0% found this document useful (0 votes)
23 views

Consider y - 1. The Stationarity Condition Is: Two Solutions of X From Are Outside The Unit Circle. 2. Rewriting The AR (2) Model

1. The document describes the properties of an AR(2) time series model, including its stationarity condition, mean, autocovariance and autocorrelation functions, and partial autocorrelations. 2. It also provides the log-likelihood function of an AR(2) model in innovation form and describes how to extend the AR(2) model to include a drift term. 3. The final section gives an example of generalizing to an AR(p) model.

Uploaded by

Vidaup40
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
23 views

Consider y - 1. The Stationarity Condition Is: Two Solutions of X From Are Outside The Unit Circle. 2. Rewriting The AR (2) Model

1. The document describes the properties of an AR(2) time series model, including its stationarity condition, mean, autocovariance and autocorrelation functions, and partial autocorrelations. 2. It also provides the log-likelihood function of an AR(2) model in innovation form and describes how to extend the AR(2) model to include a drift term. 3. The final section gives an example of generalizing to an AR(p) model.

Uploaded by

Vidaup40
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 15

Example: AR(2) Model: Consider yt = φ1 yt−1 + φ2 yt−2 + t .

1. The stationarity condition is: two solutions of x from φ(x) = 1 − φ1 x − φ2 x2 = 0


are outside the unit circle.

2. Rewriting the AR(2) model,

(1 − φ1 L − φ2 L2 )yt = t .

Let 1/α1 and 1/α2 be the solutions of φ(x) = 0.

Then, the AR(2) model is written as:

(1 − α1 L)(1 − α2 L)yt = t ,

which is rewritten as:


1
yt = t
(1 − α1 L)(1 − α2 L)

96
( )
α1 /(α1 − α2 ) −α2 /(α1 − α2 )
= + t
1 − α1 L 1 − α2 L

3. Mean of AR(2) Model:

When yt is stationary, i.e., α1 and α2 are outside the unit circle,

µ = E(yt ) = E(φ(L)t ) = 0

4. Autocovariance Function of AR(2) Model:

γ(τ) = E((yt − µ)(yt−τ − µ)) = E(yt yt−τ )


( )
= E (φ1 yt−1 + φ2 yt−2 + t )yt−τ

= φ1 E(yt−1 yt−τ ) + φ2 E(yt−2 yt−τ ) + E(t yt−τ )






 φ1 γ(τ − 1) + φ2 γ(τ − 2), for τ , 0,
=
 φ γ(τ − 1) + φ γ(τ − 2) + σ2 ,
 for τ = 0.
1 2 

97
The initial condition is obtained by solving the following three equations:

γ(0) = φ1 γ(1) + φ2 γ(2) + σ2 ,

γ(1) = φ1 γ(0) + φ2 γ(1),

γ(2) = φ1 γ(1) + φ2 γ(0).

Therefore, the initial conditions are given by:


( )
1 − φ2 σ2
γ(0) = ,
1 + φ2 (1 − φ2 )2 − φ21
( )( )
φ1 φ1 1 − φ2 σ2
γ(1) = γ(0) = .
1 − φ2 1 − φ2 1 + φ2 (1 − φ2 )2 − φ21

Given γ(0) and γ(1), we obtain γ(τ) as follows:

γ(τ) = φ1 γ(τ − 1) + φ2 γ(τ − 2), for τ = 2, 3, · · ·.

98
5. Another solution for γ(0):

From γ(0) = φ1 γ(1) + φ2 γ(2) + σ2 ,

σ2
γ(0) =
1 − φ1 ρ(1) − φ2 ρ(2)

where

φ1 φ21 + (1 − φ2 )φ2
ρ(1) = , ρ(2) = φ1 ρ(1) + φ2 = .
1 − φ2 1 − φ2

6. Autocorrelation Function of AR(2) Model:

Given ρ(1) and ρ(2),

ρ(τ) = φ1 ρ(τ − 1) + φ2 ρ(τ − 2), for τ = 3, 4, · · ·,

99
7. φ k,k = Partial Autocorrelation Coefficient of AR(2) Process:
 φ 
   k,1  
 1 ρ(1) ··· ρ(k − 2) ρ(k − 1)     ρ(1) 
   φk,2   
 ρ(1)
 1 ρ(k − 3) ρ(k − 2)   .   ρ(2) 
 .. .. .. ..   ..  =  .  ,
 . . . .     .. 
    
