Alternative Approaches to Modelling Housing Market Segmentation: Evidence from Istanbul

1

• Alternative Approaches to Modelling Housing
  Market Segmentation Evidence from Istanbul
• Berna Keskin (Ph.D Candidate)
• Town and Regional Department
• The University Of Sheffield
• Primary Supervisor Prof. Craig Watkins
• Secondary Supervisor Dr. Cath Jackson

2
Introduction Aim Objectives

• Aim
• The content of this research is to understand the
  spatial distribution of housing prices. The
  main aim of my research is to compare the
  effectiveness of different models of house
  prices that captures segmented price difference
  in Istanbul.
• Objectives
• To examine the best way to conceptualize the
  structure of owner occupied housing market
• To identify the strengths and the weaknesses of
  the segmented model structures
• To examine relationship between locations and
  housing prices
• Approach
• A standard hedonic model (market-wide model)
• A segmented model (using segmentation dummies in
  market-wide model)
• A multi-level model which includes segments and
  their interactions with each other and other
  spatial influences.

3
Motivation of the Study

• Segmented Market structure
• Housing market in Istanbul are highly segmented
• There are significant price differences, in
  different parts of the market for homes with the
  same physical features and locational attributes
• Population 10,033,478.
• Istanbul population/Turkey 14.78 in 2000
  (TUIK,2006), surpasses the population of 22 EU
  countries (Eurostat).
• 2,550,000 households and 3,391,752 housing units
• The problems
• high increase rate in population,
• the gap in the incomes
• lack of enough amounts of residential plots.
• land rent and speculation.

4
Housing Prices Per m² in Istanbul in 2000
5
Data
Variables
Property Characteristics
Socio-economic Characteristics
Neighbourhood Characteristics
Locational Characteristics
1.Housing Type 2. Rooms 3. Floor Area 4. Elevator
5. Garden 6. Balcony 7. Storey 8. Site 9. Age

• Income
• Household size
• Living period in the neighbourhood
• Living period in Istanbul

• Satisfaction from
• School
• Health service
• Cultural facilities
• Playground
• Neighbour
• Neighbourhood quality

• Earthquake risk
• Continent
• Travel time to shopping centres
• Travel time to jobs and schools

Italic variables are excluded due to
multicollinearity.
6
Market Wide Model (1st stage)

• Hedonic modelling technique
• the price of housing unit as a dependent
  variable, and
• the structural, locational

7
2nd Stage The Effects of the Segments

• Hedonic model with spatial dummy variables as a
  proxy for segments
• The need for the 2nd stage effectiveness of
  market-wide model.
• So
• Segmentation is added into the hedonic model as a
  dummy variable.
• Segmentation is determined in 3 ways
• A priori identification (5 submarket)
• Experts identification (5 submarket)
• Cluster Analysis (12 submarket)

8
2nd Stage The Effects of the Segments (A priori)

• A priori segmentations which are considered to
  be the most probable.
• Five segmentations were chosen by taking account
  of
• Housing prices
• Housing types
• Location
• Size
• Age
• Income
• Living period
• Neighborhood quality
• 1st SUBMARKET Waterside house (along bosphorus
  , literally called as yali), gated communities,
  residences, low storey apartments by the shore,
  detached houses close to the city centers.
• 2nd SUBMARKET Apartment blocks mostly
  constructed after 80s (liberal economy),
  built-sell apartments and luxury sites.
• 3rd SUBMARKET Apartment blocks and
  detached/attached houses in historical areas.
• 4th SUBMARKET Apartments blocks mostly
  constructed in 2000s, built-sell apartments and
  cooperatives.
• 5th SUBMARKET Squatter settlements, old summer
  houses (apartments)

9
2nd Stage The Effects of the Segments (a priori)
10
2nd Stage The Effects of the Segments (Experts
identification)

• segmentations which are determined by experts.
• 10 interviews were done with real estate
  managers.
• 7 maps were drawn by experts and 5 submarkets
  were identified mainly focusing on the housing
  prices.

