MA(1) and Invertibility ARMA Model

Unit 9: ARMA Models: MA(1)

Spring 2016

1 / 18
MA(1) and Invertibility ARMA Model

Readings for Unit 9

Textbook chapter 3.2.

2 / 18
MA(1) and Invertibility ARMA Model

Last Unit

1 AR(p) process
2 AR(1) process
3 AR(1) in terms of backshift operator
4 AR(1) and causality

3 / 18
MA(1) and Invertibility ARMA Model

This Unit

1 Identifiability
2 MA(1) in terms of backshift operator
3 MA(1) and invertibility
4 ARMA model

4 / 18
MA(1) and Invertibility ARMA Model

Motivation

In the previous unit, we noted a condition for an AR(1) model to

be causal. In this unit, we will explore another issue, this time
regarding MA models. This issue is called identifiability.

5 / 18
MA(1) and Invertibility ARMA Model

1 MA(1) and Invertibility

2 ARMA Model

6 / 18
MA(1) and Invertibility ARMA Model

MA(q) Process

Recall the MA(q) model

xt = wt + θ1 wt−1 + ... + θq wt−q .

We can rewrite this using backshift operator as

xt = θ(B)wt (1)

where θ(B) = 1 + θ1 B + ... + θq B q which can be called the

. We already know that an MA(q)
process is stationary.

7 / 18
MA(1) and Invertibility ARMA Model

MA(1) Process
Let’s look at the MA(1) model, i.e.

xt = wt + θwt−1

We can then calculate the autocovariance function.

γ(0) = (1 + θ2 )σw2

and
γ(1) = θσw2
and zero for larger lags. Assume that we simulate from a model
using θ = a and σw2 = b, then look at the autocovariance

γ(0) = b + a2 b, γ(1) = ab

.
8 / 18
MA(1) and Invertibility ARMA Model

Identifiability

Question: Suppose we have another MA(1) model with θ = 1/a

and σw2 = a2 b, what is the autocovariance function for this model?

9 / 18
MA(1) and Invertibility ARMA Model

Identifiability

Which one is “correct”? Both are correct: this is an identifiability

problem as we have MA models that are not unique. We’ll use the
one that gives us an infinite order AR process.

10 / 18
MA(1) and Invertibility ARMA Model

Identifiability

Rewrite the MA(1) model as

wt = −θwt−1 + xt .

This is just like the AR(1) that we saw last time but now the roles
for xt , wt are reversed. So, we could write this as
∞
X
wt = (−θ)j xt−j
j=0
∞
X
= (−θ)j B j xt , (2)
j=0

provided that |θ| < 1.

11 / 18
MA(1) and Invertibility ARMA Model

MA(1) in Terms of Backshift Operator

(2) can be written as

∞
X
wt = πj xt−j = π(B)xt , (3)
j=0
P∞ j
where π(B) = j=0 πj B and πj = (−θ)j .

12 / 18
MA(1) and Invertibility ARMA Model

Identifiability

Question: What’s an intuitive explanation as to why we want

|θ| < 1 in (2)?

13 / 18
MA(1) and Invertibility ARMA Model

Invertible MA(1)
We could write an MA(1) as

xt = wt + θwt−1
∞
X
= wt + θ (−θ)j xt−1−j
j=0
∞
X
= wt − (−θ)j xt−j .
j=1

So, we have written the MA(1) as an infinite order AR process. An

MA model which can be written like this is called an
MA. We require |θ| < 1 for invertibility. We will later extend and
formalize a condition on the parameters θ1 , θ2 , · · · , θq for an
MA(q) process to be invertible.
14 / 18
MA(1) and Invertibility ARMA Model

1 MA(1) and Invertibility

2 ARMA Model

15 / 18
MA(1) and Invertibility ARMA Model

ARMA Model

Recall that an AR(p) model is given by

xt = φ1 xt−1 + φ2 xt−2 + · · · φp xt−p + wt

or in terms of backshift operator

φ(B)xt = wt , where φ(B) = 1 − φ1 B − φ2 B 2 − . . . − φp B p .

16 / 18
MA(1) and Invertibility ARMA Model

ARMA Model

An MA(q) model is given by

xt = wt + θ1 wt−1 + θ2 wt−2 + · · · + θq wt−q

or in terms of backshift operator

xt = θ(B)wt , where θ(B) = 1 + θ1 B + θ2 B 2 + . . . + θq B q .

17 / 18
MA(1) and Invertibility ARMA Model

ARMA Model

xt is called an ARMA (p,q) model if xt is stationary and can be

written as
φ(B)xt = θ(B)wt .
Thus far, we have seen a number of issues with the general
definition of ARMA(p,q) models:
ˆ Stationary AR models that depend on the .
ˆ MA models that are not (identifiability issue).
We will next look at one more issue with ARMA(p,q) models
before formalizing conditions for causality and invertibility.

18 / 18