Content uploaded by Demetris Demetriou

Author content

All content in this area was uploaded by Demetris Demetriou on Nov 24, 2015

Content may be subject to copyright.

A GIS-BASED HEDONIC PRICE MODEL FOR AGRICULTURAL

LAND

Demetris Demetriou*a,b

aSchool of Geography, University of Leeds, Leeds LS2 9JT, UK

bLand Consolidation Department, 131 Prodromou street, 1419 Nicosia, Cyprus

ABSTRACT

Land consolidation is a very effective land management planning approach that aims towards rural/agricultural

sustainable development. Land reallocation which involves land tenure restructuring is the most important, complex and

time consuming component of land consolidation. Land reallocation relies on land valuation since its fundamental

principle provides that after consolidation, each landowner shall be granted a property of an aggregate value that is

approximately the same as the value of the property owned prior to consolidation. Therefore, land value is the crucial

factor for the land reallocation process and hence for the success and acceptance of the final land consolidation plan.

Land valuation is a process of assigning values to all parcels (and its contents) and it is usually carried out by an ad-hoc

committee. However, the process faces some problems such as it is time consuming hence costly, outcomes may present

inconsistency since it is carried out manually and empirically without employing systematic analytical tools and in

particular spatial analysis tools and techniques such as statistical/mathematical. A solution to these problems can be the

employment of mass appraisal land valuation methods using automated valuation models (AVM) based on international

standards. In this context, this paper presents a spatial based linear hedonic price model which has been developed and

tested in a case study land consolidation area in Cyprus. Results showed that the AVM is capable to produce acceptable

in terms of accuracy and reliability land values and to reduce time hence cost required by around 80%.

Keywords: Land consolidation; land valuation; automated valuation models, hedonic price model, GIS

1. INTRODUCTION

Land consolidation is considered to be the most effective land management planning approach1 for eliminating land

fragmentation2 and promoting agricultural and rural sustainable development. Land consolidation which is applied in

many countries around the world including 26 out 28 EU countries, consists of two main components: land reallocation

(or land readjustment) and the construction of agricultural infrastructure such as roads and irrigation networks. Land

reallocation is the most critical, complex and time-consuming part of the land consolidation process3,4 because it

involves the land tenure restructuring both in terms of ownership exchange and land parcels boundaries modification.

Land reallocation based on parcels’ land values because each landowner should receive after land consolidation a

property with approximately the same land value as the original. If this value exceeds the original value then the

landowner should pay the extra cost to the Land Consolidation Corporation and vice-versa. Therefore, land value is a

very critical parameter that directly affects the monetary interests of landowners hence it should be reliable, accurate and

fair among all landowners. This value can be the market value or the agronomic value5. In contrast to other countries, in

Cyprus the market value is utilized because agricultural land has broader development prospects hence it is attracted by

farmers and non-farmers as well.

In practice, in Cyprus, land valuation is a mass appraisal process6 carried out by the Land Valuation Committee (LVC)

which assigns a market value for each land parcel and its contents i.e. farmstead, wells etc. by employing the sales

comparison method that based on similar sales transactions occurred in the area concerned or the neighbourhoods. The

practical experience shows that the certain process faces some problems. In particular, it is time consuming hence costly

(it may take some months on a non-regular work basis) because the process is undertaken manually by inspecting all

parcels by the five members of the LVC. Furthermore and most importantly, outcomes present inconsistencies because

the LVC empirically tries to compare land parcels attributes that define land value, with the attributes of some typical

land parcels for which LVC assigned values based on sales transactions. As a result, due to not employing systematic

analytical tools such as GIS and thanks to the human weakness in estimating and processing so much information, it has

been shown a strong regional variation7 (within the case study area) of the coefficients of land valuation factors. This

variation that reflects the importance assigned by LVC for each factor has been revealed through geographically

weighted regression (GWR)8. Therefore, the process is not standardized, fully transparent and sometimes causes

unfairness and bias to the land reallocation hence it produces objections (officially or not) by landowners who usually

compare the land value assigned in their land parcels with other similar or neighbourhood parcels. Consequently,

Demetriou7 has proposed a framework involving a new land valuation process in land consolidation areas having in its

core automated valuation models (AVMs) 9,10 aiming to overcoming these problems.

