Models that are based on an adequate number of cases in the original data set have low shrinkage and are described as having high stability because their predictive accuracy changes very little when applied to a new set of data. If model stability is not Good then the number of case need to be added
It is also important to select properties that are not too dissimilar to the subject property in terms of the range of sales prices, year of construction, location and other property characteristics although we do need some variation, otherwise there is nothing to model. The need for some but not too much variation in the sample of properties to be analyzed (the “Goldilocks principal”) is the balancing trick required for selecting properties to model.
R-square is the most commonly used metric of regression model fit but not the only one. Because it’s value is affected by both the strength of association as well as prediction accuracy, other statistics are used to evaluate model fit and, particularly, the accuracy of the estimated values. If your model stability and MAPE don’t have good ratings your R-square could give you a false reading
The MAPE summarizes the residual scores for a particular regression model and the sample on which it was derived and provides an average residual (error)score based on all the properties evaluated by the model. In appraisal analysis, MAPE reflects the average amount the predicted values of the comparables deviated from their last known sales prices.
As the old saying goes Garbage in Garbage out. Savvi analyzes your data before any modling begins to insure you get the best results possible. Our Quality Control panel go throuhgt the data and removes any data the could return false results. If your data set falls below the minimum number of properties needed for the analysis, Savvi gives you a suggested MLS search parameters to incress a better results.