# Automated Valuation Models

Automated valuation models (AVMs) are mathematical models, which, together with appropriate computer software and databases of property information, are used to provide real estate valuations.

Types of AVMs
AVMs are categorized into at least five types. These are hedonic models, econometric forecasts, ‘intelligent’ systems, house price index models and tax assessed value models.

Hedonic Models
The basic premise of hedonic models is that the price (or value) of a marketed good such as real estate, is a function of its constituent characteristics. The implication of this is that the good being valued could be decomposed into its constituent parts, and market values attributed to those parts. The market price (or value) of a specific residential property, for example, might be regarded as a summation of the values of its constituent characteristics, such as size, location, age, and so on. It is obvious that each of these characteristics influence the price that is paid for property in the market. Typically, hedonic models use regression techniques to estimate the contribution of each feature of the property to the overall value.

Econometric Forecasts
Hedonic AVMs typically use historic transaction data to estimate current market value, introducing potential inaccuracy due to data lag. Econometric forecasts of value is one way of handling this problem, and allows for the modelling of any market movements in the period between the time of transaction data and the valuation date (ibid.). A number of econometric techniques are available for the forecasting of real estate value, all of which invariably use advanced mathematical/statistical techniques.

Intelligent Systems
The so-called Intelligent Systems refer to a number of valuation techniques that are distinguished by their attempt to replicate the thinking of market actors to arrive at an estimate of value. Hedonic models in intelligent systems are designed to identify the variables relevant to market value and to ‘learn’ about changes in the relationships between these variables and value, thus continually updating the model on the basis of new transaction data. As the authors point out, artificial intelligence models have the advantage that their structure is more transferable between different countries or market areas than is a regression model constructed to reflect only local property types and features of the locally available data.

House Price Index Models
These models rely on house price indices constructed on the basis of repeat sales. The indices, usually disaggregated to some local level, are used to track changes in house prices between time periods. In application, AVMs based on house price indices are straightforward and easy to use. The modelling takes past transaction prices, or valuations, of subject properties and updates these by reference to changes in the values of the relevant indices.

Tax Assessed Value Models
Tax Assessed Value models (TAV) are normally used for updating real state values for tax purposes. The models work on the assumption that there exists a statistical relationship between past assessed values and subsequent price data to create a ratio, disaggregated to local level. In application TAV models take a valuation assessed for tax purposes at a past date and update it to estimate current market value on the basis of the established statistical relationship.

Literature: Estate Valuation Theory, A Critical Appraisal,  Manya M. Mooya, Department of Construction Economics and Management, University of Cape Town, Rondebosch, South Africa, 2016