Our News

  1. Home
  2. Our News

This study employs the statistical method of Multiple Linear Regression analysis (MLR) to develop an Automated Valuation Model (AVM) for estimating land values by utilizing transaction-based data in Limassol, Cyprus. The authors focus on the confidence level and accuracy of the value estimated by an AVM. Thus, the developed AVM was tested in two contrasting areas of Limassol in terms of location characteristics and market conditions. Most AVMs contain a statistical method to generate the estimated value of a real estate property. However, the outcome of a statistical method is verified by statistical measures. Therefore, if the validation of the predicted value for its accuracy derives from the statistical metrics of the model, then the explanatory variables cannot remain constant. It is implied that the AVM in order to grant the highest statistical metrics for a given property valuation requires different combination of independent variables in different locations, which means that the parameters of the model should change or adjust for every case to obtain the best fit model. The authors demonstrate that the best fit model is obtained when several models are executed with alternative combinations of variables. Hence, the best fit to the regression is given by the model with the better statistical measures when compared to the other models. Consequently, the predicted value is supported by statistical significance and can be adopted at a high confidence level.

Property valuation evolved from simple empirical judgements to automated valuation models and their application have extended from single property to mass valuation. Many governments across the world have used AVMs to get a valuation in thousands of properties for tax related purposes. The literature review is extensive and it is growing day by day. The island of Cyprus was introduced to computer assistant mass appraisal (CAMA) in 2013 when the Department of Land and Surveys (DLS) performed a general valuation and then to revaluation in 2018.  The aim of this research is to provide more transparency to the reliability of the data used in the latest general valuation. An automated valuation model was developed, using the MRA method and Hedonic Pricing Model, to test the performance of the data and compare them with the minimum standards a valuation model should have according to the International Association of Assessing Officers (IAAO).  A case study using a holdout sample with data from Lakatamia Municipality was created to observe the reliability of the data but also to improve the accuracy of the Automatic Valuation Model. Three regressions were carried out: a) Basic regression with 503 observations and 10 variables, b) Regression with the previous variables plus 10 nearest neighbors as predictors and c) Regression with the previous variables plus 10 nearest neighbors as predictors, with 450 best observations – deleted outliers based on absolute error. The coefficient of determination (R-squared) measures the goodness of fit of the regression line, in other words, how close the data are to the estimated line. Initially the R-squared was 0.319 which is above IAAO standards but it was increased to 0.765 after the application of the third model. This accuracy showing better performance than the mass valuation system applied by the Department of Land and Surveys in Cyprus with accuracy of 0.384 Concluding the research ends with a critical discussion about the reliability of the data and some suggestions that could be applied by the DLS to improve the performance of the data.  It is worth mentioning that the Cypriot data have a limitation due to the high heterogeneity found between properties.

The integration of Artificial Intelligence (AI) in real estate and more specifically in property valuations marks a transformative era in the industry.
Without data, AI does not exist. AI’s capabilities in data analysis, prediction, and automation are revolutionising how properties are valued, managed, and traded.

Loading...