Cointegration
Find a stationary relation among trending series
Cointegration describes non-stationary series whose particular linear combinations are stationary. Engle-Granger estimates a long-run relation and tests its residuals using cointegration-specific critical values, usually focusing on one relation. Johansen estimates cointegrating rank and multiple vectors jointly within a VAR system. Both separate shared long-run movement from short-run adjustment without establishing a causal economic mechanism.
Choose this when theory suggests a stable long-run relation among integrated series and levels contain information that differencing alone would discard. Integration order, deterministic terms and structural stability require prior investigation.
Strengths
- Preserves long-run information alongside short-run dynamics
- Johansen accommodates multiple equilibrium relations
Limitations
- Rank and vector estimates are sensitive to lag and trend choices
- Breaks can mimic or destroy stable relations
Know the boundary
Cointegration is a stationary long-run combination, not merely high correlation between trending series. Engle-Granger and Johansen have different system scope.