Convergence in Asymmetric VECM-MGARCH-BEKK Model

[Kiplimo]

Hi
I'm attempting to fit a market data to estimate asymmetry and transmission of volatilities between 8 food markets with 160 monthly data points. The challenge however is that when I estimate the Asymmetric VECM-MGARCH-BEKK model I don't get convergence but when I run the model without asymmetry there is convergence. Similaly when I run the model with constant in the mean model and without the lags there is convergence.

Please help.

*
OPEN DATA "C:\Users\Ben\Dropbox\ABCWorking Papers\Price-Volatility\DATA\WFP_2019DEC1_Kenya_FoodPricesData.csv"
CALENDAR(PANELOBS=163,M) 2006:1
DATA(FORMAT=PRN,ORG=COLUMNS) 1//2006:01 1//2019:06 YEAR MONTH ELDW KSMW MBSW NRBW LDWR KTIR MDRR MSBR
* Check data
TABLE
* Normalize both to log relative to the first entry
SET LNRBW = LOG(NRBW)
SET LMBSW = LOG(MBSW)
SET LELDW = LOG(ELDW)
SET LKSMW = LOG(KSMW)
SET LLDWR = LOG(LDWR)
SET LKTIR = LOG(KTIR)
SET LMDRR = LOG(MDRR)
SET LMSBR = LOG(MSBR)
SET DLMSBR = LMSBR - LMSBR{1}
SET DLNRBW = LNRBW - LNRBW{1}
SET DLMBSW = LMBSW - LMBSW{1}
SET DLELDW = LELDW - LELDW{1}
SET DLKSMW = LKSMW - LKSMW{1}
SET DLLDWR = LLDWR - LLDWR{1}
SET DLKTIR = LKTIR - LKTIR{1}
SET DLMDRR = LMDRR - LMDRR{1}
@varlagselect(lags=10,crit=aic)
#LELDW LKSMW LMBSW LNRBW LLDWR LKTIR LMDRR LMSBR
@varlagselect(lags=10,crit=sbc)
#LELDW LKSMW LMBSW LNRBW LLDWR LKTIR LMDRR LMSBR
*
compute nlags=1
*
@johmle(lags=nlags,det=constant,cv=cv)
#LELDW LKSMW LMBSW LNRBW LLDWR LKTIR LMDRR LMSBR
equation(coeffs=cv) ecteq *
#DLELDW DLKSMW DLMBSW DLNRBW DLLDWR DLKTIR DLMDRR DLMSBR
set ect = %eqnprj(ecteq,t)
*
system(model=vecm)
variables DLELDW DLKSMW DLMBSW DLNRBW DLLDWR DLKTIR DLMDRR DLMSBR
lags 1
det ect{1}
end(system)
*
estimate(resids=vecmresids)
*
@mvarchtest(lags=1)
#vecmresids
*
* Estimate an asymmetric BEKK GARCH model.
*
garch(model=vecm,mv=bekk, asymm, stdresids=stdu,hmatrices=hh,rvectors=rr,$
pmethod=simplex,piters=10,method=bfgs, robust, iters=500, dist=t,vechmat=vechcomps)

[TomDoan]

1. That's not a panel data set; just a standard monthly data set.
2. That's not the correct way to set up a VECM: the ECT is on the levels, not the differences. Also RATS version 10 permits you to input the model to GARCH directly using a VECM SYSTEM.
3. Why would there be just one cointegrating vector involving eight food prices?

However, getting to the original question, that's a very complicated model for a very small amount of data. I'm actually surprised that you can get a (symmetric) BEKK model to fit---BEKK rarely works well beyond about four or five series, and even then it generally requires a lot more data than you have.