ARMA(1,1) Processes

For an ARMA(1, 1) process

image210z

Now let’s suppose that |φ1| < 1. We show how to create the MA(∞) representation as follows:

image211z

Thusimage212z

where ψ0 = 1 and for j > 0image213z

The MA(∞) representation is therefore

image214z

If |φ1| < 1, then this ARMA(1,1) process is stationary. It also turns out that when |θ1| < 1, the process is invertible.

Example 1: Find the MA(∞) form of the ARMA(1, 1) process yi =.4yi-1 +εi – .2εi-1

image219z

image028c

etc.

We get the same result using the Real Statistics PSICoeff array function as shown in Figure 1.

Psi coefficients ARMA process

Figure 1 – Finding MA(∞) Coefficients

Property 1: The following is true for an ARMA(1,1) process

Autocovariance 0

Proof: See ARMA Proofs

Property 2: The following is true for an ARMA(1,1) process

image222z

and for k > 1image223z

ProofSee ARMA Proofs

Observation: Since z1 = 1/φ1 is the root of the characteristic polynomial φ(z) = 1 – φ1z, we can express the second term in Property 2 as

image224z

It turns out that this can be generalized to other ARMA processes.

Property 3: The following is true for an ARMA(1, 1) process

rho 1

and for k > 1image226z

Proof: See ARMA Proofs

Observation: When θ1 = –φ1 for an ARMA(1, 1) process, we note that γ0 = σ2 and ρk = 0 for all k > 1, which are the characteristics of white noise.

In fact, the white noise process with zero mean takes the form

yi = εi

We see that φyi – φ1 εi  = 0  for all i, and in particular φyi-1 – φ1 εi-1  = 0. Thus, the process takes the form

image227z

which is an ARMA(1,1) process with θ1 = –φ1.

Example 2: Simulate a sample of 105 elements from the ARMA(1,1) process

image228z

where εiN(0, 1) and calculate the ACF.

This is done in Figure 2 by placing the formula =NORM.S.INV(RAND()) in cell B7 and the formula =0.7*C6+B7-0.2*B6 in cell C7, highlighting the range B7:C111 and pressing Ctrl-D (only the first 16 elements of the simulation is shown in Figure 2).

Simulated ARMA(1,1) process

Figure 2 – Simulated ARMA(1,1) process

The ACF values are shown in Figure 3 where the theoretical values are based on Property 3.

ACF ARMA(1,1) process

Figure 3 – ACF for ARMA(1,1) Process

Cell M6 contains the formula =ACF($C$12:$C$111,L6), and similarly, for the other cells in column M, cell N6 contains the formula =(Q5+Q6)*(1+Q5*Q6)/(1+2*Q5*Q6+Q5^2) and cell N7 contains the formula =N6*Q$5, and similarly for rest of the cells in column N.

Since |φ1| = .7 < 1 and |θ1| = .2 < 1, this process is both stationary and invertible.

Example 3: Simulate a sample of 105 elements from the ARMA(1,1) process

image229z

where εiN(0, 1) and calculate the ACF.

The only difference between this process and the one in Example 2 is the constant term. The approach is identical, except that this time you place the formula =3+0.7*C6+B7-0.2*B6 in cell C7. The result is shown in Figure 4. Note that the ACF values are identical to those in Figure 3.

ARMA(1,1) process zero mean

Figure 4 – Simulated ARMA(1,1) Process with non-zero mean 

8 thoughts on “ARMA(1,1) Processes”

  1. Hi Charles,
    Thank you for your amazing work.
    I would like to ask you: in a real dataset, how are the residuals calculated? Are they the residuals from (actual return – AR model predicted return) or something else? Thank you!

    Reply
  2. Dear Charles,

    I was wondering if the excel-file from “forecasting-arma” is also downloadable, so i could get a better understanding of Example 1 and Example 2.

    I am currently working on my master thesis, whereby i want to forecast some dataset i have based on the ARIMA(p,d,q) model. Here for i have succeeded to make a ACF and PaCF of the current and differenced data set. To test different type of ARMINA models i think i need to estimate these parameters for several configurations based on the ACF and PACF results.

    Thanks,
    Kelvin

    Reply

Leave a Comment