AVMs are mathematically based computer software programs that are able to estimate the value of various types of

properties based on market analysis of a specified area and the characteristics of a certain group of properties stored in

appropriate databases for a given point of time9. Two well-known properties valuation methods are the sales comparison

method noted earlier and the hedonic regression approach involving that the value of a property is composed by the

value of its various attributes. Actually the two methods can be combined since usually the aim is the estimation of a

relationship among a dependent variable and one or more independent variables (attributes) based on known land values

for a set of properties. This process called calibration and it is the core of an AVM. Various methods have been used so

far among which multiple regression analysis (MRA) is the most traditional and popular method11 which is employed in

this research. In addition, several works include a spatial component i.e. GIS12,13,14,15 or they employ artificial intelligent

(AI) techniques as an alternative calibration approach instead of MRA, such as artificial neural networks (ANN)16, 17, 18,

19, expert systems20 , case based reasoning21, agent based models22 and genetic algorithms23 (GAs) which the latter two

have been arisen very recently.

In particular, GIS technology is nowadays a necessity for land valuation modeling9 because valuation process is

inherently spatial related and involves a great amount of data that can be easily stored, managed, analyzed and visualized

(e.g. value maps) through a proprietary GIS or ideally within a Computer Aided Mass Appraisal System (CAMA)24

integrated with the former. In addition, the huge GIS toolbox for spatial analysis provided in such systems and its

capability to integrate geo-information from various resources i.e. remote sensing, photogrammetry, airborne lidar

systems, mobile GIS and cetera, make them supreme instruments that open new horizons in land valuation in terms of

efficiency, effectiveness, reliability and accuracy. A raster based GIS has been used to assign a nominal asset values25 to

urban land consolidation area. Although the study is interesting it does not involve monetary values and quality

assurance is lack. It should be noted that while there is relatively rich literature about real estate valuation and namely for

residential properties, on the other hand, research related with land valuation of agricultural land and more specifically in

land consolidation areas is very rare.

In the light of the above, the aim of this paper is to present and discuss the development and evaluation of a hedonic

price AVM for a case study land consolidation area in Cyprus by utilizing linear MRA. Because sales transactions in the

area concerned were not adequate for accurate predictions7 three sample sets of land values (i.e. 10%, 15% and 20%)

provided by the LVC employed to estimate the values of the rest land parcels. Results are evaluated based on

international standards defined by the International Association of Assessing Officers (IAAO) 6,9,26 showing that AVM is

capable to produce acceptable in terms of accuracy and reliability land values and to reduce time hence cost required by

around 80%. In this context, the structure of the rest of the paper involves Section 2 which focuses on the methodology

utilised for developing and assessing the hedonic price model followed by section 3 that presents the land consolidation

case study area and the land valuation factors involved in the model. Afterwards, section 4 discusses the outputs of the

model by employing statistical analysis and ratio studies. Eventually, conclusions and recommendations for further

research are sum up in section 5.

2. THE DEVELOPMENT AND QUALITY ASSURNACE OF A HEDONIC PRICE MODEL

2.1 The development of the hedonic price model

Hedonic regression modeling is the oldest statistical calibration methodology has been utilized for estimating property

values gained the most attention since 60s early 70s27 until nowadays 28,29,30. As noted earlier, it is based on multiple

regression analysis (MRA) which is one of the most well-known statistical methods with huge range of applications

especially for prediction and forecasting. It involves the estimations of relationship among a dependent variable and one

or more independent variables. The general regression model relates Y to a function of X and β. This function can be

either linear or non-linear.

+ (1)

The dependent variable is Y whilst the independent variables denoted by X. The unknown parameters, denoted as β,

which may represent a scalar or a vector and represents random errors. In the case of linear function the development

of the above function involves a general multiple regression model with p independent variables as shown in the

equation below:

(2)

where Xij is the ith observation on the jth independent variable, and where the first independent variable takes the value 1

for all i (so it is the regression intercept). is an error term for observation i.

It should be emphasised that the reliability of MRA outcomes depends on the satisfaction of four main assumptions33

related to the residuals i.e. the difference between the actual market value and the predicted value: (i) normality, (ii)

constant variance (homoscedasticity), (iii) linearity and (iv) independence, which discussed later in the application of the

model.

