Autoregressive Distributed Lag Model (ARDL)

The autoregressive distributed lag (ARDL) model regresses a series on its own lags and on current and lagged values of explanatory series, giving short-run effects and long-run multipliers.

자기회귀 분포시차(ARDL) 모형은 시계열을 자신의 시차와 설명 시계열의 현재·시차 값에 회귀해 단기 효과와 장기 승수를 함께 구하는 모형이다.

What is the long-run effect of interest rates on housing investment once lagged adjustment is allowed?

시차 조정을 고려했을 때 금리가 주택투자에 미치는 장기 효과는 얼마인가요?

ARDL selects the lag orders by an information criterion and estimates the dynamic regression by ordinary least squares. The long-run multiplier for each regressor is the sum of its lag coefficients divided by one minus the sum of the autoregressive coefficients.

When to use Autoregressive Distributed Lag Model (ARDL)

• Effects of a regressor spread over several periods.
• You need both short-run and long-run effects.
• Regressors may be a mix of I(0) and I(1) series, provided a long-run relation is established separately, for example with a bounds test.

When to use something else

• Several series affect each other simultaneously; use Vector Autoregression (VAR).
• Variables are I(2) or you need several cointegrating relations; use a Vector Error Correction Model (VECM).

Assumptions and requirements

• Residuals are serially uncorrelated.
• Stable dynamics (sum of AR coefficients below one).
• Regressors are weakly exogenous.
• No I(2) variables.

What the analysis reports

• Selected lag orders and information criteria
• Short-run coefficients with standard errors
• Long-run multipliers with delta-method standard errors
• Residual autocorrelation check

How to report it

An ARDL([value], [value]) model selected by AIC gave a long-run multiplier of [value] (SE = [value], p = [value]).

Related methods

• Engle-Granger Cointegration Test
• Vector Error Correction Model (VECM)
• Cross Correlation Analysis (CCF)

Run Autoregressive Distributed Lag Model (ARDL) on Quorant