Assignment 2

Economics of Forcasting & Analysis

Arjun Dua

Answer 1
Answer A
The given equation is,
𝑦! = 𝑎" + 0.75𝑦!#$ − 0.125𝑦!#% 𝜖!
The homogenous equation is given by,
𝑦! = 0.75𝑦!#$ − 0.125𝑦!#%
𝐿𝑒𝑡, 𝑦! = 𝐴𝛼 !
Plugging in the value to find the homogenous solution, we get
𝐴𝛼 ! − 0.75𝐴𝛼 !#$ + 0.125𝐴𝛼 !#% = 0
𝛼 % − 0.75𝛼 + 0.125 = 0
𝛼 % − 0.5𝛼 − 0.25𝛼 + 0.125 = 0
𝛼(𝛼 − 0.5) − 0.25(𝛼 − 0.5) = 0
(𝛼 − 0.5)(𝛼 − 0.25) = 0
Therefore the characteristic roots of the equation are:
𝛼$ = 0.5, 𝛼% = 0.25
Thus, the homogenous solution is given by,

𝑦!& = 𝐴$ (0.5)! + 𝐴% (0.25)!

where 𝐴$ and 𝐴% are arbitrary constants.
Finding the particular solution,
As we have seen in class, the particular solution has the form
𝑦!' = 𝛽( + ∑𝛼) 𝜖!#)
Here we can see that the constant 𝛽( is given by
 𝑎( 8𝑎"
𝛽" = =
(1 − 0.75 + 0.125) 3
Substituting 𝑦! = ∑𝛼) 𝜖!#)
𝛼( 𝜖! + 𝛼$ 𝜖!#$ + 𝛼% 𝜖!#% +. . . . . . .
= 0.75(𝛼( 𝜖!#$ + 𝛼$ 𝜖!#% + 𝛼% 𝜖!#* +. . . ) − 0.125(𝛼( 𝜖!#% + 𝛼$ 𝜖!#* + 𝛼% 𝜖!#+ +. . . ) + 𝜖!
Comparing both sides and matching like terms, we get
𝛼( = 1
𝛼$ = 0.75
𝛼% = 0.4375
Thus we can generalize the solution, and get
𝛼) = 0.75𝛼)#$ − 0.125𝛼)#% , ∀𝑖 ∈ [2, ∞)
Now the IRF if given by the coefficient of the the particular solutions, therefore we get
𝜕𝑦!
= 𝛼( = 1
𝜕𝜖!
𝜕𝑦!
= 𝛼$ = 0.75
𝜕𝜖!#$
Thus we can generalize the solution, and get
𝜕𝑦!
= 𝛼) = 0.75𝛼)#$ − 0.125𝛼)#% , ∀𝑖 ∈ [2, ∞)
𝜕𝜖!#)
Now since both the characteristic roots (0.25 and 0.5) are positive and less than 1, the IRF
,-
coverages to 𝑦! = * ! .

Answer B
Adding the particular and homogenous solutions from part A, to get the general solution
8𝑎(
𝑦! = 𝐴$ (0.5)! + 𝐴% (0.25)! + + 𝜖! + 0.75𝜖!#$ + 0.4375𝜖!#% +. ..
3
This implies that, Period 0:
8𝑎(
𝑦( = 𝐴$ + 𝐴% + + 𝜖( + 0.75𝜖#$ + 0.4375𝜖#% +. ..
3
Period 1:
8𝑎(
𝑦$ = (0.5)𝐴$ + 𝐴% (0.25) + + 𝜖$ + 0.75𝜖( + 0.4375𝜖#$ +. ..
3
For the sequence to be stationary 𝐴$ = 𝐴% = 0. Which would imply that,
8𝑎(
𝑦( = + 𝜖( + 0.75𝜖#$ + 0.4375𝜖#% +. ..
3
8𝑎(
𝑦$ = + 𝜖$ + 0.75𝜖( + 0.4375𝜖#$ +. ..
3

Answer C
Using the Yule-Walker equations to derive the correlogram, we get Since 𝑎( is a constant we can
ignore it as it will not have an affect on correlation.
Now we know that 𝑎$ = 0.75, 𝑎% = −0.125 and we also know that 𝜌( = 1.
Using the equation 𝜌$ = 𝑎$ + 𝑎% 𝜌$ , we get
2
𝜌$ =
3
Generally, we write
𝜌) = 0.75𝜌)#$ − 0.125𝜌)#% , ∀𝑖 ∈ [2, ∞)

Answer 2
Answer A
The given equation is,
𝑦! = 1.5𝑦!#$ − 0.5𝑦!#% + 𝜖!
The homogenous equation is given by,
𝑦! = 1.5𝑦!#$ − 0.5𝑦!#%
𝐿𝑒𝑡, 𝑦! = 𝐴𝛼 !
Plugging in the value to find the homogenous solution, we get
𝐴𝛼 ! − 1.5𝐴𝛼 !#$ + 0.5𝐴𝛼 !#% = 0
𝛼 % − 1.5𝛼 + 0.5 = 0
𝛼 % − 𝛼 − 0.5𝛼 + 0.5 = 0
𝛼(𝛼 − 1) − 0.5(𝛼 − 1) = 0
(𝛼 − 1)(𝛼 − 0.5) = 0
Therefore the characteristic roots of the equation are:
𝛼$ = 1, 𝛼% = 0.5
Thus, the homogenous solution is given by,

𝑦!& = 𝐴$ + 𝐴% (0.5)!
where 𝐴$ and 𝐴% are arbitrary constants.