2.2 International standards for quality assurnace

IAAO9 has defined standards and specifications for the development process of AVMs involving a number of steps: (i)

model specification, (ii) model calibration and (iii) model testing and quality assurance26. Specification involves the

definition of the valuation method i.e. sales comparison and the selection of independent variables will be included in the

model as predictors of the market value (depended variable) and; calibration is the process of testing model structure to

estimate variable coefficients/parameters using a different dataset employed for testing the performance of the model by

utilizing ratio studies26. Actually, specification and calibration is a common iterative process until statistical model

performance indices are acceptable. Both tasks are very important so as to develop an effective and accurate model. The

third step i.e. model testing and quality assurance involves various statistical diagnostic tests of the model performance

with a property sample (called holdout sample) that has not been used before in model calibration so as to ensure that it

fulfils the acceptable accuracy and reliability standards before its use. For this purpose ratio studies are employed that

encompasses four basic measures: (i) appraisal level (mean, median, weighted mean), (ii) variability-uniformity

(coefficient of dispersion-COD), (iii) reliability (confidence interval); and (iv) vertical inequities (price-related

differential-PRD, price related bias-PRB).

In particular, appraisal level aims to measure statistically how close appraisal are to real market values (or the LVC’s

values in our case) by employing primary measures of central tendency. While ideally the desire level is 1.0 i.e. appraisal

value equals the market value, an appraisal level between 0.90 and 1.10 is considered acceptable for any type of

property26 for certain confidence intervals. The evaluation of variability-uniformity measures the dispersion of ratios. In

general, the smaller the measure the better the uniformity, but extremely low measures suggest rather a problematic

calibration for many potential reasons i.e. sales chasing or non-representative sample. IAAO26 provides standards of

COD range for all type of properties. In our case, the standards for vacant land (with a range between 5.0 to 25.0) are

assumed because it is the closer category to agricultural land that is not exclusively involved in the list. Reliability

measures that are represented by the confidence intervals reflect the degree of confidence (for a certain percentage i.e.

95%) can be placed in a calculated statistic for a sample of appraised properties. A confidence interval noted in brackets

shows the upper and lower limits of a certain measure of central tendency and a desired level for any type of property

between 0.9-1.10. The measure of dispersion i.e. COD is a “horizontal” metric regardless of the value of individual

properties. In contrast, vertical inequities provide evidence about the accuracy of appraised individual properties. An

index for measuring vertical inequality is called price-related differential (PRD) which is calculated by dividing the mean

ratio by the weighted mean ratio. This statistic should be close to 1.0. Measures significantly above 1.0 show regressivity

i.e. low-value properties are appraised at a greater percentages market value than high-value properties. In contrast,

measures considerably lower than 1.0 indicate a progressivity, that is, low-value properties are appraised at smaller

percentages of market value than high-value properties. IAAO26 notes that PRD should be between 0.98 and 1.03.

Moreover, IAAO26 emphasizes that further to PRD, it is recommended to carry out a statistical test for price related bias

(PRB) as well, because it provides a more meaningful and easily interpreted index than the former. PRBs for which 95%

confidence interval fall outside the range of -0.10 to +0.10 indicate unacceptable vertical inequities. In addition to the

above metrics included in the ratio studies standards they have been also used the root mean squared error (RMSE), the

mean absolute percentage error (MAPE). The former measures the discrepancies between the predicted values and the

actual observations whiles the latter measures scaled discrepancies. Such measures have been also used in other case

studies 23,30. Furthermore, in order to measure how the error deviates, the absolute percentage error estimated which

called forecasting error (FE)17.

3. CASE STUDY AREA

3.1 The land consolidation area

Choirokoitia is a village in the Larnaca District of Cyprus, which is located southwest of Larnaca town (Figure 1). The

village is built on a hill with an average level of 230 m and the land consolidation area is located northwest of the village

in lowland with limited hills. The land consolidation area is included in an agricultural zone while on the east is almost

coincide with the main road that connects some of the mountainous villages of the District of Larnaca with the main

motorway in Cyprus. The land consolidation area has a size of 266 hectares and involves 488 land parcels (Figure 2).

