The RATS Software Forum

Hello Tom,

Just a quick question

Do you have any reason/suggestion why I should not do GARCH BEKK using more than 2 variables? im using 53 variables with each observations only 140. Understanding the forum i listed below, I chose to estimate it pairwise (bivariate)

I have read the forum and found your suggestions on:

Big model because one asks with 6 variables --> https://estima.com/forum/viewtopic.php?f=11&t=2617 and if i am not mistaken there are also other forum that pointed out the fault in estimating bekk with many variables.


From this, may i ask why bekk should not be estimated with many variables? and is there any rule of thumb to determine the number of variables should be included in the estimation?

Many thanks

It's very rare to see a BEKK with more than four variables---it just doesn't seem to work well beyond that. (CC/DCC are typically used instead for larger sets of variables).

In our paper (written below) published in Energy Economics, we used four variables. Usually, even in the four-variate case, estimations take too long until it achieves the convergence due to the curse of dimensionality in the BEKK model. Things become more complicated if the estimations were done within the loop in the context of moving or recursive windows.

Hasanov, A.S., Do, H.X., and Shaiban, M. (2016). Fossil fuel price uncertainty and feedstock edible oil prices: Evidence from MGARCH-M and VIRF analysis. Energy Economics, 57, pp. 16-27.