Bayesian Phillips-Perron Unit Root Test
Also known as: Bayesian PP test, Bayesian Phillips-Perron test, Bayesian nonparametric unit root test, Bayes PP unit root
The Bayesian Phillips-Perron unit root test combines the nonparametric long-run variance correction of the classical Phillips-Perron test with a Bayesian inferential framework. Instead of a p-value, it yields a posterior probability or Bayes factor quantifying evidence for or against a unit root, allowing researchers to incorporate prior economic knowledge and obtain probability statements directly about the persistence of a time series.
Key highlights
- Provides a direct posterior probability or Bayes factor rather than a frequentist p-value, enabling straightforward probability statements about persistence.
- Retains the PP test's nonparametric long-run variance correction, avoiding the need to select lag length as in the ADF test.
- Naturally accommodates prior economic knowledge, which can sharpen inference in small samples where frequentist tests have low power.
- Coherently handles model uncertainty through Bayesian model averaging across unit-root and stationary specifications.
- Avoids pre-testing distortions common in sequential lag-selection procedures used with ADF tests.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use the Bayesian PP test when you need a probability statement about persistence rather than a binary decision, when your sample is small or moderately sized and frequentist tests have low power, or when you want to incorporate prior economic knowledge (e.g., strong a priori belief in mean reversion for interest rates). It is particularly valuable when classical PP and ADF tests give conflicting results. Do not use it as a mechanical substitute for the classical PP test if you have no defensible prior and cannot justify a specific prior choice; in that case, transparent sensitivity analysis across priors is essential. It is also less appropriate when structural breaks are suspected — use Zivot-Andrews or Fourier-based unit root tests instead.
Strengths & limitations
- Provides a direct posterior probability or Bayes factor rather than a frequentist p-value, enabling straightforward probability statements about persistence.
- Retains the PP test's nonparametric long-run variance correction, avoiding the need to select lag length as in the ADF test.
- Naturally accommodates prior economic knowledge, which can sharpen inference in small samples where frequentist tests have low power.
- Coherently handles model uncertainty through Bayesian model averaging across unit-root and stationary specifications.
- Avoids pre-testing distortions common in sequential lag-selection procedures used with ADF tests.
- Results can be sensitive to the choice of prior on the autoregressive parameter, especially with small samples; this sensitivity must be reported and discussed.
- Marginal likelihoods required for Bayes factors are analytically intractable under many priors and require numerical integration or MCMC, raising computational cost.
- The nonparametric bandwidth selection for the long-run variance estimator introduces a tuning choice that can affect results.
- Less widely implemented in standard software than classical PP or ADF tests, making replication harder for practitioners.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
How does the Bayesian PP test differ from the classical PP test?
The classical PP test produces a t-statistic and p-value under the null of a unit root, using asymptotic critical values. The Bayesian version places a prior on the autoregressive parameter and computes a posterior probability or Bayes factor, giving a direct measure of evidence for each hypothesis rather than a binary decision.
How does the Bayesian PP test differ from the Bayesian ADF test?
Both apply Bayesian inference to unit root testing, but the underlying regression differs. The ADF test controls for serial correlation by adding lagged differences (requiring lag selection), while the PP test corrects the variance nonparametrically with a kernel estimator (requiring bandwidth selection). The Bayesian PP test thus avoids lag-length pre-testing at the cost of bandwidth choice.
What prior should I use?
There is no universally correct answer. A flat prior on the autoregressive coefficient is common but can implicitly favour the unit root. An informative Normal prior centred slightly below unity (e.g., around 0.9) reflects mean reversion beliefs. Always report results under at least two or three prior specifications to demonstrate robustness.
What Bayes factor value constitutes evidence against a unit root?
Using the conventional Jeffreys scale: BF01 < 1/3 is moderate evidence against the unit root (in favour of stationarity), BF01 < 1/10 is strong evidence. Values between 1/3 and 3 are inconclusive. Context and prior sensitivity should always accompany these thresholds.
Can I use this test with structural breaks?
Standard Bayesian PP tests do not model structural breaks. If the series has a known or suspected break (e.g., financial crisis, policy change), use a Bayesian Zivot-Andrews test or Fourier-based unit root approach that accommodate shifts in the deterministic component.
Sources
- 1.Phillips, P. C. B., & Perron, P. (1988). Testing for a unit root in time series regression. Biometrika, 75(2), 335-346.
- 2.Sims, C. A., & Uhlig, H. (1991). Understanding unit rooters: A helicopter tour. Econometrica, 59(6), 1591-1599.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Bayesian PP unit root test. ScholarGate. https://scholargate.app/econometrics/bayesian-pp-unit-root-test