The land use is mainly citrus, olives, various fruit trees and cereals. Some land parcels are irrigated through individual

drills and several land parcels are irrigated through a network connected with a water reservoir.

Figure 1. The location of Choirokitia village on the map Figure 2. The land consoliation case study area

of Cyprus

The land valuation in the certain area was carried out periodically from October 2008 until February 2009. The highest

land value has been defined in €35000 per decare (1000 m2) and the lowest in €2000. It should be noted that the LVC

separately valued any buildings included within the land i.e. a house, a farmstead, a drill, a fence and any isolated trees

included in the land e.g. large olive and carob trees.

Choirokitia

3.2 Land valuation factors

All information regarding the case study area built within ArcGIS 10.0. In particular, information includes a cadastral

map (Figure 2) (provided at a scale of 1:5,000) showing the original situation before land consolidation which is joined

with three databases containing information regarding landowners, ownerships and land parcels. In addition, the GIS

data model involves the official land valuation map with an associated catalogue, a soil map, a zoning map and a contour

map. All this information employed to automatically extract (using some programming routines) the scores for each

parcel for fourteen land valuation factors extensively discussed by Demetriou7. These factors are grouped31 in two major

categories: internal and external in relation to the property. Each category can be split further in two sub-categories:

physical attributes and legal factors for the former and locational characteristics and economic conditions for the latter.

In particular, each category includes the followings factors regarding each parcel with variable’s name within

parentheses: physical attributes are: size (size) in squared meters, shape (shape) measured in parcel shape index called

PSI32, mean slope (slope) measures in percentage, mean elevation from sea level (elevation) in meters, aspect (aspect)

that is measured clockwise in degrees (i.e. 0 north, 90 east, 180 south and so on), existence of a stream (stream) and soil

type (soil) has been provided by the Geological Survey Department which involves two types: Skeletic-calcaric-

REGOSOLS and calcaric-lithic-LEPTOSOLS represented by the letter “A” and calcaric-CAMBISOLS and calcaric –

REGOSOLS represented by the letter “B”. Legal factors involve the existence of irrigation rights (irrigation) for a

parcel. Locational characteristics are access through a registered road (access1), access through a registered pathway

(access2), the distance from residential zones (zone), the distance from the main road that connects the neighborhood

villages with the motorway (main_road) and the existence of sea view (sea_view). Economic conditions are land-

use/productivity (land_use) for the agricultural economic potential of a parcel which is reflected by the expected net

revenue per decare for various crops that has been provided by the Agricultural Department of Cyprus. The dependent

variable i.e. the market land value is measured in Euros per decare.

4. RESULTS AND DISCUSSION

4.1 MRA results

As noted earlier, sales transactions for the consolidated area were note sufficient to predict land values, hence, three

sample sizes i.e. 10%, 15% and 20% of land parcels values assigned by the LVC used for that purpose. In contrast to

many statistical experiments that use a random sampling, in this case it is very important that the selected land parcels

should adequately represent in terms of features the whole population. Therefore, parcels were selected based on their

attributes to cover all the range found in the population. In particular, they were selected all the parcels having the min,

max and mean score of all the continuous land valuation factors and then parcels were selected randomly to cover (as

much as possible) the whole case study area. Similarly, categorical factors (i.e. those take a binary value), parcels were

selected proportionally based on the frequency of each score (i.e. 0 or 1) and randomly taking into account to keep a

location-based balance.

The multiple regression analysis model ran with the three basic methods for selecting variables i.e. forward selection,

backward elimination and stepwise variable selection. The second method gave slightly better results from the other two.

Therefore, the basic model summary statistics and the number of variables included in the final model for each sample

are presented in the following Table 1.

Table 1: Summary statistics for the linear model

Sample

R2

Adjusted

R2

Variables

remained

Durbin-

Watson

1

0.820

0.799

Size, stream, slope, access1, zone

2.50

2

0.788

0.769

Access2, size, slope, acces1, irrigation, zone

1.98

3

0.814

0.799

Access2, size, access1,irrigation, aspect, slope, zone

2.14

The square of the correlation coefficient R2, that indicates the proportion of the observed variability of the dependent

variables which is explained by the independent variables, is quite high for all samples with the highest for sample-1

(which is close to that of sample-3). This indicates that even a small sample which includes only 10% (i.e. 49 land

parcels) of the total population may fit a model very well as shown graphically for sample 1 in the Q-Q plot of the

standardized residuals in Figure 3a where the points fall close to a straight line.

