Is the multicollinearity problem a characteristic of the population or the sample?

For example, if you are trying to estimate the effect of the unemployment rate and interest rates on stock prices, do you have a multicollinearity problem if the unemployment rate and interest rates tend to be correlated in general or if they happen to be correlated in the particular data you have?