Is it worth identifying service employment (sub)centres for modelling apartment prices? The case of Lyon, France LET, Transport Economics Laboratory (CNRS, University of Lyon, ENTPE) ERES conference 2009 Stockholm KTH Marko Kryvobokov
Jan 06, 2016
Is it worth identifying service employment (sub)centres
for modelling apartment prices?
The case of Lyon, France
LET, Transport Economics Laboratory(CNRS, University of Lyon, ENTPE)
ERES conference 2009 Stockholm KTH
Marko Kryvobokov
2
ERES conference 2009 Stockholm KTH
1. Introduction
URBAN CENTRES vs.
ALL TERRITORIAL UNITS
in hedonic price model
3
ERES conference 2009 Stockholm KTH
1. Introduction
Identification of urban centres
- Generalization – creation of higher order objects from lower order objects
- von Thünen, Alonso, Wingo, Wendt, Harris and Ullman…
- McDonald (1987): an urban center represents a distinct zone whose employment density exceeds the density of its adjacent neighborhood and whose size is sufficiently large to potentially impact the urban land and/or property market
- McDonaln (1987): employment subcentres as secondary peaks in the employment density and the employment-population ratio
4
ERES conference 2009 Stockholm KTH
1. Introduction
Identification of urban centres
McMillen (2001), McMillen and Smith (2003): the first stage: potential subcentres have significant
residuals in the locally weighted regression of employment density on distance from the CBD;
the second stage: check if they provide significant explanatory power in a semiparametric employment density regression
5
ERES conference 2009 Stockholm KTH
1. Introduction
Identification of urban centres
Empirical examples in the real estate literature:
- Söderberg and Janssen (1999): re-estimate regression for apartment properties in Stockholm changing the precise location of the CBD with the step of 50 meters
- Sivitanidou (1996): application of the definition of McDonalds for office-commercial real estate in Los Angeles
- McDonald and McMillen (1990), McMillen (1996): land values in Chicago in 1836-1990
6
ERES conference 2009 Stockholm KTH
1. Introduction
Accessibility and centrality
Des Rosiers and Thériault (2008):
accessibility is the ease with which persons, living at a given location, can move to reach activities and services which they consider as most important.
It is distinct from
centrality, which relies on structural features and relates to proximity to urban amenities
7
ERES conference 2009 Stockholm KTH
1. Introduction
All territorial units
- Thériault et al. (2005) and Des Rosiers and Thériault (2008): hedonic modelling of real estate prices with centrality and accessibility indices; accessibility index, based on interview and fuzzy logic criteria, far outweigh the centrality index in Quebec city
- With fast development in GIS and transportation analysis software, in principle, all territorial units in a city can be focused. Do we still need generalization, i.e. identification of urban centres?
8
ERES conference 2009 Stockholm KTH
2. Identification of service employment centres
0 10 205 Kilometers
The Lyon Urban Area +:
812 zones (IRISes)3,723 sq. km
1,904 thousand inhabitants (2005)
9
ERES conference 2009 Stockholm KTH
2. Identification of service employment centres
Two origine-destination (O-D) matrices of travel times from the MOSART transportation model for the Lyon Urban Area (2007), a.m. peak:
- cars- public transport (N. Ovtracht and V. Thiebaut, LET)
As in McMillen (2001), we run a simple regression model of service employment density on travel time to Bellecour-Sala (the CBD)
15 zones have positive standardized residuals higher than 3.3
10
ERES conference 2009 Stockholm KTH
2. Identification of service employment centres
The pre-identified service employment centres
14
3
