An overview of our methodology is presented in Fig. 1. We begin with detailed real estate market data for the Seattle metro area (discussed in detail in the Appendix). Using GIS methods, we spatially code the data by zip code to correspond to the three concentric attack zones, consistent with the AAS. We conduct a panel regression analysis, using the GIS-coded data, to determine the additive effect of each variable of interest on median sales price throughout each zone of the city (Section 'Real Estate Regression Results'). We utilize our macromodel (Section 'REAL ESTATE PRICE IMPACT AND FORECLOSURE ANALYSIS') to estimate macroeconomic impacts (including both direct and indirect effects), consistent with the attack scenario, and shock the panel model with those macrooutputs. That shock, discussed in detail in Section 'REAL ESTATE PRICE IMPACT AND FORECLOSURE ANALYSIS', allows us to predict real estate price changes resulting from the attack (Section 'Property Value Changes'). We then utilize HUD's 2009 American Housing Survey (AHS) to construct a citywide distribution of homeowner equity. Finally, we utilize that distribution to predict foreclosures for each spatial zone resulting from the attack (Section 'Foreclosures').
4.1. Data Utilized for Panel Analysis
Terrorist attacks affect real estate values both through housing market and macroeconomic variables. We begin with a discussion of a unique data set consisting of spatially differentiated macroeconomic variables. The sales price of any single property can be driven, to a large extent, by exogenous factors specific to that property: use requirements of the buyer, preferences, tastes and characteristics inherent to the specific location (e.g., square footage) or the community in which the property is located. It is community and regional drivers of real estate prices that would be most altered by a region-specific terrorism event.
We utilize zip code as our unit of analysis because it enables us to make both community-specific and region-specific assessments of the macroeconomic drivers of real estate prices, while avoiding the large quantity of property-specific and exogenous information that may otherwise provide noise in our model. Although there are 61 zip codes within the City of Seattle, our analysis focuses on the main 32 of these.5 Our unit of analysis for time scale is months, from January 1994 through December of 2010. This historical time range is important because of the region-specific history of the Seattle metro area. In 2001, the Nisqually Earthquake hit the Puget Sound region, shocking the regional economy in the short run, having a minor downward influence on real estate prices. Our historical range of time, therefore, incorporates periods before, during, and after that disaster, and also encompasses events such as recessions.6
Zip code 98101 is the epicenter, which means that an anthrax attack on the Seattle CBD would contaminate an area of approximately 10 square miles. We use the U.S. Census's OnTheMap GIS tool to divide three zones: (1) Contamination Zone (includes the CBD), (2) Adjacent/Fringe Zone (a four-mile radius perimeter surrounding Zone 1), and (3) Outlying Zone (includes outlying zip codes of the City of Seattle).
4.2. Real Estate Regression Model Specification
The dependent variable of this analysis is median sales price of residential real estate. It is measured at the zip code level, and varies monthly across our 17-year panel. To adjust for constant dollars we used the Seattle-Tacoma housing Consumer Price Index as the deflator. Our method for analyzing these data is a time series cross-sectional (TSCS), or longitudinal model. This class of model provides unbiased estimates of multiple units across time that do not fall prey to the same criticisms as pooled models applied to panel data. Our analysis uses a fixed effects cross-sectional time series (FE) model. The model includes a time-stationary error term that corrects for time dependency, and enables the same style of estimates that are sought through pooled models or OLS cross-sectional, nonpanel models. This model is given by:
where uz are unit (zip code) specific fixed effects/unobservables, and εzt is the identically and independently distributed (i.i.d.) error term.
The regressors included in this equation are both the fundamentals of the local and regional macroeconomy, as well as the fundamentals of the local and regional housing market. Because the analysis is driven by the aim of estimating spatially disaggregated effects, this estimating equation is broken down into five separate regression models. First, the regression is applied to the entire City of Seattle. Second, the model is applied to each of the three zones of spatial analysis, which include the Contaminated Zone (CBD), the Fringe Zone, and the Outlying Zone. Finally, a separate regression is estimated only for those zip codes that lie outside of the Contaminated Zone, for purposes of comparison to the Contaminated Zone.
4.3. Real Estate Regression Results
The panel analysis results are provided in Table II. The regressions are robust overall and explain anywhere between 35% and 75% of the variance in median residential sales prices between 1994 and 2010 for the City of Seattle. The models have strong fitness measures, indicating that a proper set of both macro and real estate variables has been included.7 The regression model labeled “Citywide” indicates a regression that includes each of the 32 zip codes for the entire Seattle area. Spatial disaggregation at the zip-code level is important to the results because it is the ideal geographic unit for this analysis. A zip code captures significant demographic, economic, and housing data and is a good indicator of income and socioeconomic levels. Furthermore, zip codes distinguish areas of a city such as financial centers, public places, and neighborhoods mainly by these variables. The models labeled “Zone #1, Zone #2, and Zone #3” indicate individual zone-specific regression models. And, the model labeled “Zone #2 and 3” indicates a pooled model that includes the Fringe Zone and the Outlying Area, excluding the Contaminated Zone.
