Panel VECM Heteroskedasticity and Non-Normality

[JVallid]

Hi,

I am testing the causal relationship between financial development and economic growth within a VAR/VECM framework. I have 4 endogenous variables: GDP, 2 financial development indicators and a control. My panel consists of 84 countries between 1975 - 2014.

My procedure is as follows:
1) Unit Root Tests
- conclude all I (1)
2) Run unrestricted VAR
- Correct Lag Specification
- Diagnostic Tests: serial correlation, hetero and non-normality. It fails the latter 2.
- Check non-spurious regression
- Impulse Response
3) Johansen Cointegration Test
- 1 Cointegrating equation
4) VECM @ lags specified in (2)
- Coefficient Analysis
- Block Exogeneity Test / Granger Causality
- AR Roots Graph

Are my results valid given the presence of heteroskedasticity, where I believe it is caused by cross-country differences? If not, how may I correct this in EViews 9?

Also with regards to non-normality, am I correct in believing this is non-madatory given symmetrical distribution / large sample (3000 obs).

Many thanks,

J