Figure 3. a) The normal Q-Q plot of the standardized residuals; b) The distribution of residuals; c) The scatterplot of

studentised residuals against the standardized predicted value, all for sample 1

Similarly, the adjusted R2 (adjusted for the number of parameters in the equation and the number of data observations)

which is always smaller than R2 reflects how well the model would fit another sample from the same population is

equally the highest for samples 1 and 3. The estimated regression equations for each sample revealed that the signs of all

coefficients are reasonable, namely size, stream, slope, zone and aspect having negative signs whilst access1, access2,

and irrigation having positive signs. A critical issue among the independent variables included in each final model is the

potential correlation and the multicollinearity i.e. the linear relationship. The former statistic is measured by Pearson

correlation coefficient and the latter by the tolerance. The Pearson’s correlation matrix showed that for all samples all

potential combination of variables have a correlation less or equal to 0.5 which is a very acceptable result. Similarly,

tolerance that may take values between 0 to 1 (0 means there is a linear relations and 1 means there is not a linear

relationship) is in most cases equal or more than 0.7 except in two cases where irrigation and zone variables have a

tolerance around 0.5-0.6. However, this fact does not suggest a problem since the value is quite far from 0.

The four MRA assumptions examined in the LVC model are all fulfilled indicating reliability of results. In particular,

normality is confirmed for all three samples by utilizing Shapiro-Wilk test with an output value for sample-1 0.979 (1

means that sample data are perfectly normal) with a level 0.132 which is more than 0.05 and a standard deviation of

standardized residual 0.963. Similarly, sample-2 model results in a value 0.983, with a level 0.436 and a standard

deviation of standardized residual 0.957; and Sample -3 model results a value of 0.980, with a level 0.584 and a standard

deviation of standardized residual 0.946. Normality assumption is evident from the distribution chart which is close to

normal as shown for sample 1 in Figure 3b. Homoscedasticity is also confirmed since the scatterplot of studentised

residuals against the standardized predicted value do not present any pattern as shown for instance in Figure 3c for

sample 1.

Regarding linearity assumption, F-test showed that there is a linear relationship between the dependent and the

independent variables because the null hypothesis (i.e. there is no linear relationship between the dependent and the

independent variables) has been rejected since the F ratio is small (based on the magnitude of the regression coefficients)

for all samples with a high significance level. However, visual inspection showed that slightly non-linear function (e.g.

exponential) could fit for the variables size, slope and zone. Independence of observations also confirmed by Durbin-

Watson test for the three samples since all values are between1.50-2.5 while values close to 2 involves no correlation. In

addition, spatial autocorrelation tests (p-value and z-score) within ArcGIS showed that the p-value is not statistically

significant hence we cannot reject the null hypothesis, meaning that it is quite possible for 90% confidence level that the

spatial distribution of residuals’ values is the result of a random spatial process.

(a)

(c)

(b)

4.2 Ratio analysis results

All the quality assurance measures noted in section 2.2 estimated. In particular, results show that all the appraisal level

statistics, that is the three main measures of central tendency the mean, median and weighted mean of ratios are within

the acceptable by IAAO26 range, namely, between 0.9 and 1.10 revealing an initial positive sign. A stronger relevant

indication is the use of confidence intervals to determine whether it can be reasonably concluded that the noted appraisal

levels differ from the established performance standards in a particular instance. Therefore, it has been confirmed that

with 95% confidence interval estimates for the mean and the median relevant standard has been met since all measures

fall within the noted range suggesting reliability evidence. Furthermore, the coefficient of dispersion-COD is within the

noted standard range for vacant land (i.e. between 5.0-25.0), ranging from 11.66 the minimum (for sample 3) to 13.83

the maximum (for sample 1), which are quite far from the highest acceptable value i.e. 25. COD is improved as the

sample increases while samples 3 and 2 present similar COD. Similarly, the price-related differential-PRD is almost the

same for all samples and it is marginally within the acceptable range 0.98-1.03 but that for sample 1. This potentially

shows a slight regressive tend i.e. low-value properties are appraised at a greater percentages market value than high-

value properties. Similarly, the estimated price related bias-PRB ranges from -0.087 to -0.036 for the three samples

which is within the acceptable limits i.e. -0.10 to 0.10, showing a trend towards the lower limit. Therefore, PRB shows

that for 95% confidence interval, assessment levels do not change by more than 10% when values are halved or doubled.

