Traditionally, if $x_t$ and $y_t$ are both $I(1)$, they are cointegrated when there exists some linear combination $z_t=y_t-$$\gamma$ $x_t$ such that $z_t$ is stationary or $I(0)$.

My question is if I can claim that $x_t$ and $y_t$ are cointegrated (perhaps a special case) when $z_t=y_t-$$\gamma_t$$x_t$ where $z_t$~$I(0)$? Here, I allow the cointegrating vector (coefficient) $\gamma_t$ to vary over time.

In this situation, $\gamma_t$ adapts to covariance structure between the $x_t$ and $y_t$ if said cov structure evolves over time. Since $z_t$ is still stationary, is there some term that is appropriate I can use in place of traditional cointegration? I was thinking 'time-varying cointegration' or something along those lines.


1 Answer 1


Yes, the concept of non-linear cointegration has been introduced in the litterature, and there are various authors who adressed the case of time varying cointegration. Look for example at:

  • Bierens, Martins, Time Varying Cointegration, Econometric Theory, 2010, Page 1 of 38
  • Park, Hahn, Cointegrating Regressions With Time Varying Coefficients, Econometric Theory, 15, 1999,

There is also a strand of papers looking at very flexible (time) varying models in a non-parametric way, so-called functional cointegration, see among others:

  • Xiao, Functional-coefficient cointegration models, Journal of Econometrics 152 (2009) 81–92

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.