MerQuR+Statistics Glossary

Statistics Glossary → Johansen trace and maximum-eigenvalue tests

Johansen trace and maximum-eigenvalue testsconcept

Maximum likelihood tests, within the vector error correction model framework, of how many cointegrating (long-run equilibrium) relations link several non-stationary time series.

The vector error correction model is written Δy_t = Π y_{t-1} + Σ Γ_i Δy_{t-i} + ε_t; the rank r of the matrix Π is the number of independent cointegrating relations tying the series together. If r = 0 there is no long-run relationship; if r equals the number of series, the series are already stationary. Johansen's method estimates the eigenvalues of Π and tests the rank through them.

The trace test examines the null 'rank is at most r' against 'greater than r', and the maximum-eigenvalue test examines 'rank is r' against 'rank is r + 1'. The tests are applied in sequence starting from r = 0, and the first r at which the null is not rejected is accepted. The critical values depend on how the deterministic terms enter. In small samples the tests tend to reject too often and are sensitive to the choice of lag length; the two tests sometimes suggest different ranks. Unlike the two-step Engle-Granger method, it handles several relations at once and does not depend on the choice of normalization variable.

An everyday example

The prices of the same product in two neighbouring cities fluctuate separately in the short run, but if the gap grows too wide, traders buy in the cheaper city and sell in the dearer one, and the prices converge again. The Johansen test can show a cointegrating relation (r = 1) between these two price series.

In MerQur

'Method' (Johansen + Engle-Granger / Johansen / Engle-Granger), 'VECM rank r' (-1 = 5% decision of the trace test) and 'Normalization variable' in the Cointegration analysis.

Analyses where this concept appears

Cointegration (Johansen / Engle-Granger)

Related concepts

Vector error correction model (VECM)Engle-Granger two-step method