   φk,k−1   
ρ(k − 1) ρ(k − 2) ··· ρ(1) 1   ρ(k)
φk,k
for k = 1, 2, · · ·.

100

1 ρ(1) · · · ρ(k − 2) ρ(1)

ρ(1) 1 ρ(k − 3) ρ(2)
.. .. .. ..
. . . .

ρ(k − 1) ρ(k − 2) · · · ρ(1) ρ(k)
φk,k =
1 ρ(1) · · · ρ(k − 2) ρ(k − 1)


ρ(1) 1 ρ(k − 3) ρ(k − 2)
.. .. .. ..
. . . .

ρ(k − 1) ρ(k − 2) · · · ρ(1) 1

101
Autocovariance Functions:

γ(1) = φ1 γ(0) + φ2 γ(1),

γ(2) = φ1 γ(1) + φ2 γ(0),

γ(τ) = φ1 γ(τ − 1) + φ2 γ(τ − 2), for τ = 3, 4, · · ·.

Autocorrelation Functions:

φ1
ρ(1) = φ1 + φ2 ρ(1) = ,
1 − φ2
φ21
ρ(2) = φ1 ρ(1) + φ2 = + φ2 ,
1 − φ2
ρ(τ) = φ1 ρ(τ − 1) + φ2 ρ(τ − 2), for τ = 3, 4, · · ·.

102
φ1
φ1,1 = ρ(1) =
1 − φ2

1 ρ(1)

ρ(1) ρ(2) ρ(2) − ρ(1)2
φ2,2 = = = φ2
1 ρ(1) 1 − ρ(1)2

ρ(1) 1

1 ρ(1) ρ(1)

ρ(1) 1 ρ(2)

ρ(2) ρ(1) ρ(3)
φ3,3 =
1 ρ(1) ρ(2)

ρ(1) 1 ρ(1)

ρ(2) ρ(1) 1

103
(ρ(3) − ρ(1)ρ(2)) − ρ(1)2 (ρ(3) − ρ(1)) + ρ(2)ρ(1)(ρ(2) − 1)
= = 0.
(1 − ρ(1)2 ) − ρ(1)2 (1 − ρ(2)) + ρ(2)(ρ(1)2 − ρ(2))

8. Log-Likelihood Function — Innovation Form:



T
log f (yT , · · · , y1 ) = log f (y2 , y1 ) + log f (yt |yt−1 , · · · , y1 )
t=3

where
 ( γ(0) γ(1) )−1 ( y )
1 γ(0) γ(1) −1/2

 1 1 

f (y2 , y1 ) = exp − (y1 y2 )  ,
2π γ(1) γ(0) 2 γ(1) γ(0) y2
( )
1 1
f (yt |yt−1 , · · · , y1 ) = √ exp − 2 (yt − φ1 yt−1 − φ2 yt−2 ) .
2
2πσ2 2σ

Note as follows:
( γ(0) γ(1) ) ( 1 ρ(1) ) ( 1 φ1 /(1 − φ2 ) )
= γ(0) = γ(0) .
γ(1) γ(0) ρ(1) 1 φ1 /(1 − φ2 ) 1

104
9. AR(2) +drift: yt = µ + φ1 yt−1 + φ2 yt−2 + t

Mean:

Rewriting the AR(2)+drift model,

φ(L)yt = µ + t

where φ(L) = 1 − φ1 L − φ2 L2 .

Under the stationarity assumption, we can rewrite the AR(2)+drift model as


follows:
yt = φ(L)−1 µ + φ(L)−1 t .

