STAT8003 Chap.7 ARMA Models and Invertibility
ARMA Models and Invertibility
Keep the two memory channels visible
This chapter combines autoregressive feedback and moving-average shock memory in one polynomial model. The AR side carries earlier process values forward, while the MA side distributes a finite set of innovations. Writing phi and theta on separate lines protects their different roles.
Match coefficients in chronological order
The AR polynomial controls stationarity.
Once its roots pass, substitute a general innovation expansion and compare the coefficient of each shock. The MA contribution affects the early weights and then ends after q, while the AR recursion continues.
At sufficiently late lags, the ACF likewise follows the AR recurrence because the finite MA overlap has disappeared.
Reverse direction for invertibility
Stationarity maps innovations into a stable observed process. Invertibility asks whether current innovations can be recovered from current and past observations through a convergent reverse representation.
This property supports residual analysis but is not another name for stationarity. Each conclusion should identify the direction it has actually proved.
What this chapter covers
- 01
ARMA equation
- 02
AR and MA polynomials
- 03
Stationarity roots
- 04
Psi coefficient matching
- 05
Late-lag ACF recurrence
- 06
Invertibility
Worked example · free
Expand an ARMA(1,1)
- 2Match the current innovation.
- 3Combine AR and MA contributions at lag one.
- 3Continue with the AR recursion.
Key terms
- ARMA Process
- A stationary process combining autoregressive feedback with a finite moving-average innovation filter. Its early weights reflect both components, while later weights continue through the AR recursion.
- AR Polynomial
- The feedback polynomial whose roots control stationary existence. It must be kept separate from the MA polynomial, which governs the finite shock filter and invertibility conditions.
- Invertibility
- A convergent reverse representation that recovers innovations from observed values. It supports interpreting fitted residuals as estimated shocks and is distinct from forward stationarity.
ARMA Models and Invertibility FAQ
Which part of an ARMA model controls stationarity?
The roots of the AR characteristic polynomial control whether a unique stationary solution exists. Adding MA terms changes early covariance and innovation weights but does not replace the AR root condition.
When does the ARMA ACF follow an AR recurrence?
For lags greater than the moving-average order q, the finite MA contribution no longer overlaps the older process value in the covariance calculation. The autocovariance and autocorrelation then obey the AR recurrence.
Is stationarity the same as invertibility?
No. Stationarity gives a stable process representation from innovations to observations and is governed by the AR side. Invertibility runs the relationship back from observations to innovations and concerns the MA representation.
How should I expand an ARMA model into innovations?
Test the AR roots first, then substitute a general psi-weight expansion. Match the coefficient of the current shock, each finite MA shock and every later innovation in chronological order. Mark the lag q where theta terms end; beyond it, only the AR recursion continues.
Finally substitute the first few weights back into the equation and state whether the result establishes stationarity, causality or merely a formal coefficient sequence.
Why should phi and theta be written separately?
The phi polynomial carries autoregressive feedback and supplies the stationarity root test. The theta polynomial carries the finite shock filter and enters invertibility. Separating them prevents an MA coefficient from being used in the wrong property test.
Exam move
Write the AR and MA polynomials on separate lines. Test AR roots for stationarity, then match innovation coefficients in chronological order. Mark the lag where the MA terms end, because the recursion simplifies after that point. Finish by stating whether the question concerns forward stationarity or reverse innovation recovery.
Translate every visible plus or minus sign into the course theta convention before using a coefficient formula. For ARMA(1,1), derive psi one from phi minus theta rather than recalling it without context, then apply the AR multiplier to later weights. For a higher order model, write theta j as zero beyond q so one recursion can be used without inventing further MA terms.
Check that the generated psi sequence decays under the accepted AR root condition. Next, separate second-order moments into early and late lags. At lags no greater than q, MA overlap contributes extra terms; at later lags, the AR recurrence takes over. Practise explaining this boundary in words before calculating.
For invertibility, reverse the direction of the diagram: start from observed Z values and ask whether the current innovation can be written as a convergent series of current and past observations. Do not use a stationary AR-root result as evidence for that reverse claim. Close each exercise with a property sentence naming exactly what has been proved: stationary, causal, invertible or merely linear.
That sentence is a strong defence against treating the four labels as synonyms.
Working through ARMA Models and Invertibility in STAT8003? Sia is AskSia’s AI Statistics tutor — ask any STAT8003 ARMA Models and Invertibility question and get a clear, step-by-step explanation grounded in how STAT8003 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.