The Effect of Terror on Behavior in the

The Effect of Terror on Behavior in the Jerusalem Housing Market Daniel Felsenstein Shlomie Hazam Funded by the German-Israel Fund (GIF) Institute of Urban and Regional Studies, Hebrew University of Jerusalem

Objectives Descriptive: Terror Patterns n G. I. S. depiction of Terror Incidence n Identification of Terror Hot Spots and Intensity Analytic: Modeling Impact of Terror n Effect of Terror on House Prices and Rents n Spatial Spillover Effects

Theory Terror Generates (1) Risk (2) Fear. [Becker and Rubinstein 2003]. Risk(β) = probabilities very small - BUT small probabilities events can have great impact. (SARS mad – cow disease) DUE TO: Fear(γ) = subjective, different threshold, accommodation levels. Y= observed behavior Y 1 t = α+β+γ+μ 1 t Y 2 t = α+β+μ 2 t t=1…. T β = risk γ = fear μ = unexplained factors

Empirical Evidence Effect of Terror: Behavior with respect to: Public Transport Demand Side Supply Side Passengers 20% decrease [High Frequency: no change] [Low Frequency : great change] Drivers No differences in compensation No greater likelihood to quit Security Guard Employment 10% increase in demand [likelihood to enter employment] No change in compensation Coffee Shops Negative Effect [High Frequency: no change] [Low Frequency : great change] ------ Long Term: small effect (purchasers) Short Term: significant effect (renters) ------ Housing Market

Macro Effects • interest rates • permanent income Property Characteristics • housing conditions • housing quality Neighborhood Characteristics • population density • economic level • distance from the seam line Terror • terror attacks Model House Prices Rental Prices

Model Estimation (levels) Pi 1 = α 1+β 1 X 1+μi 1 | [μi 1 = Vi + εi 1] (levels) Pi 2 = α 2+β 2 X 2+μi 2 | [μi 2 = Vi + εi 2] (differences) ΔP = α 2 - α 1 + β 2 X 2 - β 1 X 1+ Δ μ V = neighborhood attributes ε = property attributes

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

This map presents the terror attacks which took place during the years 2002 -2003 (and two bench marks 1990, 1995). Most of the attacks are located in the vicinity of the “seam line”. We notice an infiltration of attacks on the west side of the seam line during 2002 -2003.

Total Terror Casualties by Years and Type of Attack Data 1990 1995 2000 2001 2002 2003 Total arson Casualties 0 0 0 0 Attacks 0 0 0 1 1 0 2 attack Casualties 7 3 6 6 2 0 24 Attacks 19 5 3 5 4 0 36 explosive device Casualties 16 111 12 435 614 306 1494 Attacks 24 1 3 34 30 6 98 grenade Casualties 7 0 0 0 7 Attacks 14 0 4 1 0 0 19 molotov cocktail Casualties 0 0 28 0 0 0 28 Attacks 0 0 7 1 2 0 10 mortar bomb Casualties 0 0 0 1 Attacks 0 0 0 6 2 0 8 shooting Casualties 0 1 4 75 75 4 159 Attacks 1 1 6 24 11 2 45 stabbing Casualties 24 10 11 8 3 2 58 Attacks 29 14 10 9 5 2 69 Total Casualties 54 125 61 525 694 312 1771 Total Attacks 87 21 33 81 55 10 287

G. I. S. Method Data Standardization (Size and Money Values) n. Price Assignment to G. I. S. street/Buildings cover. n Spatial Geographic Weighted Means n Delta– House Prices + Rentals 1999 -2004 n

This map clearly expresses the dwelling prices in each of the streets. The green color stands for the cheaper streets, and the red color stands for the expensive ones.

The price information was attached to each of the buildings on every street. This procedure is significant for creating price surfaces which will be presented later on. street

The red zones, the most expensive areas in the city, are located in the west and in the center of Jerusalem. The green zones, the cheaper areas, are located in the vicinity of the seam line and in the marginal neighborhoods of Jerusalem.

In the year 2004 the real estate price level in Israel was lower than in the year 1999, due to global processes and the burst of the high tech bubble. The distribution of the dwelling prices changes mainly in the marginal areas, which became cheaper. The city’s

This map shows the difference in dwelling prices between 1999 and 2004 (the real estate price index was taken into an account). The green areas presents a rise in the prices and the red areas presents a decline.

Findings Descriptive Patterns of Terror n Spatial Distribution of Changes in House + Rental Prices n Factors Affecting House + Rental Prices n

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 :

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. The most crowded areas in the city, with the highest number of casualties are not dwelling areas, but the central business

Zooming in to the south part of the city shows us the difference in dwelling prices more clearly: the prices in marginal neighborhoods in 2004 were much lower than in 1999 (abu-Tor, Talpiot, Arnona, Armon Ha’Natziv). In the western city (Rasko, Katamonim, pat) prices were higher in 2004 than in 1999. The margin areas were those who

Terror Intensity we used the neighborhood statistics GIS function. This function computes an output raster where the value at each location is a function of the input cells in some specified neighborhood of the location. For each cell in the input raster, the Neighborhood Statistics function computes a statistic based on the value of the processing cell and the value of the cells within a specified neighborhood, then sends this value to the corresponding cell location on the output raster.

