Asymmetric Error Correction Models - some Comments and Improvements
In this essay the method to estimate asymmetric relations with error correction models by splitting the error correction term u[t-1] in [+] u[t-1] and [-] u[t-1] , proposed by Granger and Lee (1989), is proven. The results can be summarized in the following aspects:
- The sign-depending splitting of the error correction term does not enable the quantification of asymmetric effects.
- Symmetric as well as asymmetric error correction models are not completely specified, if the x-variable with level data and the time lag [t-1], on which the co-integrating regression is based, is not included in the function. The neglecting leads to considerably distorted parameters.
- The separate calculation of the error correction terms for the variables [+] y[t-1] and - y[t-1] , derived by splitting the y[t-1] -variable, proposed by the author,proves to be an exact approach to quantify asymmetric relations with error correction models.
CramonEngleGrangerLeeasymmetric error correction modelscointegrationreversibilitysymmetric error correction models