Project topics and materials
Get Complete Project Material File(s) Now! »
Background to the Study
The market value of a property is a matter of great interest to local authorities, mortgage institutions, dissolved companies and other market participants, as either of the parties might be disadvantaged should there be an error in the assessment process. Though litigation arising from inconsistent and unreliable estimates by disadvantage parties rarely occurs, it is necessary to guard against its occurrence by ensuring that estimates reflect the market price of properties. Appraisal of a property or properties is a complex procedure due to the different influential factors that constitute the market price(s). While it appears relatively easy to conceptualise the features that will most considerably be associated with a property market price, quantifying their magnitude and contributions is another difficulty. Traditionally, income, cost and market approaches are utilised in the estimation of market values of residential properties. But these methods are increasingly becoming unsustainable for mass appraisal of properties because of valuers’ subjectivity, time, cost and insufficient number of comparable properties for assessment. Having realised these limitations particularly as it concerns cost, time and accuracy, various municipalities have introduced computer assisted mass appraisal (CAMA) with the use of hedonic regression models (HRMs). In South Africa, the most successful application is probably the city valuation office (CVO) Cape Town. According to KPMG 2015 report, modernisation of CVO resulted in the reduction of the general assessment cost by R94m (US$7.7m) from 2000 to 2009. Consequently the total revenue from property tax that accrue to the city of Cape Town for 2014 alone was R6 billion (US$ @ 12.21).
The study of Bourassa, Cantoni & Hoesli (2010: 139) reported the widespread use of the HRMs to include price index construction, mass appraisal of properties for taxation, mortgage underwriting and portfolio management. The study noted also that the model is relevant in the assessment of the externalities on property values. Relative to mass appraisal of properties, the model has a long history of use among academics and practitioners (Zurada, Levitan & Guan, 2011: 350). However despite its extensive used, the method is fraught with a number of limitations including inability to handle specification error exacerbated by nonlinearity, multicollinearity and functional form (Do & Grudnitski, 1992: 38; Worzala, Lenk &Silva, 1995: 185). The study of Bourassa et al. (2010: 139) having identified these limitations gave a caveat that should follow the use of HRM including careful selection and measurement of relevant variables and ensuring independence of errors one from another.
Recent events in the real estate market have shown a gradual increase in the emergence of several alternative methods used in the assessment of property prices (Lin & Mohan, 2011: 224). This development is noticed in the shift in emphasis towards the additive nonparametric regression (ANR) (Lin & Mohan, 2011); support vector machines (SVMs) (Lam, Yu & Lam, 2009; Zurada et al., 2011); artificial neural networks (ANNs) (Borst, 1991; Worzala et al., 1995; Zurada et al., 2011; Lin & Mohan, 2011; McCluskey, McCord, Davis, Haran, &McIlhatton, 2013); M5P trees (Zurada et al., 2011) and spatial and temporal models including geographically weighted regression (GWR); geographically and temporally weighted regression; simultaneous autoregressive model (SAR) (Pace, Barry, Gilley & Sirmans, 2000; Sun, Tu & Yu, 2005; Huang, Wu & Barry, 2010; McCluskey et al., 2013 and Fotheringham, Crespo & Yao, 2015) amongst others. While some of these models have their root in the HRMs (global and local), others have completely different underlying philosophies. However, it has generally been established that all models have their individual strengths and weaknesses (Kauko & d’Amato, 2008: 17; McCluskey et al., 2013: 240) and there has not been a general consensus on any of the appraisal techniques. Again McCluskey, Davis, Haran, McCord & Mcllhatton (2012: 275) observed that there is no generally accepted class of nonlinear models that can be applied to explore multivariate relationships because of the myriads number of available models.
1. CHAPTER ONE Introduction
1.1 Background to the Study
1.2 Problem Statement.
1.3 Aim of the Study .
1.4 Objectives of the Study
1.5 Research Methodology.
1.5.1 Literature Review.
1.6 Contributions to Knowledge.
1.7 Organisation of the thesis
2. CHAPTER TWO. Literature Review.
2.1 Introduction
2.2 Pricing of properties with traditional hedonic regression
2.3 The application of ANNs in mass appraisal of properties.
2.4 Criteria for the selection of appropriate pricing model .
2.5 Chapter analysis, summary and conclusion.
3. CHAPTER THREE. Mass Appraisal Modelling Techniques
3.1 Introduction
3.2 Hedonic Regression Models
3.3 Additive Nonparametric Regression
3.4 M5P Trees .
3.5 Artificial Neural Network Models
3.6 Support vector machines (SVMs)
3.7 Chapter summary and conclusion
4. CHAPTER FOUR . Data and Modelling Procedures .
4.1 Introduction
4.2 Performance measurement .
4.3 Chapter summary and conclusion
5. CHAPTER FIVE Analyses and Discussion of Results .
5.1 Introduction
5.2 Establishing a baseline model for the Cape Town property market.
5.3 Prediction of property prices with spatially varying and weighted regression models.
5.4 Prediction of property prices with support vector machines, M5P trees and additive nonparametric regression models
5.5 The influence of ANNs training algorithms in mass appraisal
5.6 Combining GABP and PSOBP in weight optimisation and ANNs training.
5.7 Effective comparison of the performance of models
5.8 Chapter summary and conclusion
6. CHAPTER SIX Conclusion and Recommendations for Future research.
6.1 Introduction
6.2 Realisation of study objectives
6.3 Conclusion.
6.4 Practical application of the study
6.5 Recommendations for Future Research.
7. References .
