ATLASResearch
methods
Quantitative/ Analysis

ARDL bounds testing

Test a long-run relation with mixed integration orders

ARDL models combine lags of an outcome with current and lagged regressors. Bounds testing assesses whether lagged level terms form a long-run relation when regressors may be I(0), I(1), or a mixture. Critical bounds depend on deterministic specification; an intermediate statistic is inconclusive. Standard bounds procedures exclude I(2) variables and do not remove endogeneity concerns.

WHEN IT FITS

Use for a theory-led single-equation long-run relationship when integration orders are mixed but no variable is I(2). The time series must support plausible lag choices and a stable adjustment process.

Strengths

  • Allows a mixture of I(0) and I(1) regressors
  • Provides short-run and long-run parameterisations in one framework

Limitations

  • An intermediate bounds result can be inconclusive
  • Small samples, breaks and endogenous regressors can undermine interpretation

Know the boundary

ARDL is not a licence to skip integration checks or claim causality from a long-run coefficient.

USED ACROSS
Banking & financeBusiness & MBA