Skip to content
All library documents

Mean Reversion in AR and ARMA Models as Discrete OU Analogues

Article Quant Q&A · Author: Sane
Read the original

Summary

The document discusses how mean reversion in an Ornstein–Uhlenbeck process relates to autoregressive and moving-average time-series models. Its answer identifies the sum of the autoregressive coefficients in an AR(p) model as governing mean-reverting behavior in the OU analogy. It extends this distinction to ARMA(p,q): the autoregressive terms mainly control long-run reversion, while moving-average terms shape the process’s short-term path.

The explanation rests on the intuition that MA terms affect near-term behavior without determining the long-run behavior. It points to a technical reference, but the document itself gives no derivation, assumptions, parameter restrictions, or worked example. Thus, it offers a qualitative coefficient interpretation rather than a complete test for stationarity or mean reversion; the exact behavior still depends on model specification and conditions not discussed here.

Key ideas

  • For an AR(p) model viewed as a discrete OU analogue, the sum of the autoregressive coefficients is identified as governing mean-reverting behavior.
  • In an ARMA model, autoregressive terms mainly determine long-run behavior.
  • Moving-average terms affect short-term behavior and the path the process takes as it reverts.
  • The explanation is qualitative and does not state assumptions or provide a formal derivation.

Tags

StatisticsMean reversionFixed income
Full text
# Is it possible to discretize OU with a more general AR(p) / ARMA (p,q) models?


# Is it possible to discretize OU with a more general AR(p) / ARMA (p,q) models?












The discrete analogue of an OU process is a simple AR(1) model. More general AR(p) or ARMA(p,q) models can also be regarded as discrete analogues of an OU process? If so, which coefficients describe its mean-reverting behavior?

My assumption is that for AR(p) model, sum of all coefficients should capture the mean-reversion behavior of the process. If so, is this true for ARMA as well or mean-reversion pattern is also governed by MA coefficients?

## Answer by Mahavir Bhattacharya (score 1)

https://quant.stackexchange.com/a/79151

Yes, you're correct.

In an AR(p) model, the sum of all autoregressive coefficients governs the mean-reverting behaviour in case of an OU process.

In an ARMA(p,q) model, the AR terms predominantly dictate the mean revertive behaviour, but the MA terms influence the path taken for the mean reversion.

An intuition would be to realize that the MA terms affect the short term behavior, but not the long term behavior of the process.

Reference: https://www.cs.upc.edu/~argimiro/mypapers/Journals/2014/ACCOUp2014.pdf (Page 12)

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.