Johansen procedure requires all the variables to be in the same integration order i.e., I(1). Therefore, ARDL bounds testing is recommended for the variables with mixture of integration order instead.

Nevertheless, having an I(2) variable is extremely rare case in practice.

Try to test for distribution for each component separately.

-How does government spending on health behave? What kind of distribution does the data show?

-How does private spending on health behave? What kind of distribution does

the data show?

-Does government spending on health and private spending on health manifest same or similar pattern? distribution?

This may be a case of combining apple and orange. The healthcare spending by the government is part of G in the GDP equation. The health care spending by non-government is the sub-part of the C in GDP. For the government spending,

the government budget is something quite predictable. However, healthcare

spending as part of private consumption may not follow such predictable pattern.

This may explain why they behave differently in your combined model. test each group of data first may give clearer picture, then decide on modeling approach.

Thank you all of you for your time and efforts.