Glossary
Augmented Dickey-Fuller (ADF) Test
The ADF test is a widely used unit-root test that augments the basic Dickey-Fuller regression with lagged first differences of the dependent variable to account for serial correlation in the error term. Said and Dickey (1984) showed that approximating an ARMA process with a su...
Definition
The ADF test is a widely used unit-root test that augments the basic Dickey-Fuller regression with lagged first differences of the dependent variable to account for serial correlation in the error term. Said and Dickey (1984) showed that approximating an ARMA process with a sufficiently long autoregression yields a test statistic whose limiting distribution matches the standard Dickey-Fuller tables. The ADF test is typically the first step in time-series analysis to determine whether differencing is needed to achieve stationarity.
Why It Matters
Before estimating any time-series model, researchers must establish whether the data are stationary. Running regressions on non-stationary series can produce spurious results with misleadingly high R-squared values and significant t-statistics. The ADF test provides a formal, widely accepted procedure for this critical diagnostic step and is a prerequisite for cointegration analysis.
Example
A researcher studying the Turkish lira / US dollar exchange rate suspects the series has a unit root. Running an ADF test with a trend and four lagged differences yields a test statistic of -2.1, which is above the 5% critical value of -3.45. The null hypothesis of a unit root cannot be rejected, confirming that the series is non-stationary and must be differenced before further modelling.
Related Terms
Software Notes
- SPSS: Analyze > Forecasting > Unit Root Test; select ADF and specify lag length or use automatic selection.
- R: Use
adf.test()from thetseriespackage orur.df()from theurcapackage for more detailed output including lag selection. - Stata: Use
dfuller variable, trend lags(4)for an ADF test with trend and 4 lags. Usedfuller variable, trendfor automatic lag selection via the Ng-Perron criterion.