We computed the sum of the casualties in the radius of 500 m 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 Ya’akov, the French Hill, and Gilo which suffered up to 100 casualties per

Terror Over Time: Clustered or Random? The G statistics were developed by Getis and Ord (1992). This measures concentrations of high or low values for an entire study area where is the value of i point, is the weight for point i and j for distance d

1990 2001 1995 2000 2001 2002 Observed General G = 0. 00037921665857368672 Expected General G = 0. 00033026303185345842 General G Variance = 1. 4722722834068171 e-009 Z Score = 1. 2758241065266294 Standard Deviations 2003

Spatial Descriptive of Changes in House Prices Surface Interpolating n Kriging Interpolation Method n Statistic areas Distributions n

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, the prices values between these input points will be predicted.

Several interpolation methods were tested – The best results were obtained by Kriging interpolation, that assume the distance or direction between sample points shows spatial correlation that helps describe the surface. The assumption is that nearby dwelling price points has similar values, therefore we proceeded with increment of the model radius until the surface had homogeneous levels of distribution of prices .

The output interpolated grid of 1999, shows that there are relatively expensive dwelling areas (colored orange/red) in the some of the marginal neighborhoods.

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.

The following map shows the interpolated grid of the difference in dwelling prices between 19992004, 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.

Even though the contours allow us to visualize flat and steep areas It might be a bit difficult to notice the prices ‘valley’ and perceive the difference in ‘height’ between the ‘valley’ floor and a ‘ridge’. Viewing data in three dimensions gives us new perspectives. 3 D viewing can provide insights that would not be readily apparent from a planimetric map of the same data.

The ‘height’ in the 3 D map is presented by z-values of the difference in dwelling prices between 1999 -2004. The high steep ‘mountains’ are the marginal neighborhoods. The next following maps shows this same map from different angles.

South West View Ramot the old city Talpiot Armon Ha’Natziv Gilo

Zonal Overlay Statistics Zonal functions take a value raster as input and calculate for each cell some function or statistic using the value for that cell and all cells belonging to the same zone. Zonal functions quantify the characteristics of the geometry of the input zones.

COUNT TERROR TYPE MEAN STD 45 shooting -25. 71% 7. 17 8 mortar bomb -25. 65% 2. 2 10 molotov cocktail -25. 45% 6. 39 69 stabbing -22. 06% 6. 87 19 grenade -21. 78% 5. 64 98 explosive device -21. 41% 7. 44 36 attack -21. 11% 8. 14 2 arson -15. 32% 8. 32

St. area_# Terror attacks casualties Average delta per st. area Neighborhood 721 3 0 -1219 Shuaafat 824 3 0 -1064 Armon Hanatziv 822 3 1 -1064 Armon Hanatziv 831 1 0 -926. 25 Armon Hanatziv 833 1 1 -854. 2 Armon Hanatziv 852 3 1 -808 Gilo 641 8 6 -796 Old City 642 5 13 -796 Old City 532 6 6 -794. 4932 Abu Tor 834 2 1 -788. 3455 Armon Hanatziv 544 2 4 -756. 1882 Arnona 313 1 52 -755. 4299 Qiryat Menachem 545 3 5 -716. 6129 Talpiot 543 2 0 -706. 5984 Talpiot 771 11 77 -641. 3303 French Hill 772 2 1 -614. 5333 French Hill 844 11 14 -493. 5 793 16 9 -304 843 12 13 -298. 5932 Gilo city center Gilo

St. area_# Terror attacks casualties Average delta per st. area Neighborhood 322 0 0 146. 5211 332 0 0 4. 4938 422 0 0 -27. 5366 Rasko 511 0 0 -70. 9561 Northern Qatamons 124 0 0 -112. 5625 Maalot Dafna 512 0 0 -162. 7882 Northern Qatamons 211 0 0 -204. 641 Qiriyat Tzanz 423 0 0 -222. 5075 Rasko 132 0 0 -243 176 0 0 -245. 7108 Bucharim 411 0 0 -252. 6939 Gonen 412 0 0 -255. 9195 San Martin 212 0 0 -274. 75 417 0 0 -280. 2515 Gonen 414 0 0 -296. 8101 Northern Qatamons 213 0 0 -302. 1 163 0 0 -307. 0426 Zichron Yosef 128 0 0 -311. 2642 Meah Shearim 123 0 0 -318. 1654 Bucharim Qiriyat Yovel Ir Ganim Nachalat Shivaa Romema

Regression Purchase Prices 1999 avg. price per m (log) 2004 avg. price per m (log) delta price % (constant) 1. 866* 3. 121** -4. 146** terror attacks (log) -0. 160 -0. 008 -0. 681 distance from the seam line (log) -0. 529 -1. 081 -0. 859 population density per built sq. km **4. 338 **4. 911 3. 001** housing conditions (sq. m of dwelling person (log)) 1. 656* 1. 712* 3. 705** housing quality (% ‘dilapidated’ buildings) -2. 460* -4. 548** -0. 247 0. 15 0. 25 0. 10 Variable (average per st. area) adjusted R sq

Regression Rental Prices 1999 avg. price per m (log) 2004 avg. price per m (log) delta price % 0. 398 1. 576 -0. 129 -4. 048** -3. 581** -0. 771 distance from the seam line (log) -0. 400 -0. 656 -0. 679 population density per built sq. km **4. 100 **3. 139 2. 559** housing conditions (sq. m of dwelling person (log)) 3. 266** 3. 767** 3. 260** 0. 19 0. 18 0. 07 Variable (average per st. area) (constant) terror attacks (log) adjusted R sq

Conclusions Terror intensity – increases over time n Terror patterns – increasingly random n House + rental prices – largest declines in peripheral neighborhoods and adjacent to seam line n

Differences between purchasing + rental behavior. n Terror is significant in rental decisions, less so in purchasing. n Explanation: if fear is main component of terror ( and not risk), more likely to be expressed in short term behavior (rental) than in long term (purchasing). n