Shlomie Hazam

1

The Effect of Terror on Behavior in the
Jerusalem Housing Market

• Shlomie Hazam
• Daniel Felsenstein

Funded by the German-Israel Fund Institute of
Urban and Regional Studies, Hebrew University of
Jerusalem
2
Objectives

• Descriptive Terror Patterns
• Has center of gravity moved over time?
• Can we identify terror Hot Spots
• Terror over time Increasingly random or
  clustered?
• Analytic Modeling Impact of Terror
• Effect of Terror on House Prices and Rents?
• Do Spatial Spillover Effects Exist?

3
Theory

• Terror Generates (1) Risk (2) Fear. Becker and
  Rubinstein 2003.
• Risk(ß) numerical probability. Not sufficient
  to change behavior patterns. BUT in combination
  with FEAR can have great impact on behavior (SARS
  mad-cow disease)
• Fear(?) subjective, different threshold,
  accommodation levels.
• Y1t aß?µ1t
• Y2t aßµ2t
• t1.T

Y observed behavior ß risk ? fear µ
unexplained factors
4
Model

• Macro Effects
• interest rates
• permanent income

• Property Characteristics
• housing conditions
• housing quality

House Prices

• Neighborhood Characteristics
• population density
• economic level
• distance from the seam line

Rental Prices

• Terror
• terror attacks

5
Estimation Model

• (levels) Pi1 a1ß1X1µi1 µi1
  Vi ei1
• (levels) Pi2 a2ß2X2µi2 µi2
  Vi ei2
• (differences) ?P a2 - a1 ß2X2 - ß1X1 ? µ

V neighborhood attributes e property
attributes
6
Data

• Terror Incident Data Police Diaries
• House Prices and Rents Levi Yitzhak Guide
• Terror Monetary Damage Data Property Tax
  Bureau
• G.I.S. Data Assignment

7
Terror attacks which took place over the periods
2000-2003 (with 1990, 1995 benchmarks) Most of
the attacks are located in the vicinity of the
seam line. Note the infiltration of attacks on
the west side of the seam line over period
2002-2003.
8
Data cont.- G.I.S. Method

• Data Standardization (average price per meter in
  )
• Price Assignment to G.I.S. street/Buildings
  cover.
• Examining Spatial Geographic Weighted Means (hot
  spots)

9
Dwelling prices by street. Green color stands for
the cheaper streets, and red color stands for the
expensive ones.
10
The price information was attached to each of the
buildings on every street. This procedure is
necessary for creating price surfaces ( to be
presented)
11
Red zones, the most expensive areas in the city,
are located in the west and in the center of
Jerusalem. Green zones, the cheaper areas, are
located in the vicinity of the seam line and in
the peripheral neighborhoods.
12
In 2004 real estate prices were lower than in the
1999, due to global processes and the high-tech
bust. The distribution of the dwelling prices
changes mainly in the marginal areas, which
became cheaper. The citys center remains
expensive.
13
The difference in dwelling prices between 1999
and 2004 (accounting for the real estate price
index). The green areas presents a rise in the
prices and the red areas presents a decline.
14
GIS Descriptive Results

• Descriptive Patterns of Terror (movement of
  center of gravity, creation of hot spots,
  increasing randomization)
• Spatial Changes in House / Rental Prices
• The Factors that Affected House / Rental Prices

15
1. Movement of Center of Gravity
The main mass of terror attacks was in the city
center. In the next map we calculated the
geographic center of terror attacks of each year.
The square symbol points in the map, present the
geographic center of all of the recorded attacks
of a single year, and the triangle symbol points
present the weighted mean center of each year.
The weighting factor is the number of casualties
Weighted Mean uses the following equations to
calculate the weighted mean center of a cluster
of points
16
The geographic center of the terror attacks in
both cases is in the city center and in the
vicinity of the seam line. The movement of the
mean points over time is in the general direction
of north-south (seam line). The most crowded
areas in the city, with the highest number of
casualties are not dwelling areas, but the
central business district of Jerusalem.
17
2.Terror Intensity (Hot Spots)
In order to find where were the most intense
terror activity in the city in terms of
causalities, we used the GIS neighborhood
statistics function. This function computes a
statistic raster based on the value of the
processing cell and the value of the cells within
a specified neighborhood.
18
We computed the sum of the casualties in the
radius of 500m from each attack point. We notice
that the city center and the seam line zone
suffered the most nearly 200-400 casualties per
square km. Other significant areas were the
marginal neighborhoods Neve Yaakov, the French
Hill, and Gilo which suffered up to 100
casualties per square km.
19
3. Terror Over Time Clustered or Random?