8
9
4
13
116
5
10
7
1512
1
2
0 0.5 10.25 Kilometers
11
ERES conference 2009 Stockholm KTH
3. Centrality index
– attraction of zone j (either service employment density or service employment to population ratio);
N
j ij
ji tt
ACI
1
jA
ijtt – travel time from zone i to zone j;
N – number of zones
12
ERES conference 2009 Stockholm KTH
3. Centrality index
0 6 123 Kilometers
Centrality index
466.92 - 1181.14
1181.15 - 2400.61
2400.62 - 2981.00
2981.01 - 3521.00
3521.01 - 4713.31
0 6 123 Kilometers
Centrality index
282.21
282.22 - 834.18
834.19 - 1933.40
1933.41 - 3479.00
3479.01 - 7903.88
Clusters of centrality index for cars with service employment density
Clusters of centrality index for cars with service employment to population ratio
13
ERES conference 2009 Stockholm KTH
4. Accessibility index
– 50th percentile of the observed travel time from travel survey;
N – number of zones
As in Thériault et al. (2005): suitability index Sij for travelling from zone i to zone j1ijS 50Cttij
5090
501CC
CttS ij
ij 9050 CttC ij
0ijS 90Cttij
ijtt – travel time from zone i to zone j;
50C
– 90th percentile of the observed travel time from travel survey90C
N
jjiji ASAI
1
– attraction of zone j (either service employment density or service employment to population ratio);
jA
14
ERES conference 2009 Stockholm KTH
4. Accessibility index
Clusters of accessibility index for cars with service employment density
Clusters of accessibility index for cars with service employment to population ratio
0 6 123 Kilometers
Accessibility index
118.03 - 2871.12
2871.13 - 5497.00
5497.01 - 7457.00
7457.01 - 9667.30
9667.31 - 14822.52
0 6 123 Kilometers
Accessibility index
20.63 - 1387.71
1387.72 - 2581.95
2581.96 - 3485.00
3485.01 - 4602.00
4602.01 - 7099.83
15
ERES conference 2009 Stockholm KTH
5. Hedonic model of apartment prices
Lyon
16
ERES conference 2009 Stockholm KTH
5. Hedonic model of apartment prices
Data from Perval: 4,362 apartments sold in 1997-2008
Location: mainly in Lyon and Villeurbanne
Number of rooms: 1 to 9
Apartment priceper square metre,Euros
0 1 20.5 Kilometers
317 - 1200
1201 - 1800
1801 - 2500
2501 - 3000
3001 - 5112
17
ERES conference 2009 Stockholm KTH
5. Hedonic model of apartment prices
Apartment variables:
- dummies for year of transaction- apartment area- dummies for number of bathrooms- dummies for number of parking places- dummies for floor- dummies for period of construction- dummies for apartment’s state (conditions)- dummies for the quality of view- dummies for number of cellars- dummy for existence of garden- dummy for existence of terrace
18
ERES conference 2009 Stockholm KTH
5. Hedonic model of apartment pricesLocation variables:- dummy for location within a 100 m buffer of water- dummy for location in an ad hoc district- % middle income households- % high income households- travel times by car to each of the 15 pre-identified centres- travel times by public transport to each of the 15 pre-identified
centres- centrality index for cars with service employment density- centrality index for cars with service employment to population ratio- centrality index for public transport with service employment density- centrality index for public transport with service employment to
population ratio- accessibility index for cars with service employment density- accessibility index for cars with service employment to population
ratio- accessibility index for public transport with service employment
density- accessibility index for public transport with service employment to
population ratio
19
ERES conference 2009 Stockholm KTH
5. Hedonic model of apartment prices
Dependent variable: log price
42 or 43 independent variables
OLS regression:- global- geographically weighted regression (GWR) (Brunsdon et al., 1996) GWR with a Gaussian error term; fixed kernel type
