MATH7970 Chap.5 Autoregressive Models and Stability
Autoregressive Models and Stability
Feedback from observed history
An AR model links the current value to lagged observations and a new innovation. Coefficients govern persistence and sign patterns, but the intercept is not generally the long-run mean.
Roots control the recursion
First-order stability requires a coefficient magnitude below one.
The alternating impulse response makes that interpretation visible across successive future dates. Higher-order stability depends on characteristic roots, so coefficients must be assessed jointly rather than one at a time.
Centre around the long-run mean
Writing an AR equation in deviations from its mean separates equilibrium level from dynamic persistence.
The coefficient then states how much of the previous deviation survives. This form prevents the intercept from being mistaken for the mean and makes multi-step forecasts clear: repeated powers carry the current gap toward zero when the recursion is stable.
Substituting the fixed point back into the original equation verifies the conversion.
Read sign and persistence together
A positive coefficient preserves the direction of a shock, while a negative coefficient produces alternating responses. Magnitude controls the rate of decay under stability. Coefficients with equal absolute values can therefore share persistence length but imply opposite paths.
Sketch the first several impulse responses before calculating them. The expected sign sequence is a powerful check on powers and on whether the effect is measured relative to the mean.
Higher-order stability is joint
Individual coefficient magnitudes do not settle a higher-order recursion. Characteristic roots combine all coefficients and determine whether effects decay, oscillate or explode.
State the polynomial convention before using inside or outside unit-circle language, because equivalent forms reverse that description. Complex roots can generate damped cycles that no single coefficient reveals. Stability must be established before a long-run mean or mean-reverting forecast is interpreted.
Forecast recursively and track uncertainty
At one step, required lags are observed.
At later horizons, earlier forecasts replace unknown future observations. Future innovations are set to zero only for the conditional point forecast; their uncertainty still expands the interval. Label values by date to avoid inserting the wrong horizon. Compare the calculated path with the impulse-response shape and stability result.
If those stories conflict, recheck indices, signs and whether levels were confused with deviations.
Distinguish persistence from causation
An AR coefficient describes predictive dependence on the series own history after the included structure is considered. It does not show that the previous observation causes the current one. A common driver, aggregation or omitted seasonal process can create persistence.
Interpret the coefficient as a dynamic statistical relation, then use subject knowledge to discuss mechanisms separately. This distinction matters in the project: a strong autoregression can forecast effectively without supporting a causal policy claim, and a causal story does not excuse unstable roots or patterned residuals in the forecasting model.
What this chapter covers
- 01
AR recursion
- 02
Long-run mean
- 03
Shock decay
- 04
Negative persistence
- 05
Characteristic roots
- 06
Multi-step forecasting
Worked example · free
Trace a stable shock
- 3Find the mean.
- 4Propagate the shock.
- 1State stability.
Key terms
- Autoregressive Process
- A process depending on its lagged observations plus a current innovation.
- Long-run Mean
- The stable expectation to which forecasts revert when the model permits it.
- Innovation
- New unpredictable information arriving at the current time.
- Characteristic Root
- A polynomial root governing stability and dynamic shape.
- Impulse Response
- The sequence of effects generated by one innovation over future dates.
Autoregressive Models and Stability FAQ
Why is the intercept not the mean?
Persistence carries part of the expected level forward, so the mean solves a fixed-point equation and equals c divided by one minus phi in a stable AR(1). Substituting the mean into the expectation equation shows why this fixed point is required.
What does a negative AR coefficient do?
It makes shock effects alternate in sign while shrinking when the coefficient magnitude is below one. The alternating impulse response makes that interpretation visible across successive future dates.
Why inspect roots in higher-order models?
Stability depends on combinations of coefficients represented by the characteristic polynomial, not on each coefficient separately. Complex roots can also generate damped cycles that individual signs do not reveal.
Exam move
For each AR equation, calculate the mean, check stability and trace one unit shock. Predict the sign pattern before multiplying powers of the coefficient. Build a table for several AR coefficients showing mean, stability, sign pattern and the first five impulse responses. Add one higher-order example and inspect its characteristic roots before interpreting coefficients.
Forecast recursively for three horizons, labelling when an observed lag is replaced by an earlier forecast. Compare the point path with uncertainty from unseen innovations. Finally rewrite the equation around its mean and verify that both forms yield the same forecast. This sequence catches intercept, index and stability mistakes before final interpretation.
Explain every impulse path aloud using direction, decay and reference level. If those words do not match the calculated signs or root condition, repair the mismatch before forecasting. Preserve the root calculation beside every claimed long-run mean or stable response.
Working through Autoregressive Models and Stability in MATH7970? Sia is AskSia’s AI Mathematics tutor — ask any MATH7970 Autoregressive Models and Stability question and get a clear, step-by-step explanation grounded in how MATH7970 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.
