When dealing with financial datasets such as stock or house prices, practitioners generally care more about the relative error $\frac{y - \hat{y}}{y}$ than the absolute error $y - \hat{y}$. For instance, a prediction error of $100,000 is massive when estimating the price of a typical $125,000 rural house, but the exact same absolute error represents a stunningly accurate prediction for a median house price exceeding $4 million in a wealthy neighborhood.