• The G statistic (Getis and Ord 1992) measures
  concentrations of high or low values for an
  entire study area
• randomgtfear factorgtconsumer behaviorgthousing
  prices

• where is the value of i point,
• is the weight for point i and j for distance
  d

20
1990
2001
1995
2000
2001
2002
Observed General G 0.00037921665857368672 Expect
ed General G 0.00033026303185345842 General G
Variance 1.4722722834068171e-009 Z Score
1.2758241065266294 Standard Deviations
2003
21
Spatial Changes in House Prices in relation to
terror activity

• Surface Interpolating
• Visiting every location in a study area to
  measure the prices is difficult. Instead, we use
  the input point locations, and a predicted value
  can be assigned to all other locations.
• By interpolating, we predict prices values
  between these input points.

22
Several interpolation methods were tested The
best results were obtained by Kriging
interpolation - that assumes nearby dwelling
price points have similar values and that the
distance or direction between sample points shows
spatial correlation that helps to describe the
surface. (this is the logical price structure of
neighborhoods).
23
The output interpolated grid of 1999, shows that
there are relatively expensive dwelling areas
(colored orange/red) in the some of the marginal
neighborhoods.
24
The output interpolated grid of 2004, shows
that the relatively expensive dwelling areas in
the marginal neighborhoods of 1999 map
disappeared and now are cheaper. Other areas in
the western city became more expensive.
25
The following map shows the interpolated grid of
the difference in dwelling prices between
1999-2004, over background buffers from the seam
line. The terror attacks are the black points.
The red zones are the areas where prices were
lower in 2004. These are the marginal
neighborhoods, which suffered most of the terror
attacks.
26
The height in the 3D map is presented by
z-values of the difference in dwelling prices
between 1999-2004 . The steep mountains are
the peripheral neighborhoods. The following map
shows this result from different angle.
27
South East View
Ramot
the old city
Talpiot
Gilo
Armon HaNatziv
28
South West View
the old city
Ramot
Talpiot
Armon HaNatziv
Gilo
29
Correlating price data and terror activity data
Zonal Overlay Statistics Zonal functions take a
value raster as input and calculate for each cell
some function or statistic using the attack value
for that cell and all cells belonging to the same
attack zone. Zonal functions quantify the
characteristics of the geometry of the input
zones.
30
Distribution of average decline in price by
terror type of activity
31
Running a regressionof points of terror
(intensity) on prices points derived from the
grid, produced non significant explanation with
great errors
Using price points from the grid
This led us to enlarge the unit of investigation
to the statistical zone (i.e. neighborhoods)
32
(No Transcript)
33
(No Transcript)
34
(No Transcript)
35
analysis OLS

• Distance to the seam line had a negative affect
  on prices, but insignificant.
• The variables population density and housing
  conditions had a positive and significant affect
  on housing prices, as expected.

• Terror has a negative affect on housing prices.
  Larger and more significant for rental prices
  than purchase prices.
• Terror intensity (measured by casualties and
  damage) had a lager, significant and positive
  impact on housing prices in 2004 than in 1999,
  contrary to our expectations.

• The Lagrange tests implies spatial
  autocorrelation , therefore we should run spatial
  lag regressions

36
Spatial autocorrelation

• Spatial autocorrelation is when the value at one
  point in space is dependent on values at the
  surrounding points. That is, the arrangement of
  values is not just random. Positive spatial
  correlation means that similar values tend to be
  near each other.
• We model spatially dependent data by using
  Spatial Lag Model which estimates for an effect
  of neighboring areas.

37


38
Regression Spatial Lag

• The new dependent variable is the housing prices
  level in neighbor statistical areas.
• .Housing prices are negatively affected by
  terror. Rental prices were more significantly
  affected as in the OLS model.
• Significant negative lag effect neighboring
  prices lower prices in the statistical area. Due
  to the unique, non continuous nature of
  Jerusalem housing market?.

39
Conclusions

• Descriptive results
• Most attacks took place in the peripheral
  neighborhoods. A spatial pattern of terror
  exists unarmed attacks and stabbings exists in
  the vicinity of the seam line, shootings mainly
  in South (Gilo) and suicide bombing in crowded
  areas, especially city center.
• Geographic center of gravity for terror events
  shifted over time towards the seam line.
• Neighborhood statistics method emphasized the
  vulnerability of the city center and creation of
  hot spots
• The G statistic shows that terror became
  increasingly random over the course of time. This
  increases the fear factor.

40
Conclusions

• Analytical Results
• Terror has a significant and negative impact on
  housing prices. Greater significance for rental
  than purchasing behavior. Shows fear as main
  component of terror. More likely to be expressed
  in short term behavior (rental) than in long term
  (purchasing).
• Population density and housing conditions have a
  positive and significant affect on housing
  prices, as expected
• Significant negative Moran's I coefficient the
  impact of terror on housing prices is not clean
  it is also affected by neighboring statistical
  areas
• Surprising negative and significant spatial lag
  effect on purchase and rental prices. Perhaps due
  to the unique, non continuous nature of the
  Jerusalem housing market?