After the first global OLS run, observations with standardised residuals
higher than 3 were deleted. 4,308 observations remained
Variance inflationary factor (VIF) checks multicollinearity
Moran’s I measures spatial autocorrelation
20
ERES conference 2009 Stockholm KTH
5. Hedonic model of apartment prices
Examination of the influence of the pre-identified centres:- global model without travel times - travel time to each of the pre-identified centres is added one
at a time; fifteen global models for each transport mode- sorting their adjusted R-squared high to low, all fifteen
variables are added to the equation and then excluded one by one from the bottom until it is obtained a model with acceptable VIF
- the best global models include two centres
Global model with travel time to the CBD only
Global model with centrality index
Global model with accessibility index
GWR models for the same cases
21
ERES conference 2009 Stockholm KTH
5. Hedonic model of apartment prices
Extracted location variables
Adjusted R2 Coefficient t-value Maximum VIF Moran’s I
TT_C_10TT_C_3
0.880 (0.896) -0.163 (-0.149)-0.126 (-0.104)
-22.965-13.612
6.307 0.29 (0.20)
TT_C_10TT_C_6
0.880 (0.896) -0.094 (-0.097)-0.155 (-0.122)
-9.439-13.562
6.305 0.29 (0.20)
TT_C_3 0.865 (0.892) -0.196 (-0.130) -21.188 6.301 0.36 (0.23)
CI_C_SD 0.876 (0.893) 0.007 (0.006) 29.490 6.298 0.30 (0.21)
CI_C_SP 0.860 (0.891) 0.010 (0.009) 17.144 6.306 0.36 (0.22)
AI_C_SD 0.873 (0.893) 0.004 (0.005) 27.266 6.304 0.32 (0.22)
AI_C_SP 0.867 (0.892) 0.003 (0.003) 22.699 6.312 0.34 (0.22)
Global regression and GWR for cars
22
ERES conference 2009 Stockholm KTH
5. Hedonic model of apartment prices
Extracted location variables
Adjusted R2 Coefficient t-value Maximum VIF Moran’s I
TT_PT_10TT_PT_6
0.874 (0.892) -0.114 (-0.123)-0.132 (-0.074)
-10.737-10.238
6.307 0.31 (0.22)
TT_PT_10TT_PT_3
0.873 (0.894) -0.154 (-0.147)-0.089 (-0.059)
-17.629-8.266
6.309 0.31 (0.21)
TT_PT_3 0.864 (0.892) -0.192 (-0.134) -20.695 6.298 0.35 (0.22)
CI_PT_SD 0.867 (0.891) 0.007 (0.005) 23.514 6.299 0.33 (0.22)
CI_PT_SP 0.859 (0.890) 0.012 (0.010) 16.638 6.309 0.36 (0.22)
AI_PT_SD 0.867 (0.891) 0.004 (0.005) 23.449 6.309 0.34 (0.22)
AI_PT_SP 0.867 (0.887) 0.003 (0.003) 23.052 6.301 0.34 (0.25)
Global regression and GWR for public transport
23
ERES conference 2009 Stockholm KTH
5. Hedonic model of apartment prices
The highlighted centres:
3 – Bellecour-Sala10 – Les Belges
14
3
8
9
4
13
116
5
10
7
1512
1
2
0 0.5 10.25 Kilometers
24
ERES conference 2009 Stockholm KTH
5. Hedonic model of apartment prices
The highlighted subcentres:
6 – Jussieu10 – Les Belges
14
3
8
9
4
13
116
5
10
7
1512
1
2
0 0.5 10.25 Kilometers
25
ERES conference 2009 Stockholm KTH
5. Hedonic model of apartment pricesApplication of principal component analysis
Extracted location variables
Factor number % of variance
Adjusted R2 for factor 1
and othervariables
Moran’s I for factor 1
and other variables
Adjusted R2 for all factors
CarsTravel times to 15
centres 1 18.6 0.880 0.29 0.606
Travel times to 3 centres 1 6.8 0.850 0.40 0.607
Public transportTravel times to 15
centres 1 22.1 0.871 0.32 0.601
Travel times to 3 centres 1 8.3 0.854 0.38 0.618
26
ERES conference 2009 Stockholm KTH
6. Conclusions
The best results for travel times were obtained with three centres: Bellecour-Sala, Les Belges, and Jussieu.
Among them, it is difficult to find a leader.
Duocentric models are better than the monocentric one.
Centrality index and accessibility index behave differently in comparison with each other, but in most cases outperform the monocentric model.
Both global and GWR models with travel times to two centres, either with or without the CBD, are the best among all, including centrality and accessibility indices.