Table II. Real Estate Regression Analysis Results
| || ||Zone #1||Zone #2||Zone #3||Zone #2|
| ||Citywide Model||Model||Model||Model||& 3 Model|
|Median Household Income||2.58***||3.71***||3.91***||1.65***||2.14***|
|Personal Income [$B US]||2,982.83***||3,108.16***||3,119.14***||2,755.75***||2,860.35***|
The citywide regression explains the average tendency of real estate prices throughout the City of Seattle, irrespective of geospatial location. Throughout the panel period, Median Household Income provides one of the strongest macro drivers of real estate prices, as it does throughout individual zones. Holding all other variables constant at their mean, a $1,000 increase in median household income would increase the median sales price of residential property in the mean zip code by more than $2,500. That effect is more than 40% higher in the area with the CBD (Zone #1) and the fringe area (Zone #2). In those districts, the median sales price of real estate is much more responsive to changes in household income. Household income is also a highly statistically significant driver of real estate prices, safely rejecting the null hypothesis that the asymptotic relationship is zero at even the most stringent levels of significance (p-value) across all geospatial parameters.
Employment (full-time equivalent persons by place of work), however, is a strong but less statistically significant driver of real estate prices throughout this panel. In those regression models in which the coefficient is statistically significant, the employment rate is positively related to real estate prices. In the average Seattle zip code, for every 1,000 job increase in its total employment, the median sales price of residential real estate in that zip code increases by over $680.
Personal Income, which is an aggregate measure of the strength of the regional macroeconomy, is shown to be a highly robust and significant determinant of real estate prices. For every $1 billion increase in regional personal income, the average Seattle zip code's median sales price increases by nearly $3,000. That relationship is slightly stronger in the Contaminated Zone (Zone #1), and fringe (Zone #2), as the same effect leads to a more than $3,100 increase in median sales price. Put another way, across our 17-year panel, the mean regional personal income is just below $150 billion. A single standard deviation increase in the personal income ($18 billion) would lead to the average zip code's median sales price increasing by more than $52,000, holding all other variables constant at their mean.
The real estate measures also provide for robust drivers of changes in real estate prices. The variable for Home Sales, which is a measure of the strength of the real estate market, particularly the demand side, is robust, significant, and positively related to sales price in all of our regression models except Zone 1. The CBD constitutes the downtown area, and housing prices in that area are not affected by normal changes in demand patterns for housing. Demand for housing in that community is fairly inelastic. Depending upon the geospatial location, for every additional property sale in a zip code, the median sales price can increase between $50 and $500.
Foreclosures are also a robust indicator of the health of local and regional real estate markets. Foreclosures signal to other homeowners that the demand for housing may be in decline, and are often both a cause and a consequence of a depressed market. Homes that are foreclosed may also be sold for significant decreases in sales price if a “short sale” is made or if a repossession or bank ownership occurs. As with the housing market regressors, the median sales price in the CBD is less responsive to foreclosures. Similarly, the mean monthly foreclosure rate of an average Zone 1 zip code is just above one foreclosure per month, whereas that number outside of that zone is just over two foreclosures per month.
The supply side, however, provides a different effect altogether. Housing Inventory is our proxy for the supply of available residential real estate. Ceteris paribus, an increase in the supply of available real estate should lead to a decrease in the price of real estate, relative to a fixed demand for housing. We find this effect to be accurate and in the appropriate direction in each of the regressions for which the coefficient is highly statistically significant. For every additional residential property added to the housing inventory in the average Zone 1 zip code, the median sales price of housing decreases by approximately $15. That is, a single standard deviation increase (approximately 2,800) in the number of residences would decrease the average Zone 1 zip code mean sales price by over $44,000.
The supply of housing provides an interesting departure from expected disaster outcomes. That is, under a nondisaster scenario, an increase in the housing stock relative to a fixed housing demand would lead to a decline in the price of housing. However, under a disaster scenario, a decline in the housing stock stemming from property damage or condemned properties would constitute a decline in both the supply and demand for housing. There would be fewer livable homes and fewer people would prefer to reside in them. Under an anthrax attack disaster scenario, we would expect a likely sign reversal of this coefficient for housing stock, as the demand for housing in both the attack zone and neighboring communities would decline both in the short run and long run, and the supply of housing would be suppressed in the short run, with likely sticky rebound in redevelopment. It should be noted, however, for all coefficients in each of the regression-estimating equations, that the assessments were made based on a panel of data from periods that did not include terrorism events.8