Furthermore, it was found that RMSE and MAPE decreases as the sample increases. However, the decrease is very slight

between sample-2 and sample-3 suggesting that no significant performance improvement can be achieved if the sample

increased more than a limit. Similarly, the best percentage of FE for 10% resulted from sample 2 and 3 with around 57%

of land parcels values estimated with accuracy better than 10% and around 32% of land parcels values estimated with

accuracy better than 20% compared with the true values. Therefore, it seems that a sample of 15% is adequate to

accurately predict with a discrepancy between 0%-20% from true values for the 90% of parcels. In alignment with this

finding, the AVM could carry out the site work of the LVC by employing around 80% less resources because the LVC

could carry out only 5 site valuation visits for assigning land values to the selected 15% sample of parcels (i.e. 73

parcels) instead of 25 days to assess all the 488 land parcels of the consolidated area. These figures show a proportional

reduction of both time and costs. Further to these savings, the most beneficial parameters provided by the AVM is the

quality of valuation in terms of comprehensiveness (a great number of variables are taken into account) consistency

(precise comparison of variables scores), reliability (predictions based on international standards) and transparency (the

analytical explanation of a land value is possible) of outcomes with an equity impact to the landowners involved.

5. CONCLUSIONS

The current mass land valuation process employed in land consolidation schemes is not efficient and reliable thus

authorities involved need to introduce AVMs based on international standards. Clearly, the AVM presented in this paper

is considerably more efficient than the traditional empirical process in terms of time, costs, accuracy, reliability,

consistency and transparency. Ongoing research focused on employing different calibration methods for developing

AVMs based on a non-linear function and artificial neural networks, in order to investigate and compare the performance

for all the three models.

ACKNOWLEDGEMENTS

The data used for this research have been kindly provided by various Departments of the Public Sector in Cyprus.

Therefore, I would like to thank very much the Departments of: Land Consolidation, Land and Surveys, Geological

Survey and Agriculture and the people assisted me to get and discuss further details regarding these data.

REFERENCES

[1] Demetriou, D., Stillwell, J., and See, L., 2012 “Land consolidation in Cyprus: why is an integrated planning

and decision support system required?” Land Use Policy, 29 (1), 131–142 (2012).

[2] Van Dijk, T., [Dealing with Central European land fragmentation], Delft, The Netherlands: Eburon, (2003).

[3] Thomas, J., “What’s on regarding land consolidation in Europe?” Proceedings of the XXIII FIG congress,

shaping the change, October 8–13, 2006, Munich, Germany. Available from: http://www.fig.

net/pub/fig2006/papers/ts80/ts80_03_thomas_0311.pdf [Accessed 19 January 2013].

[4] Ayranci, Y., “Re-allocation aspects in land consolidation: a new model and its application” Journal of

Agronomy, 6 (2), 270–277 (2007).

[5] FAO, [Overview of land value conditions], FAO, Rome, (2003).

[6] International Association of Assessing Officers (IAAO)., [Standard on Mass Appraisal of Real property].

IAAO, Kansas City, Missouri, (2013).

[7] Demetriou, D., “Assessing the land valuation process carried out in land consolidation: the need for a new

land valuation framework”, (to be submitted), (2015).

[8] Brunsdon C., Fotheringham S., Charlton M., “Geographically weighted regression: a method for exploring

spatial nonstationarity” Geographical Analysis, 28, 281-289, (1996).

[9] International Association of Assessing Officers (IAAO) “Standard on Automated Valuation Models

(AVMs)”, IAAO, Chicago, Illinois, (2003).

[10] Downie, M., and Robson, G., “Automated valuation models: an international perspective” The Council of

Mortgage Lenders, London, (2007).

