Deriving Housing Preferences from advertising on the web for ...

2 downloads 42 Views 1MB Size Report
RC43 conference, At home on the housing market, Amsterdam, Netherlands July 10-12, 2013. 1. Deriving Housing Preferences from advertising on the web for.
RC43 conference, At home on the housing market, Amsterdam, Netherlands July 10-12, 2013

Deriving Housing Preferences from advertising on the web for improving decision making by Economic and Social actors

Tam Van Nguyen* and Frank De Troyer Division of Architectural Engineering, Department of Architecture, Urbanism and Planning, KU Leuven, Kasteelpark Arenberg 1 - box 2431, 3001 Heverlee, Belgium ABSTRACT In the South Asian region, mass production approaches of large housing projects have been used for several decades. A large number of housing units have been built but are not fully occupied. Insight in housing preference is crucial information for designers and developers. Many studies have analyzed housing preferences but they are not used, moreover, most data sources from questionnaires and historical documents are elaborated with high cost and outdated when available. To understand housing preference and housing plot preferences, we propose a method to derive them from selected internet websites, and include them in a decision model at the early investment stage. The text elaborates how to derive a utility function with diminishing marginal utility via regression analysis and how to include this in a decision model. There are two clear conclusions: First, housing preference is only slightly better described with a non-linear utility function than linear one. Second, visual summary of the model can facilitate the selection process regarding housing layout in planning phase for stakeholders. In conclusion, the model can stimulate the housing market by linking the “willingness to pay of customers” and “profit expectations of economic actors”. The model is also useful for policy formulations and social interventions in the market. The model can be further elaborated by including more design variables and can be applied in other contexts. Keywords: Housing preference, Decision- Making, Multi-Objective optimization, web based data collection, early design stage, willingness to pay,

Introduction Inadequate condition of urban housing is a global problem, and it is the most serious issues in developing countries. One billion people, accounting for thirty percent of the world’s urban population, live in poor housing condition (UN-HABITAT, 2008). A high housing demand in the urban areas can be predicted based on the already dense population now, fast population growth and high rural-urban migration. With population of 88 million, Vietnam is the third highest populated country in South East Asia, and is ranked the thirteenth position in the world (GSO, 2011). As a consequence, more than 30% of total household lives in dwellings with less than 36 square meters floor area and 19% of total households live in temporary houses (MOC, 2009).

*Correspondence Address: [email protected], [email protected], Cantho University.

1

RC43 conference, At home on the housing market, Amsterdam, Netherlands July 10-12, 2013

A stated hedonic price, elaborated with questionnaire method, can estimate trade-offs between customer value and the construction cost. By doing so a priority list of housing attributes can be elaborated (Hofman, Halman, & Ion, 2006). Many researchers try to fill the gap from the preferences to design decisions. The theory of means-end chain was proposed to deal with user preferences of housing and to link three aspects: attributes, consequences and values of buildings (Zinas & Jusan, 2012). Other researchers used questionnaires to obtain housing preferences and then identified relationships between willingness to pay and total cost of main housing attributes (Delgado & De Troyer, 2011). To understand housing preferences, we propose a method to derive preferences from selected internet websites, and include them in a model for decision making in the early design stage. This text report a first elaboration whereby only the value and cost of developed land in is predicted in 3 steps: (1) select residential land characteristics for large scale housing projects; (2) derive a function to predict selling price based on selected housing attributes via regression analysis and (3) link cost and willingness to pay to generate graphs that provide design decision information for designers and developers taking into account users’ preferences. The remainder of the paper is organized as follows: Section 2 describes the housing status in Vietnam and Cantho city. Section 3 explains the methodologies. In Section 4 the results are presented and discussed in detail. The final section formulates conclusions. Overview Housing status in Vietnam and Cantho city In recent years Vietnam, with urbanization ratio from 30% to 40%, has been forced to develop many residential projects. Mainly three types of dwelling units are developed: terraced house, detached house and apartment. In most cases, the single private houses in urban areas are built by owners. They are named “Tube houses” or “Neo tube house”. These houses are terraced houses which a depth/width ratios from 3 to 6 and the widths are form 4 to 8 meters. Therefore, daylight and natural ventilation are limited. Two benefits of this housing type are low cost for land and access roads. Meanwhile, governments and private developers have chosen to develop large scale housing projects combining different types for both social housing and real estate market.

