[AstralStorm]

Weak correlations are fine, as long as they're consistent and statistically significant.

This generally means that there is a factor causing a misfit with the simplistic base model. It's fine if the factor is properly identified, which it is not in this case. I think a logarithmic fit would give much higher R2 than linear. (Logistic fit is commonly used for prices and salaries.)

This problem goes away with multifactor analysis as you see correlation in errors or not, or using nonlinear least squares, NLS.