The term “accuracy” refers to the ability of a statistical valuation method to produce results close to the respective Benchmark Values.
Benchmark values are intended as the correct market value; hence it typically consists of either a reliable surveyor valuation or sale price.
Accuracy of statistical valuation methods incorporates two distinct dimensions:
Bias, intended as any overall tendency to systematically overvalue or undervalue properties when compared to the benchmark value.
It represents the “true accuracy of the statistical valuation method” and it is based on the median of actual error in percentage terms (MdPE).
Dispersion, intended as the relative frequency of all different sizes of errors. This typically displays the shape of a bell curve with a tall narrow peak and thin tails if dispersion is low, or a low broad peak and thicker tails if dispersion is high.
It can be quantified by Median Absolute Percentages Error (MdAPE) or by Percentage Prediction bands that represent the percentage of model results that had an error within a specific tolerance (PPE5, PPE10, PPE20).
Measures of bias:
MdPE, intended as the median value of the valuation error. It is preferred to the Mean (Average) Error for its robustness (Outliers Neutralized).
Measures of dispersion:
MdAPE, intended as the median of the absolute value of the valuation error.
Percentage within x%, intended as the percent of model results that had an error within a specific tolerance. PPE05, PPE10, and PPE20 represent the most commonly used tolerance threshold and they mean an Error ≤ 5%, 10% or 20% respectively, regardless of its + or – sign.
They do not require the data to be normally distributed, and they mean exactly what they say.