Answer B
Finding the roots of the given equation, we get
1 − 1.5𝐿 + 0.5𝐿%
1 − 𝐿 − 0.5𝐿 + 0.5𝐿%
1(1 − 𝐿) − 0.5𝐿(1 − 𝐿)
(1 − 0.5𝐿)(1 − 𝐿)
𝐿 = 1,2
As we can see that the two inverse characteristic roots are the reciprocals of characteristic roots
found in Part A. The stability condition is for the characteristic roots to lie inside of the unit
circle or for the roots of the inverse characteristic equation to lie outside of the unit circle.

Answer C
Using forward iteration method to find the solution, we get
𝑦% == 1.5𝑦$ − 0.5𝑦( + 𝜖%
𝑦* = 1.5𝑦% − 0.5𝑦$ + 𝜖*
𝑦+ = 1.5𝑦* − 0.5𝑦% + 𝜖+
Substituting values from the above equation, we get
𝑦4 = −0.875𝑦( + 1.875𝑦$ + 1.75𝜖% + 1.5𝜖* + 𝜖+
Generalizing the result, we get
!#%

𝑦! = A 𝛼) 𝜖!#) + 𝛼!#$ 𝑦$ + 𝛼! 𝑦(
).(

where: 𝛼( = 1, 𝛼$ = 1.5, 𝛼! = 1 − 𝛼!#$ , and the remaining coefficients 𝛼) solve the difference
equation 𝛼) = 1.5𝛼)#$ − 0.5𝛼)#% .

Answer D
/#%

𝑦/ = A 𝛼) 𝜖/#) + 𝛼/#$ 𝑦$ + 𝛼/ 𝑦(
).(
Updating 𝑡 time periods we get
/#%

𝑦/0! = A 𝛼) 𝜖/0!#) + 𝛼/#$ 𝑦!0$ + 𝛼/ 𝑦!

).(

Answer E
1. 𝐸(𝑦! ) = 𝛼!#$ 𝑦$ + 𝛼! 𝑦(
2. 𝐸(𝑦!0$ ) = 𝛼! 𝑦$ + 𝛼!0$ 𝑦(
3. 𝑉𝑎𝑟(𝑦! ) = [1 + (𝛼$ )2 + (𝛼% )2+. . . +(𝛼!#% )2]𝜎 %
4. 𝑉𝑎𝑟(𝑦!0$ ) = [1 + (𝛼$ )2 + (𝛼% )2+. . . +(𝛼!#$ )2]𝜎 %
5. 𝐶𝑜𝑣(𝑦!0$ , 𝑦! ) = [𝛼( 𝛼$ + 𝛼$ 𝛼% +. . . +𝛼!#* 𝛼!#% ]𝜎 %

Answer 3
Answer A
Answer i
The equation for an AR(1) model is given by 𝑦! = 𝑎( + 𝑎$ 𝑦!#$ + 𝜖! .
Solving the above equation through iteration, we get
1
𝑎(
𝑦! = + A 𝑎$) 𝜖!#)
1 − 𝑎$
).(

Taking unconditional expectation, we get

𝑎(
𝑦=
1 − 𝑎$
𝑎( = 𝑦(1 − 𝑎$ )
Replacing 𝑎( in the equation, we will get
𝑦! = 𝑦(1 − 𝑎$ ) + 𝑎$ 𝑦!#$ + 𝜖!
𝑦! − 𝑦 = 𝑎$ (𝑦!#$ − 𝑦) + 𝜖!
Hence, the equivalent relationship is proved.

Answer ii
We use the forward-looking solution to derive the given equivalent form of the AR(1) model.
𝑦! = 𝑎( + 𝑎$ 𝑦!#$ + 𝜖!
Solving for 𝑦!0$ we will get,
 𝑦!0$ (𝑎( + 𝜖!0$ )
𝑦! = −
𝑎$ 𝑎$
Iterating the above equation n times, we get
UNABLE TO SOLVE

Answer B
The time series model is
𝑦! = 𝑎( + 𝑎$ 𝑦!#$ + 𝑎% 𝑡 + 𝜖!
The expected value of the above model is 𝐸(𝑦! ) = 𝑎( + 𝑎% 𝑡. This value changes with t,
therefore the series is not stationary.
𝑦! is stationary about the trend line 𝑎( + 𝑎% 𝑡 because when we will remove this underlying trend
which is a function of time only, the remaining series will be stationary. Therefore, 𝑦! =
𝑎$ 𝑦!#$ + 𝜖! is stationary, after removing the trend line. When 𝑦! is trend stationary variance will
be finite, any shock will only change 𝑦! temporarily, and the spectrum of 𝑦! will be finite when
frequency is zero.
Answer 4
Answer A
AIC is lowest for AR(3) Model, whereas BIC is lowest for ARMA(1,1). We do not see
autocorrelation. Since the t-statistics of coefficient on Lag 3 is not significant we drop the AIC
criterion. We know that, SBC selects more parsimonious models. ARMA (1,1) has white noise
residuals hence this is the best fit model.