ARDL & Cointegration Services
Economic and financial time series trend over time, and regressing one on another naively can produce a spurious result. The ARDL bounds-testing approach estimates short-run dynamics and long-run (cointegrating) relationships together—robustly, even when variables are integrated of different orders.
Cointegration means two or more non-stationary time series share a stable long-run relationship, so they move together over time even while diverging in the short run. The autoregressive distributed lag (ARDL) bounds-testing approach estimates the short-run dynamics and the long-run relationship in one model, and applies whether the variables are stationary, non-stationary, or a mix—but not if any is integrated of order two.
What ARDL and cointegration analysis do
Most macro-economic and financial series—GDP, prices, exchange rates, credit, stock indices—are non-stationary: they trend over time rather than fluctuating around a fixed mean. Regressing one trending series on another can generate a spurious relationship—an impressive-looking result that reflects shared trending rather than any genuine link. The concept of cointegration resolves this: two or more non-stationary series are cointegrated if a linear combination of them is stationary, meaning they share a stable long-run equilibrium and cannot drift arbitrarily far apart, even though each wanders in the short run.
The autoregressive distributed lag (ARDL) bounds-testing approach is a widely used and flexible way to analyse such relationships. Its central advantage is that it does not require all variables to be integrated of the same order—it works whether the series are stationary, non-stationary, or a mixture (I(0) and I(1)), provided none is I(2). From a single ARDL model it delivers both the long-run relationship (the equilibrium) and the short-run dynamics via an error-correction representation, whose error-correction term measures how quickly deviations from equilibrium are corrected. This makes ARDL especially suited to the small samples and mixed integration orders common in applied economics and finance.
When to use it, and the NARDL/QARDL extensions
Use ARDL and cointegration analysis when your data are time series with plausible long-run relationships—questions about how variables move together over time and how quickly they return to equilibrium after a shock. It suits the moderate sample sizes typical of country-level and financial studies, and its flexibility on integration order removes a common obstacle of older cointegration methods.
Two important extensions widen its reach. The nonlinear ARDL (NARDL) allows asymmetric effects—where increases and decreases in a variable have different impacts (for example, prices responding differently to rising versus falling costs). The quantile ARDL (QARDL) lets the relationship differ across the distribution of the outcome. For the classical framework, the Johansen approach handles systems where all variables are I(1) and there may be multiple cointegrating relationships, and FMOLS/DOLS provide efficient long-run estimators. We select among ARDL, its nonlinear and quantile variants, and system methods based on the integration properties, the sample, and the economic question.
ARDL and its relatives
| Method | Best for | Integration order |
|---|---|---|
| ARDL bounds test | Single long-run relationship, small samples | Mix of I(0) and I(1); no I(2) |
| NARDL | Asymmetric (sign-dependent) effects | Mix of I(0) and I(1) |
| QARDL | Effects varying across the distribution | Mix of I(0) and I(1) |
| Johansen (VECM) | Systems with multiple relationships | All I(1) |
| FMOLS / DOLS | Efficient long-run coefficient estimates | Cointegrated I(1) |
Doing time-series econometrics credibly
Rigorous time-series work begins before the main model. Unit-root and stationarity testing establishes each series’ order of integration—which method is even valid depends on it, and ARDL specifically requires confirming that no variable is I(2), since the bounds test is invalid in that case. Because standard unit-root tests can be misled by structural breaks (shifts in level or trend from crises, policy changes, or regime shifts), tests that allow for breaks are used where the data warrant, and breaks are modelled rather than ignored.
Once the model is estimated, its diagnostics decide its credibility: the ARDL bounds test itself (comparing the F-statistic to the I(0) and I(1) critical bounds) to confirm a long-run relationship exists, and residual checks for serial correlation, heteroskedasticity, functional form, and normality—along with parameter-stability tests (CUSUM/CUSUMSQ). A significant, negative error-correction term confirms a stable adjustment to equilibrium. We report the integration orders, the bounds test, the long- and short-run estimates, and the full diagnostic battery, because a cointegration result without these is not evaluable.
Establish the integration order first—and rule out I(2). Time-series methods are only valid for particular integration orders, and ARDL bounds testing breaks down if any variable is I(2). Unit-root testing (allowing for structural breaks) and the full diagnostic set are what separate a genuine long-run relationship from a spurious regression.
Software
We deliver ARDL, NARDL, QARDL, and system cointegration methods in established, reproducible tools—EViews, Stata, and R—with unit-root and break testing, the bounds test, error-correction estimation, FMOLS/DOLS where appropriate, and the full diagnostic and stability battery, with versioned code.
How we deliver an ARDL / cointegration study
This work sits within our wider econometrics practice—so integration orders are established first, the estimator matches the data, and every diagnostic is reported.
We start by testing each series’ order of integration (allowing for structural breaks where relevant) and confirming ARDL is appropriate—in particular that no variable is I(2). We select the model (ARDL, NARDL, QARDL, or a system approach) to fit the integration properties and the economic question, run the bounds test for a long-run relationship, and estimate the long-run and error-correction dynamics.
Reporting sets out the unit-root and break results, the bounds test, the long- and short-run estimates with the error-correction term, and the full diagnostic and stability checks—so the relationship can be judged, not just asserted.
You receive the long-run (cointegrating) estimates and the short-run error-correction dynamics with the adjustment speed, the bounds-test and unit-root results, the diagnostic and CUSUM stability checks, any asymmetry (NARDL) or quantile (QARDL) results, and reproducible analytical code and analysis-ready files (where appropriate and permitted).
ARDL & cointegration across Management & Allied Studies
Long-run relationships among trending series are central to economics and finance—so ARDL and cointegration analysis are core tools across the quantitative disciplines we serve.
Economics & Public Policy
Long-run relationships among macro variables—growth, inflation, trade, energy, and policy—and how fast they adjust to equilibrium. A core ARDL setting.
Finance & Financial Markets
Long-run links among prices, rates, exchange rates, and indices, and the short-run dynamics of adjustment.
Energy & Environmental Economics
Relationships among energy use, emissions, growth, and prices—often with asymmetric (NARDL) effects.
Banking & Monetary Economics
Long-run dynamics of credit, money, and policy variables, with error-correction adjustment.
Development & International Economics
Country-level long-run relationships where moderate samples and mixed integration orders suit ARDL.
Accounting & Corporate Finance
Time-series relationships among financial aggregates and market variables over the long run.
ARDL & cointegration: common questions
Estimating long-run relationships in time-series data?
If your variables trend over time and you need the long-run relationship and its adjustment dynamics, ARDL and cointegration analysis deliver both—with integration orders established, the bounds test applied, and the full diagnostics reported.