Therefore,

µ
E(yt ) = φ(L)−1 µ + φ(L)−1 E(t ) = φ(1)−1 µ =
1 − φ1 − φ2

105
Example: AR(p) model: Consider yt = φ1 yt−1 + φ2 yt−2 + · · · + φ p yt−p + t .

1. Variance of AR(p) Process:

Under the stationarity condition (i.e., the p solutions of x from φ(x) = 0 are
outside the unit circle),

σ2
γ(0) = .
1 − φ1 ρ(1) − · · · − φ p ρ(p)

Note that γ(τ) = ρ(τ)γ(0).

Solve the following simultaneous equations for τ = 0, 1, · · · , p:

γ(τ) = E((yt − µ)(yt−τ − µ)) = E(yt yt−τ )






 φ1 γ(τ − 1) + φ2 γ(τ − 2) + · · · + φ p γ(τ − p), for τ , 0,
=

 φ1 γ(τ − 1) + φ2 γ(τ − 2) + · · · + φ p γ(τ − p) + σ2 , for τ = 0.

106
2. Estimation of AR(p) Model:

1. OLS:

T
min (yt − φ1 yt−1 − φ2 yt−2 − · · · − φ p yt−p )2
φ1 , · · · , φ p t=p+1

2. MLE:

max log f (yT , · · · , y1 )


φ1 , · · · , φ p

where

T
log f (yT , · · · , y1 ) = log f (y p , · · · , y2 , y1 ) + log f (yt |yt−1 , · · · , y1 ),
t=p+1
  y 
  1 
  
 1  y2 
f (y p , · · · , y2 , y1 ) = (2π)−p/2 |V|−1/2 exp − (y1 y2 · · · y p )V −1  . 
 2  .. 
  
yp

107
 ρ(1) ··· ρ(p − 2) ρ(p − 1) 
 1


 ρ(1) 1 ρ(p − 3) ρ(p − 2) 
V = γ(0)  .. .. .. .. 
 . . . . 
 
ρ(p − 1) ρ(p − 2) ··· ρ(1) 1
1 ( 1 )
f (yt |yt−1 , · · · , y1 ) = √ exp − 2 (yt − φ1 yt−1 − φ2 yt−2 − · · · − φ p yt−p )2
2πσ2 2σ

3. Yule=Walker (ユール・ウォーカー) Equation:

Multiply yt−1 , yt−2 , · · ·, yt−p on both sides of yt = φ1 yt−1 + φ2 yt−2 + · · · + φ p yt−p +

108
t = yt , take expectations for each case, and divide by the sample variance γ̂(0).
 φ 
   1  
 1 ρ̂(1) · · · ρ̂(p − 2) ρ̂(p − 1)     ρ̂(1) 
   φ2   
 ρ̂(1)
 1 ρ̂(p − 3) ρ̂(p − 2)   .   ρ̂(2) 
 .. .. .. ..   ..  =  . 
 . . . .     .. 
    
   φ p−1   
ρ̂(p − 1) ρ̂(p − 2) ··· ρ̂(1) 1   ρ̂(p)
φp
where

1 ∑
T
γ̂(τ)
γ̂(τ) = (yt − µ̂)(yt−τ − µ̂), ρ̂(τ) = .
T t=τ+1 γ̂(0)

3. AR(p) +drift: yt = µ + φ1 yt−1 + φ2 yt−2 + · · · φ p yt−p + t

Mean:
φ(L)yt = µ + t

109
where φ(L) = 1 − φ1 L − φ2 L2 − · · · − φ p L p .

yt = φ(L)−1 µ + φ(L)−1 t

Taking the expectation on both sides,

E(yt ) = φ(L)−1 µ + φ(L)−1 E(t ) = φ(1)−1 µ


µ
=
1 − φ1 − φ2 − · · · − φ p

4. Partial Autocorrelation of AR( p) Process:

φk,k = 0 for k = p + 1, p + 2, · · ·.

110

You might also like