The RATS Software Forum

Dear Tom,
I am looking for the code of the following paper by Xiao (2009) entitled "Quantile cointegrating regression" available at
https://www.sciencedirect.com/science/a ... 7608002224
Is it possible to estimate with RATS?
Best Regards

Isn't that just RREG applied to a Stock-Watson DOLS style regression?

yes could you please show me how to modify this estimator to obtain quantile coefficients?

Is it possible to include those methodology to the recently announced unit root and cointegration course since it became popular in financial econometrics literature?

ege_man wrote:yes could you please show me how to modify this estimator to obtain quantile coefficients?

You use RREG rather than LINREG.

ege_man wrote:Is it possible to include those methodology to the recently announced unit root and cointegration course since it became popular in financial econometrics literature?

I don't see 44 citations in 10 years as showing something extremely popular, particularly since some of them are clearly using some other form of quantile analysis and simply citing this one as another use of that general concept.

I just scanned it, but I'm not even seeing what the point of this is. The underlying idea is that the cointegrating vector is somehow "time-varying". OK. Possible, though not easy to handle. But what does quantile estimation have to do with that? Quantile estimation is designed to get something that is more robust to outliers than least squares. Now, I can see where a robust estimator in a DOLS regression might make some sense in a price on dividends regression which presumably has some very fat tails. Whether this is a good approach to handle that doesn't seem to be addressed.

Dear Tom,
Thank you for your prompt reply.
I know it is not easy to handle a time-varying structure in cointegrated models and also know there are some residual based tests like Gregory and Hansen (1995) or Maki (2012) allowing for endogenous breaks in the cointegrated system. Do you have better suggestion for a convenient methodology?
Best Regards

Note well that those are not ways to estimate a time-varying cointegrating vector, but a way to test for cointegration allowing for breaks in the cointegrating vector. CATS has some recursive tests which can be used to reject a time-invariant cointegrating vector (or cointegrating rank); again, it doesn't give a specific alternative, just shows that the data don't behave properly for time-invariant cointegration.