Artificial Intelligence (AI) Based Automated Valuation Models (AVMs)

Is valuation a science of an art? The science aspect has attracted strong attention from researchers and advocates of using technology, especially the artificial intelligence (AI), into valuation. This would further enhance the capability of the traditional mass appraisal/valuation approach, which invariably relies on the standard hedonic regression estimation.  Does the AI improve the efficiency and accuracy of the mass appraisal system?

The mass appraisal system, or sometimes referred to in practice as the automated valuation model (AVM), estimates real estate (housing) prices as a function of a set hedonic attributes and spatial characteristics. The hedonic attributes include unit size, floor height, age, tenure, sale type and others, and the spatial characteristics include local amenities, such as distance to the nearest Mass Rapid Transit (MRT) stations, schools, bus stations, park, shopping malls and others. The AVM system requires the inputs of a large sample of historical transaction evidence of the comparable real estate, and feed the data into the OLS model that minimizes the squared error terms. With the estimated coefficients on the hedonic and spatial attribute, the model could then predict, or from the AVM term, to assess a new subject property based on the historical inputs. If the market is stable and not subject to exogenous policy shocks, we may expect the prediction by the AVM to follow closely to the objective values of the markets. In other words, the AVM could produce valuations for subject properties that are as close as possible to the valuation of professional valuers/appraisers. The AVM estimates could be rather straightforward when applied to valuation of relatively homogenous properties, which do not have significant variations in their hedonic and spatial characteristics. In other words, the valuation of these properties require little “art” in the adjustment to the values.

This study aims to explore the AI and Artificial Neural Network (ANN) based methodology, and its applications in improving the AVM estimation. It would also examine how the AI-ANN could produce more reliable and accurate estimation compared to the traditional regression-based AVM. The issues of the “substitutability” of the art in the valuation by the machine input also causes serious concerns of many practitioners, and the issues will also be investigated in greater details in the study.

Objectives of the Study

The objective of the study are broadly defined below:

  1. To review the current state of applications of AI and ANN technology into mass appraisal and AVM;
  2. To develop AI and ANN-Based AVM using housing transaction data in Singapore; and
  3. To study potential effects of the use of advance technology on the valuation professionals and practices in Singapore.

The study may produce useful contributions to the literature in term of advancement of the valuation knowledge with the use of AI and ANN technology. More importantly, the study hopes also to develop an application of AI and ANN-based AVM to raise the professional standard and also efficiency of the valuation professionals and practices in the industry.

 

Working paper

Sing, Tien Foo, Yang, Jesse, Yu, Shi Ming, “ Artificial Intelligence (AI) Based Automated Valuation Models in Predicting Property Values: A Comparison of Tree-Based Methods and Multiple Regression Analysis”