AVMs (Automated Valuation Models) are often touted as something appraisers should understand and do more of.  This is a problem of words.

AVM” is not descriptive of a formula to calculate value.  It is an industry.  Each company in the industry does use algorithms, formulas, and calculations to predict (estimate) value.

The use of the acronym AVM does not describe a way of calculating value.

In order to get clear on the relationship between ‘automated’ models and what appraisers and asset analysts can do – we need to be completely, rigorously clear on what these words mean.

Words control our thinking and our opinions.  Each AVM company uses computer algorithms, which follow the directions of a model.  An individual AVM is not an algorithm, although it uses algorithms to arrive at a prediction of value.  AVM companies generally keep their algorithms secret.

The AVM algorithms follow models selected and applied by computer programmers and subject-matter experts.  Nearly all AVMs were originally created by appraisers.  Each followed a model they believed adequately represented reality.  Each model is designed to direct the computer to which algorithm (calculation/logic) to follow.

Model” is a small copy or representation of reality.  As an appraiser cannot bring the properties and relationships to the desk of a client – the appraiser/analyst must find a way to represent the larger reality.  Traditionally, appraisers applied three (or more) ‘approaches to value’.  Each of the ‘approaches’ is itself a model, providing a different way of observing and representing reality.

Algorithm” is a series of calculations and logical decisions (instructions).  Computers are quite good, fast, and mistake-free in following instructions.  When mistakes happen, it is in the instructions. Computers follow instructions.  Humans create instructions.  Instructions they believe follow good models.

Some mistakes can be trivial, and cause small errors or bias.  Other mistakes are from wrong algorithms.  This type of error can come from one of two sources:

  1. The wrong question was asked, or misunderstood.
  2. The wrong model was selected, or neglected.

Humans select models.  Models determine algorithms.  Algorithms deliver estimates or predictions.

Humans can then explain human reasoning and any algorithms (approaches) and provide an opinion.  The opinion (for appraisers) is a point value (or range).

The appraisal problem is almost entirely subjective, starting from ‘picking comps’, identifying ‘elements of comparison’, ‘making adjustments’, ‘reconciling’ the differing results of the ‘approaches, and emphasizing it is just an opinion, not an estimate nor an analytic result.

The valuation problem (applying the science-of-data method) starts with a similarity matching to define the data selection, one of three basic predictive algorithms, and risk/reliability scoring.  These results are delivered via summary numbers, visuals, and even interactive dashboard displays.

They can be directly input to collateral risk or portfolio management schemes, investment objectives, or equity enforcement (such as tax assessment or litigation judgments).