Error /(Equilbrium) Correction Model Question

[opticchickenmcnugget]

Dear E-Views Forum Members,

I'm using E Views version 11.

I have a model where I am looking to estimate the following:
- Speed of adjustment
- Short run elasticity
- Long run elasticity

I've had a suggestion to run this model
d(y) c d(x(-1)) d(x(-2)) d(y(-1)) d(y(-2)) (y-x)(-1)
- How would i look to interpret a model that takes a lagged difference as one of the independent variables in the context of an error correction model.

KEY ISSUE HERE!!
- The differenced version of my indendent variables are insignificant!
WHAT IF...?
- I took the log to surpress the movement a little bit, or would it be more prudent to remove a time period?

Whereby:
Y Denotes the dependent variable
X denotes the independent variable
(Y-X) Is the difference between the Y and X variable (As a substitute for running the y = c + x regression and storing the residuals - since the financial data I'm using is conducive to do so).
- However the problem here is the fact that my independent variable, with 1 lag, are statistically insignificant - however my speed of adjustment is significant.

I've developed a different model:
d(y) c d(x) d(x(-1)) d(y) d(y(-1)) (y-x)(-1)
Whereby there is no heteroskedasticity or serial correlation
- And my elasticity coefficients are significant so I can actually make inference.

My question is:
- Should I be using model 2, instead of model 1?
- When running an ECM model, how important is it to have homoskedasticity and no serial correlation? Does it completely invalidate the test? Or is there some leeway?

Thanks in advance for any help!

Happy New Decade,
Opticchickenmcnugget