Lenny D. answered • 05/11/19
Former professor of economics at Tufts University
If the variables exhibit growth, There will be a compounding effect effect and you will be trying to fit a line to a non linear relationship. You with want to use Log differences or percent changes. A simple example.
Suppore you put $100 in a bank account that earned 5% per year The balance in Your Account will grow geometrically. The Account will Grow by $5 in year 2 it will grow by 5.25. In Year 9 the Account will be worth $155.1. In year 10 the account will be worth 162.9. Simple differencing will lead to model misspecification, Coefficient bias and introduce heteroskedacity. Otherwise there is no difference.
