Time-Varying Parameter ARDL Bounds Test
Also known as: TVP-ARDL bounds test, time-varying ARDL cointegration, TVP bounds testing approach, dynamic ARDL bounds test
The time-varying parameter ARDL bounds test extends the classic Pesaran-Shin-Smith (2001) bounds testing framework by allowing regression coefficients to evolve continuously over time. It detects whether a long-run cointegrating relationship between variables exists and whether that relationship has been stable or shifting across the sample period.
Key highlights
- Handles regressors of mixed or uncertain integration order (I(0)/I(1)) without pre-testing, inheriting the robustness of the bounds approach.
- Reveals not just the existence but the temporal evolution of cointegrating relationships, capturing structural breaks and gradual parameter drift endogenously.
- More flexible than single-break or rolling-window methods because the coefficient path is estimated continuously rather than in discrete sub-samples.
- The state-space formulation provides a principled probabilistic framework with confidence bands for time-varying estimates.
- Can detect whether cointegration appeared only in a sub-period, giving richer policy or forecasting insights.
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 TVP-ARDL bounds test when you suspect a long-run relationship between two or more time-series variables but also expect that the strength or sign of the relationship has changed over time — common in macroeconomic, energy-economics, or financial data spanning policy shifts, crises, or regime changes. It is appropriate when the integration order of regressors is unknown or mixed I(0)/I(1). Avoid it with very short series (fewer than ~60 observations) because the Kalman filter needs enough data to identify time-varying paths reliably, and avoid it when theory strongly predicts a stable structure — a constant-parameter ARDL will then be more efficient.
Strengths & limitations
- Handles regressors of mixed or uncertain integration order (I(0)/I(1)) without pre-testing, inheriting the robustness of the bounds approach.
- Reveals not just the existence but the temporal evolution of cointegrating relationships, capturing structural breaks and gradual parameter drift endogenously.
- More flexible than single-break or rolling-window methods because the coefficient path is estimated continuously rather than in discrete sub-samples.
- The state-space formulation provides a principled probabilistic framework with confidence bands for time-varying estimates.
- Can detect whether cointegration appeared only in a sub-period, giving richer policy or forecasting insights.
- Requires a relatively long time series (ideally 60+ observations) for the Kalman filter to estimate stable hyperparameters.
- Results depend on the assumed state-equation variance Q; misspecification of Q can oversmooth (masking real breaks) or undersmooth (producing spurious drift).
- Computationally more demanding than standard ARDL and requires specialised software or custom coding.
- Critical value tables from Pesaran et al. (2001) were derived for constant-parameter models; their applicability to time-varying settings is not fully theoretically established.
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 is TVP-ARDL different from rolling-window ARDL?
Rolling-window ARDL re-estimates a constant-parameter model on successive windows of data, producing step-wise changes in coefficients. TVP-ARDL estimates a single state-space model in which all coefficients evolve smoothly each period via the Kalman filter, providing a more continuous trajectory and formal probabilistic uncertainty bands.
Do I still need to test for unit roots before applying TVP-ARDL?
Not strictly. One of the key advantages of the bounds approach (inherited from Pesaran et al. 2001) is that it is valid for regressors that are I(0), I(1), or a mix, so preliminary unit root testing is not required for the bounds F-test to be valid.
How do I choose the state-equation variance Q?
The preferred approach is maximum likelihood estimation of Q alongside the other hyperparameters of the state-space model. Sensitivity analysis — comparing coefficient paths under a range of Q values — is also recommended to assess how heavily the results depend on this choice.
What software can estimate TVP-ARDL?
There is no single standard package. Researchers typically implement it in R (using dlm, KFAS, or custom code), Python (statsmodels state-space framework), or MATLAB. Some applied papers provide replication code that can be adapted.
Can the bounds test be applied period by period with time-varying coefficients?
Yes, the time-varying level-relationship coefficients can be tested at each point in time (or averaged over sub-periods) using the Pesaran et al. critical value bands, though the theoretical justification is approximate and results should be interpreted with appropriate caution.
Sources
- 1.Pesaran, M. H., Shin, Y., & Smith, R. J. (2001). Bounds testing approaches to the analysis of level relationships. Journal of Applied Econometrics, 16(3), 289–326.
- 2.Chow, G. C. (1960). Tests of equality between sets of coefficients in two linear regressions. Econometrica, 28(3), 591–605.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Time-varying parameter ARDL bounds test. ScholarGate. https://scholargate.app/econometrics/time-varying-parameter-ardl-bounds-test