VECM-GARCH Model

[faaequah13]

as i am beginner for rats and working on two variables (stock return and currency exchange rate) and they are cointergated and i have run VECM in eviews, but i want to run vecm bekk model on it, but i am not getting codes regarding the vecm and vecm bekk model,kindly help me out by helping the codes.

[TomDoan]

The VECMGARCH.RPF example does a GARCH model with the VECM mean model. (It's possible that you're looking at it more like a VECM with a GARCH error process, but technically, the "GARCH" part of it controls how it's estimated).

Are you saying that stock returns and the exchange rate are cointegrated? That doesn't sound right. Stock returns are almost always very I(0), close to white noise and thus can't be cointegrated with anything.

[faaequah13]

i am using companies closing price instead of index and among the 10 companies almost four companies are co-integrated with different exchange rates. thats why i need VECM, and thats true stock return are never stationary at level but in my case i found.

[faaequah13]

I have one more query regarding the lag, in some cases i found 0 lag for var, but in that case we cant run var model and same not even GARCH, in that case how i can estimate volatility spillover, please kindly help me out for your kind reply.

[TomDoan]

i am using companies closing price instead of index and among the 10 companies almost four companies are co-integrated with different exchange rates. thats why i need VECM, and thats true stock return are never stationary at level but in my case i found.

You found that the (log??) closing price is cointegrated with an exchange rate? That doesn't seem to make much sense---if a stock price is stationary when measured in another currency, there are massive arbitrage opportunities.

[TomDoan]

I have one more query regarding the lag, in some cases i found 0 lag for var, but in that case we cant run var model and same not even GARCH, in that case how i can estimate volatility spillover, please kindly help me out for your kind reply.

If you specify the model in returns then a zero lag is not at all unreasonable. (In fact, theoretically, it should be zero.) If you do the model in log prices, that shouldn't happen since the lags have to be at least one.

[faaequah13]

like i have two variable one is closing price of infosys compony and other is exchange return won/dollar, both are stationary at level. according to you , am i doing something wrong with my variables. on other hand i have one variable is infosys closing price and inr/usdollar, infosys is stationary at level while inr/us dollar integrated at I(1), for the cointegration in above case i had applied johansen cointegration but for later case ARDL model. as the case somewhere it found cointegration and somwhere no co integration, in whole senerio u mean something wrong.

[faaequah13]

in eviews i was using log series in that case i found lag 1, but rats deals with return series that why it is showing 0 lag, as per my knowledge VAR is run on raw data series not on return series.

[faaequah13]

if mqstat is significant then and how we rectify it, what we interrupt it. as its null hypothesis is there is no auto correlation?

[TomDoan]

in eviews i was using log series in that case i found lag 1, but rats deals with return series that why it is showing 0 lag, as per my knowledge VAR is run on raw data series not on return series.

If you're trying to do cointegration analysis, you would use log data (100 x log(x) is usually the best transformation to use). You would not use returns, which is basically log differences---that's misspecified in the presence of cointegration.

[TomDoan]

if mqstat is significant then and how we rectify it, what we interrupt it. as its null hypothesis is there is no auto correlation?

The null in @MVQSTAT is no serial correlation. Sometimes it will be fixed by added lags, but not necessarily. If there's a clear break in the series, it's unlikely that you can get a single model that will give acceptable behavior. Also, if the data set is long enough, you might be running into a situation where the test is statistically "significant" but practically meaningless. See "Diagnostics in Large Data Sets".

[faaequah13]

as u shared this link (https://estima.com/forum/viewtopic.php? ... 13&p=14222), i want to ask you why there is assymmetric term in garch code, as i do not need to watch asymmetric effect. can u explain it?

[TomDoan]

as u shared this link (https://estima.com/forum/viewtopic.php? ... 13&p=14222), i want to ask you why there is assymmetric term in garch code, as i do not need to watch asymmetric effect. can u explain it?

From the comments:

*
* Example of some of the calculations done in Pardo and Torro(2007),
* "Trading with asymmetric volatility spillovers", JBFA, vol.
* 34(9-10), 1548-1568.
*

You would need to read Pardo and Torro to see why they used asymmetry.

[TomDoan]

hi
hope you are doing well, when i was looking for vecm bekk i found this link( https://estima.com/forum/viewtopic.php?f=11&t=2390) in this link i have found these codes

<<code deleted by moderator>>

*
can u help me out, what is the problem with the coding?

1. You seem to have stray characters at the end of the instruction:

set ect = %eqnprj(ecteq,t). I

It should be

set ect = %eqnprj(ecteq,t)

2. That's a 5 year old thread. You can now use a SYSTEM with an ECT in a GARCH instruction as is done in the VECMGARCH.RPF example. If you need a BEKK without asymmetry, leave the ASYMMETRIC option out. VECMGARCH has about three pages worth of notes that go through pretty much every step in detail.

[faaequah13]

apart of vecm bekk, can we run vecm cc and dcc model? if yes what will be the code.