STAT4005: TIME SERIES
TUTORIAL 3: AR Model
Michael Mun Lau, Cheung
Department of Statistics, The Chinese University of Hong Kong
Week 4, 2018 Semester 2
P∞ i
Exercise MA(∞) Consider the process Yt = [ i=0 (0.8B) ] Zt , where {Zt } ∼ N (0, σ 2 ).
a) Find E(Yt ) and V ar(Yt ).
b) Find the autocovariance function of {Yt }.
c) Find the autocorrelation function of {Yt }.
d) Is Yt weakly stationary?
1
i.i.d.
Exercise AR (1) Let Yt = 0.8Yt−1 + Zt , where Zt ∼ N (0, σ 2 ).
a) Find P
the values of ψ0 , ψ1 and ψ2 if the above process is written in the form of MA model
Yt = ∞ j=0 ψj Zt−j .
b) Find the values
P∞ of ψj , j = 0, 1, 2, . . . if the above process is written in the form of MA
model Yt = j=0 ψj Zt−j .
c) Find the autocovariance function of Yt by Yule-Walker Equation.
d) Is Yt stationary?
2
STRATEGY
1) AR format ⇔ MA format 2) Yule-Walker equation
i.i.d.
Motivation Let Yt = 2Yt−1 + Zt , where Zt ∼ N (0, σ 2 ).
a) Try to express Yt in the form of ∞
P
j=0 ψj Zt−j with all |ψ| < ∞.
b) Try to express Yt−1 in the form of ∞
P
j=0 ψj Zt−1+j with all |ψ| < ∞.
c) Is Yt stationary? Yt is not causal!
3
Exercise AR (2) Let (1 − 1.1B + 0.24B 2 )Yt = at , where {at } ∼ W N (0, σ 2 ).
a) Find P
the values of ψj , j = 0, 1, 2 if the above process is written in the form of MA model
at = ∞ j=0 ψj at−j .
b) Find the values
P∞ of ψj , j = 0, 1, 2, . . . if the above process is written in the form of MA
model Zt = j=0 ψj at−j .
c) Find ACF ρ1 and ρ2 of Yt .
d) Is Yt stationary? Is Yt causal?
4
Exercise MA (2) Let Xt = at − 0.4at−2 , where {at } ∼ N (0, σ 2 ).
a) Find the values
P∞ of φj , j = 0, 1, 2, 3, . . . if the above process is written in the form of AR
model at = j=0 φj Xt−j .
b) Is Xt stationary? Is Xt causal? Is Xt invertible?
MODEL IDENTIFICATION
Suppose {at } is white noise sequence, types of model:
• MA(q): Xt = at + φ1 at−1 + · · · + φq at−q
• AR(p): Xt + ψ1 Xt−1 + · · · + ψp Xt−p = at
• ARI(p,d): (1 − B)d (Xt + ψ1 Xt−1 + · · · + ψp Xt−p ) = at
AR or MA characteristic equation
• AR polynomial: 1 + ψ1 B + · · · + ψp B p ⇒ AR equation: 1 + ψ1 x + · · · + ψp xp = 0
• MA polynomial: 1 + φ1 B + · · · + φp B p ⇒ MA equation: 1 + φ1 x + · · · + φp xp = 0
• Relationship between ROOTS & polynomial FACTORIZATION:
• ROOTS = r1 , r2 , . . . , rn ⇔ Polynomial = (1 − r1−1 B)(1 − r2−1 B) . . . (1 − rn−1 B)
5
DEFINITIONS
A model Xt is Stationary if it can be written as Xt = ∞
P
• k=−∞ ψk at−k , with
all ψ < ∞ and {at } is white noise. !!! Can have future white noise !!!
• CHECK: All roots in AR characteristic equation: |x| =
6 1
A model Xt is Causal if it can be written as Xt = ∞
P
• k=0 ψk at−k , with
all ψ < ∞ and {at } is white noise. !!! Only past white noise !!!
• CHECK: All roots in AR characteristic equation: |x| > 1
A model Xt is Invertible if it can be written as ∞
P
• k=0 ψk Xt−k = at , with
all ψ < ∞ and {at } is white noise. !!! Past observations !!!
• CHECK: All roots in MA characteristic equation: |x| > 1
i.i.d.
Exercise Given at ∼ N (0, 1). Identify the following models (state the type and the order
of the models) and state whether the models are stationary, causal and invertible.
a) Zt = 1.4Zt−1 − 0.4Zt−2 + at
b) Zt = −0.5Zt−1 + 0.14Zt−2 + at
c) Zt = at − 3at−1 + 3at−2 − at−3
d) Zt = 2at − 2.2at−1 + 0.5at−2
-THE END-