[11] Bernknopf, R., Gillen, K., Wachter S., Wein, A., “Using econometric and geographic information systems

for property valuation: A spatial hedonic pricing model”. Working paper, Spatial Integration Laboratory for

Urban Systems (SILUS), Wharton University of Pennsylvania, (2008).

[12] Longley, P., Higgs, G., Martin, D., “The predictive use of GIS to model property valuations” International

Journal of Geographical Information Systems, 8(2), 217-235, (1994).

[13] Wyatt, P., “The development of a GIS-based property information system for real estate valuation”

International Journal of Geographical Information Science, 11 (5), 435-450, (1997).

[14] Iman, A., "GIS-MRA techniques in property valuation: a framework for implementation” Bulletin

Geoinformaci, 3(1), 63-79, (1999).

[15] Hamilton S., and Morgan, A., “Integrating lidar, GIS and hedonic price modeling to measure amenity values

in urban beach residential property markets” Computers, Environment and Urban Systems, 34(2), 133-141,

(2010).

[16] Kathman, R., “Neural networks for the mass appraisal of real estate” Computers, Environment and urban

Systems, 17 373-384, (1993).

[17] Nguyen, N., and Cripps A., “Predicting housing value: a comparison multiple regression analysis and

artificial neural network” Journal of Real Estate Research, 22 (3), 313-336, (2001).

[18] Garcia, N., Gamez, M., Alfaro, E.,”ANN+GIS: An automated system for property valuation”

Neurocomputing, 71, 733-742, (2008).

[19] Pao, H., “A comparison of neural network and multiple regression analysis in modeling capital structure”

Expert Systems with Applications, 35, 720-727, (2008).

[20] Kilpatrick, J., “Experts systems and mass appraisal” Journal of Property Investment & Finance, 29 (4/5),

529-550, (2011).

[21] Gonzalez, A., Laureano-Ortiz, R., “A case-based reasoning approach to real estate property appraisal”

Experts Systems with Applications, 4, 229-246, (1992).

[22] Breen, J., Goffette-Nagot, F., Jensen, P., “An agent-based simulation of rental housing markets” Working

Paper, Archive ouverte HAL, France, (2009).

[23] Ahn, J., Byun, H., Oh, K., Kim, T., “Using ridge regression with genetic algorithm to enhance real estate

appraisal forecasting” Expert Systems with Applications, 39, 8369-8379, (2012).

[24] Pashoulis, V., “The general valuation law and the CAMAS in the Land and Surveys Department in Cyprus”

Proc. FIG Working 2011. Marrakech, Morocco, May 18-22, (2011).

[25] Yomralioglu, T., Nisanci, R., Yildirim, V., “An Implementation of Nominal Asset Based Land

Readjustment” Proc. Strategic Integration of Surveying Services, FIG Working Week 2007, Hong Kong

SAR, China, 13-17, (2007)

[26] International Association of Assessing Officers (IAAO), “Standard on Ratio Studies” IAAO, Kansas City,

Missouri, (2013).

[27] Smith, T.R., “Multiple regression and the appraisal of single family residential properties” The Appraisal

Journal, 39 (2), 277-84, (1971).

[28] Milla, K., Thomas, M.H.., Ansine, W., “Evaluating the Effect of Proximity to Hog Farms on Residential

Property Values: A GIS-Based Hedonic Price Model Approach” URISA Journal, 17,(1), 27-32, (2005).

[29] Eckert, J., “Computer-assisted mass appraisal options for transition and developing countries” International

Studies Program, Working paper 06-43, Georgia State University, Andrew Young School of Policy Studies,

(2006).

[30] Schulz, R., Wersing, M., Werwatz, A., “Automated valuation modeling: A specification exercise” SFB 649

Discussion Paper 2013-046, University of Aberdeen Business School, Technical University of Berlin,

(2013).

[31] Wyatt, P., “The development of a property information system for valuation using geographical information

system (GIS)” Journal of Property Research, 13 (4), 317-336, (1996).

[32] Demetriou, D., See, L. and Stillwell, J., “A Parcel Shape Index for Use in Land Consolidation Planning”

Transactions in GIS, 17(6), 861–882 (2013).

[33] Norusis, M., [SPSS 13.0 Guide to Data Analysis], Prentice Hall, New Jersey, (2005).