Hung Phu

Phu An project (Phuoc Thoi, 2011)

Figure 1 Perspective and construction progress of Phu An project, 145 ha, in Can Tho city. 2

RC43 conference, At home on the housing market, Amsterdam, Netherlands July 10-12, 2013

Typical large scale housing projects include social and commercial facilities with the three housing types mentioned before (Figure 1). Besides location, an appropriate layout with appropriate fractions allocated for residential, commercial, social functions are core factors of housing quality. The large scale housing projects with good planning, architecture and engineering designs should become achievable choices for all households. However, most of the residential projects in most cities of Vietnam are not fully occupied. With restructuring the city center area, resettlements for reconstruction of some areas at city center were combined with the large scale housing projects. Hence, governments supports with social loan, reduced interest rate and decreasing tax for land use. This approach ended up however with project minimizing open space and social infrastructure provisions. Some new housing projects were designed with only terraced houses and apartments. A high urbanization rate increases high pressure on housing supply chains in urban areas. The housing supply chain comes from individual, private developers and social projects. Residents build their individual housing projects with various designs. Most of these housing units are terraced houses with high depth/width ratio. Hence, these houses provide limited living qualities in terms of daylight, thermal comfort and air quality (To, 2008). Conceptual model Housing cost model - Element method for cost control

Figure 2 Element method for cost control, (De Troyer, 2008) The element method for cost control can analyze costs of buildings based on cost of elements (Figure 2), (De Troyer, 1990). The method was defined in RICS (Royal Institute of Chartered Surveyors) first published in 1969. Element are 3

RC43 conference, At home on the housing market, Amsterdam, Netherlands July 10-12, 2013

component, which fulfils a specific function such as external walls, roof, internal walls and floor (Smith & Jaggar, 2007). When technical parameters of a building are varied, then unit rates of elements can automatically adapted (De Troyer, 2003). This model has been employed in Vietnam and other countries. The problem of location and land The main difference between real estate and other products is location value because the possible link to other facilities such as schools, commercial center, green areas and transportation system. Hence, the method described below can be used if plots have identical location characteristics. In fact, the quality of residential lands can affect the quality of housing units later due to size and location (Huu Phe & Wakely, 2000). Overview of approach

Figure 3 Decision based on preferences and designs flowchart

4

RC43 conference, At home on the housing market, Amsterdam, Netherlands July 10-12, 2013

Diminishing marginal utility functions of housing preferences The intuitive behavior “law” is accepted as a working hypothesis in this study. This means diminishing marginal utility of each housing characteristic. For example, when continuous increasing a width of housing plot, the price per square meter of land is increasing not in a linear way but following a curve of which slope is less and less. Another example is one additional square meter for apartment with 36 m2 is more appreciated than for apartment with 100m2 (Figure 4). This hypothesis will require a transformation from power function to a linear function to apply a linear regression. Land price

Width of plot

Figure 4 price of land versus width of plot, utility function: y = axb +error and (0 < b < 1); or ln(y) = ln(a) + bln(x) + error. Multi-Attribute Utility theory The approach of a power function for a housing characteristic can extend to multiple housing attributes (Delgado & De Troyer, 2011). Two attributes can be presented in three dimension graph to search for an optimal alternative within budget constraint or other constraints. The concept of this approach also used in the “MultiAttribute Utility Theory” which supports decision-making when a selection is made out a limited number of available alternatives (Jansen, Coolen, & Goetgeluk, 2011). The method is first illustrated with two parameters in a graphical way and in a second step the elaboration to an approach for a multitude of attributes is explained. As example, a plot is considered with two basic attributes the width (W) and the depth (D) (Figure 5).

Figure 5 selling price of land is related to width and depth of plot by a power function. The total price is proportional to the total quality that can be expressed as the sum of two power functions, P = a*Wb + c*Dd with (0 ≤ b ≤ 1; 0 ≤ d ≤ 0). The cost for providing this land is estimated in this example based on the cost per running meter 5

RC43 conference, At home on the housing market, Amsterdam, Netherlands July 10-12, 2013

street of a given type (s in VND/m) and cost per meter square of land (p in VND/m2) C = W*s + W*D*p (Figure 5).

