Page 38 - Regression Guideline for AMC
P. 38
MAPE
(Mean Absolute Percentage Error)
• Interpreta6on of the MAPE for a given appraisal regression is straighyorward. For example, if the last sales price of a property was $200,000 and the regression model MAPE was 5%, this means that the model predic6on would be in the range of $190,000 to $210,000 ($200,000 +-‐ $10,000 or 5%). Smaller MAPE scores represent be.er fiRng models and more accurate predic6ons.
• Unfortunately, as with R-‐square, there is no agreement on what is an acceptable or “good” MAPE because the percentage tolerance will vary with the context. One rule of thumb is that a MAPE in the range of 20% to 30% is acceptable, 10% to 20% is good, and below 10% is excellent.
• In analyses conducted thus far using the SAVVI Analy6cs regression process on different MLS lis6ngs, we have found that the main determinant of model quality has been (again!) the quality and completeness of the input data. With complete, high quality data, the regression models typically have had a MAPE under 10% and an R-‐square over 50% to 60%, sugges6ng excellent performance and enhanced predic6ve valua6on of property worth.
38
