
1

= (1 +
i=1

i Li )

3.c)

C(L) (L) = (1 (L)) (1 +


i=1

i Li ) = 1

0=
i=1

i Li
i=1

i+1 Li+1 L

0=0

4
A star indicates a mean-deviated process.

4.a)
zt =
t

+ 2ut

E[zt ] = 0
E[zt zt ] = 1 + 4 = 5 E[zt zt1 ] = 0

The ACF of zt is 5 for h = 0, 0 otherwise; it is covariance-stationary. This indicates a white noise process.

4.b)
xt = 2 + E[xt ] = 2 E[x x ] = 1 + 4 = 5 t t E[x x ] = 4 t t1 E[x x ] = 0 t t2
t

+2

t1

The ACF of xt is 5 for h = 0, 4 for h = 1 and 0 otherwise; it is covariance-stationary. This indicates an MA(1) process.

4.c)
yt = 3 + 0.5yt1 +
t

yt = (1 +
i=0

0.5i Li )(3 + t )

3 E[yt ] = =6 1 0.5
E[yt yt ] = E[ t1 i=0

0.5i

t1 i=0

0.5i ]

1 = E[ 1 0.52 4 = 3
E[yt yt1 ]

t1 t1 ]

i i t2 i=0

= E[ = E[

t1 i=0

(0.5 L )
t1i )

(0.5i Li )]
t2i )]

(0.5i
i=0

i=0

(0.5i (0.5i
i=0

= E[
i=1

(0.5i

t2i )

t2i )]

= E[
i=0

(0.5i+i+1) (0.5(2i+1) 1

2 t2i )]

=
i=0

(1) =
i=0

0.52i+1 0.52i+h
i=0

(h) =

|(h)| =
h=0 i=0

0.52i+

h=0

The ACF of xt is 6 for h = 0 and otherwise as described above; it is covariancestationary. This indicates an M A(inf) process (in order to know what rank its AR representation would have, I would need to compile the PACF. Since that was not asked, I am too lazy to do it here). As 0.5 is smaller than 1, the last expression above is smaller than innity, hence this function is also covariance-stationary. 5

5
The rst process is stationary, its ACF for any h > 0 is 0. Hence it is strictly stationary. The other two processes are not strictly stationary, since the ACF is not the same for all h. It would be the same, if the variance of t would be zero (then the ACF would be 0 h).

6
Since the ACF goes to zero as h rises for all three processes, they are all ergodic for the mean. This also means that the sum of the absolute values of the ACF is smaller than innite1 ; hence they are also ergodic for the second moment. This implies that they are asymptotically independent in the 1st and 2nd moments and we can estimate these.

7
My variables of choice come from Penn World Tables and are all for Spain: ki, which is the investment share of GDP per capita. PPP, which is the purchase parity power cgdp, which is PPP converted GDP per capita at current prices

7.a)
While log values of ki have a mean at 3 and do not change over time, the logs of cgdp and ppp do; ppp is increasing constantly over time while cgdp has a quite strong jump between 1970 and 1990. Growth values of all three variables are quite spiky. They jump around the mean quite often, yet there is some dependence over time (i.e. eects from one period last over several years).

7.b)

For the third process, the sum goes to zero because the coecient is smaller as 1, as shown previously

11 10 9 8 7 6 5 4 3 2

l_ki (left) l_cgdp (left) l_ppp (right)

-0.2 -0.4 -0.6 -0.8 -1 -1.2 -1.4 -1.6 -1.8 -2 -2.2 -2.4 2010

1950

1960 d_l_ki

1970

1980

1990

2000 d_l_cgdp

0.15 0.1 0.05 0 -0.05 -0.1 -0.15 -0.2 -0.25 1950

0.25 0.2 0.15 0.1 0.05 0 1960 1970 1980 d_l_ppp 1990 2000 -0.05 1950 1960 1970 1980 1990 2000

0.15 0.1 0.05 0 -0.05 -0.1 -0.15 1950 1960 1970 1980 1990 2000

ACF for l_ki 1 0.5 0 -0.5 -1 0 5 10 lag PACF for l_ki 1 0.5 0 -0.5 -1 0 5 10 lag 15 20 +- 1.96/T^0.5 15 20 +- 1.96/T^0.5

Figure 1: The PACF of log(ki) has a strong value for 1 and other values that appear to be close to zero. Its ACF is decreasing over time (but somehow not perfectly diminishing after 10 years). Hence I assume that it could be some kind of autoregressive time series, with a small autoregressive part (perhaps 1 or 2).

ACF for d_l_ki 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 0 5 10 lag PACF for d_l_ki 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 0 5 10 lag 15 +- 1.96/T^0.5 15 +- 1.96/T^0.5

20

20

Figure 2: I can see no known pattern in the PACF and ACF of the growth of ki.

ACF for l_cgdp 1 0.5 0 -0.5 -1 0 5 10 lag PACF for l_cgdp 1 0.5 0 -0.5 -1 0 5 10 lag 15 20 +- 1.96/T^0.5 15 20 +- 1.96/T^0.5

Figure 3: The PACF of log(cgdp) and its ACF resemble very strongly to an AR(1) process: PACF(1) is strongly positive and its other values are very close to zero. The ACF shows a perfectly diminishing pattern.

ACF for d_l_cgdp 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 0 5 10 lag PACF for d_l_cgdp 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 0 5 10 lag 15 +- 1.96/T^0.5 15 +- 1.96/T^0.5

20

20

Figure 4: Also here, I cannot identify the patterns for the growth of cgdp.

ACF for l_ppp 1 0.5 0 -0.5 -1 0 5 10 lag PACF for l_ppp 1 0.5 0 -0.5 -1 0 5 10 lag 15 20 +- 1.96/T^0.5 15 20 +- 1.96/T^0.5

Figure 5: Also PPP appears to be an AR(1) process, its characteristics are similar to those of cgdp.

ACF for d_l_ppp 1 0.5 0 -0.5 -1 0 5 10 lag PACF for d_l_ppp 1 0.5 0 -0.5 -1 0 5 10 lag 15 20 +- 1.96/T^0.5 15 20 +- 1.96/T^0.5

Figure 6: At least for PPP also its growth rates are making sense a little bit; both the PACF and ACF are diminishing over time. With some goodwill, one could interpret this as an ARMA process.

10

You might also like