11
2nd StageThe Effects of the Submarkets (Cluster
Analysis)

• Cluster Analysis is done in order to group the
  neighborhoods into submarkets.
• 12 clusters are displayed by the programme by
  taking account of these variables
• Housing prices
• Floor area
• Age
• Rooms
• Income of households
• Living period in Istanbul
• Neighborhood quality
• Travel time to jobs, school, shops
• Transportation satisfaction
• Earthquake Risk

12
2nd Stage The Effects of the Submarkets (Cluster
Analysis)
13
Comparison of Models
Basic Hedonic Model P f ( Fa, I, Lp, -Eq, S,
A, Ls, N) Fa Floor Area S Site A
Age Ls Low Storey I Income of the
household Lp Living Period in Istanbul N
Neighbor satisfaction Eq (-)Earthquake
Damage Rsquare 0.60
Hedonic Model with a priori Submarket
Variables P f ( Fa, I, Lp, -Eq, S, A, C, N,Sm1,
Sm3, -Sm4, -Sm5) Fa Floor Area S Site A
Age C Continent I Income of the
household Lp Living Period in Istanbul N
Neighbor satisfaction Eq (-)Earthquake
Damage Sm1 1st submarket Sm3 3rd
submarket Sm4 (-)4th submarket Sm5 (-)5th
submarket Rsquare 0.67
Hedonic Model (experts) submarket variables P
f ( Fa, Ls, Lp, HS, A,Sm1, -Sm3, -Sm4, -Sm5 Fa
Floor Area S Site A Age Lp Living
Period in Istanbul Hs Household size Sm1 1st
submarket Sm3 3rd submarket Sm4 (-)4th
submarket Sm5 (-)5th submarket Rsquare 0.68
Hedonic Model with Cluster Submarket
Variables P f ( Fa, I, Lp, Eq, S, A, C, Sc,Sm4,
Sm5, Sm7, -Sm8) Fa Floor Area S Site A
Age I Income of the household Lp Living
Period in Istanbul Sc School satisfaction Eq
(-)Earthquake Damage Sm4 4th submarket Sm5
5th submarket Sm7 7th submarket Sm8 (-)8th
submarket Rsquare 0.64
14
Multi-level modelling

• multilevel modeling how the individual level
  (micro level) outcomes are affected by the
  individual level variables and group level (macro
  level or contextual level) variables.
• multi-level modelling provides assessing
  variation in housing prices at several levels
  simultaneously

15
Contextual Level of Multi-level Modelling

• Segmentation is added into the multi-level model
  as level 2
• Segmentation (Level 2-macro level-contextual
  level) is determined in 3 ways
• A priori identification (5 submarket)
• Experts identification (5 submarket)
• Cluster Analysis (12 submarket)

16
Multi-level modelling (comparison)
2 level model Estimated Variance Standard Error Intra class correlation
Submarket (a priori) 0.0961 0.03467 0.23
Housing Unit 0.1785 0.90941 0.77
2 level model Estimated Variance Standard Error Intra class correlation
Submarket (experts) 0.1266 0.045476 0.34
Housing Unit 0.17462 0.9719 0.66
2 level model Estimated Variance Standard Error Intra class correlation
Submarket (cluster) 0.132033 0.037396 0.34
Housing Unit 0.1820813 2.762246 0.66
17
Multi-level modelling (a priori)
18
Multi-level modelling (experts)
19
Multi-level modelling (cluster analysis)
20
Effectiveness of models
21
Effectiveness of Models
22
Conclusions

• Housing submarkets matter in explaining the
  structure of the urban housing market system.
• From the three-stage methodology different
  models have different effectiveness. However the
  submarket aggregation plays an important role in
  the improvement of the models.
• Models were performing better with the expert
  identified submarket dummies are employed.
  Experts have a better, realistic and more
  detailed information about submarkets rather than
  a priori or statistical tools.
• To overcome the problems of hedonic models,
  multi-level modelling approach may be a solution.
  Multi-level modelling can be an alternative
  method to capture and model the housing system.