Figure 6 Cost and profit of residential plots. For each of the considered combinations of W and D the margin can be visualized with M = P - C (Figure 6). This three dimensional graph can be transformed in a two dimensional if on the left-right axis the cost of providing the land is represented and on the other axis is the price. For each combination of W and D a point can be calculated. If this is done for several combinations, we obtain a cloud of points of which points on the solid border line at the top and above the line price is equal to cost are interesting. Under this line the costs are higher than the selling values and thus combinations are not interesting for the developers. For every point one can conclude that a point more to the right or/and below is not interesting because it generates more costs and/or less profit. Starting from the left, most points on the dash border line one can analyze what is the best additional spending for a developer. P

P Price = Cost

D

Price = Cost

C P = aWb +cDd

(W*s) + (W*D*p)

C

B A C

Figure 7 Graphical representation of price and cost with two parameters: width and depth of plot. Stepping from A to C will generate a higher additional income per additional cost than from A to B. From each point the next point is selected with the highest slope illustrated with the solid line (Figure 7). This reasoning can be repeated from the next point and so on. The new solid line has the slope that is less and less, generated by the diminishing marginal utilities. The last line will be called Pareto sub optimal points, the pervious Pareto optimal points. This approach can be extended to cases where more than two characteristics are considered. The cost may depend in a complex inter related way of the physical parameters.

6

RC43 conference, At home on the housing market, Amsterdam, Netherlands July 10-12, 2013

Deriving preferences from the web data and test for small scale plot Data source selecting There are two typical methods to develop insight in housing preferences: “revealed” and “stated” preference method. The revealed method requires historical data sets and tries to predict prices based on real transactions (Rosen, 1974 and Goodman, 1978). While the stated method with questionnaires whereby users express their willingness to pay for selected attributes. In Vietnam, there are four data sources for residential land prices. First, the data of residential land transactions can be obtained from government organizations. However, these data cannot explain actual housing preferences, because land prices in norms of government are lower than in real estate market (Thien Thu & Perera, 2011). Second, transactions in private sector, these data are secret because of competition, so they are not accessible for research. Third, using questionnaires, this consumes a lot of time and funds. Fourth, data on internet can be collected by low cost, fast and based on a large data set. The advertisings of the residential land are primarily posted by real estate agencies and the owners. The advertising messages contain selling prices of residential lands. The characteristics are depth, width and road width. The location attributes describe possibilities to connect with relevant social and service facilities. This information is free for all potential buyers. However, the information collected from selected websites has some inherent limitations. The information is structured in different ways. Sometimes characteristics are omitted and some information may be described in a very positive way to attract the buyers for example a bird flights distance instead of distance based on street pattern. This study uses data contained in the internet advertisements on residential plots offered for sale in Cantho. The data have been downloaded from three Vietnamese realestate websites: batdongsan.com.vn, canthoinfo.com, mekongland24h.com.vn. There are, of course, other sites, offering similar information. However, due to their high market share, these three sites are representative. For instance, “batdongsan.com.vn” posts 240 messages for selling lands and houses per month. While the “canthoinfo.com” posts 5750 selling messages of both lands and buildings per month. Data collection During the data collection stage, some inconsistencies of data have to be identified because the data quality could significant affect the analysis later on. Therefore, some cleaning processes were employed to remove messages that lack information on location or attributes. A program was written to download data and update automatically. There are three criteria that are employed to select the land price data. Firstly, one full year of data was used to avoid problems with time variations over a long period. Secondly, the data were chosen at the new large scale housing projects in Cantho city. Thirdly, plots located along main roads are excluded because these roads connect to road network of the city. The main roads can generate commercial activities for lands that are located along them. Therefore, there are 146 selected residential plots that are used to analyze in the model.

7

RC43 conference, At home on the housing market, Amsterdam, Netherlands July 10-12, 2013

Residential land attributes and attribute selection Table 1 attributes of residential land for detached and terraced housing unites ID Attribute name 0 Price 1 Width 2 Depth 3 Road width 4 Circulation 5 Commercial area 6 Apartment 7 Terraced house 8 Detached house 9 Orientation 10 Social area 11 Nearest city centre 12 Park

Unit 106 VND m m m % % % % % 1 to 8 % 1/km %

Variable name pri wid dep roa cir com apa ter det ori soc nea par

Description Price of residential land Width of the residential land. Depth of the residential land. Width of road in front of residential land. % Circulation or road area of the project % Land to build the commercial facilities. % Land to build the apartments % Land to build the terraced houses. % Land to build the detached houses. 8 orientations in step of 45o, from North % Land for hospital, school and park Inverse of distance to city center % Green or water area within projects

Table 2: Attribute selection by using multiple linear regression method Attribute name Intercept Wid Dep Roa Com Apa Ter Det Soc Cir Or Nea Par R Square Akaike information criterion (AIC) Significance F *p