Local housing price index analysis in wind-disaster-prone areas

Abstract

This study examines the effect of severe wind events on the mean and variance of housing price indices of six metropolitan statistical areas (MSA) that are vulnerable to hurricanes and/or tornadoes. The research focuses on three areas that experienced significant tornado activity (Fort Worth-Arlington, Nashville, and Oklahoma City) and three hurricane-prone areas (Corpus Christi, Miami, and Wilmington, NC). An econometric time series model that captures the housing market responses to severe windstorms is utilized. The model estimates changes in the local housing price index (HPI) as a function of several control variables as well as dichotomous variables that correspond to the tornadoes and hurricanes. As expected, the statistical findings indicate an immediate but short-lived decline in housing prices following a tornado or hurricane. Somewhat surprising is the result that the impact on the housing market is remarkably consistent whether the wind event was a hurricane or a tornado. Hurricanes and tornadoes are vastly different in terms of the point probabilities of a hit, the scope of the affected area and the lead time that supports last minute preparation to mitigate damage. It appears that the market response to destruction of real property does not distinguish between the types of wind event that produced the damage to the region. Results suggest that windstorms result in an immediate one-half to two percent reduction in total MSA housing value. This corresponds to a range of $34 million to $580 million in lost housing value. Estimates indicate some differences in how long market values continue to decline in the periods following the wind event; however, most of the decline occurs within four quarters after the windstorm. These differences can be attributed to the particular time series characteristics of the specific housing markets and their respective housing price indices. The market serves the purpose of integrating and normalizing the losses. In so doing the market provides a metric— a method for calibrating and comparing structural damage caused by different phenomenon.

Appendix

With the pulse dummy specification, the coefficient of the dummy variable in the nth period determines the effect of a disaster for that period. In the following derivation, this study uses an ARMA(1,1) model for illustration purposes and assume one disaster in period t and no other disasters during the n periods after t.¹

In time period t:

$$L_t=c+\psi L_{t-1}+\rho S_t+\varphi N_t+\delta D_t+\eta T_t +\varepsilon_t +\lambda \varepsilon_{t-1}$$

(4)

If :

$$A_t=c+\rho S_t +\varphi N_t +\eta T_t +\varepsilon _t +\lambda \varepsilon _{t-1}$$

then,

$$L_t =\psi L_{t-1} +\delta D_t +A_t$$

(5)

And in period t + 1:

$$L_{t+1} =\psi L_t +A_{t+1}$$

(6)

Substituting (2) into (3) gives:

$$L_{t+1}=\psi ^2L_{t-1}+\psi\delta D_t +(A_{t+1}+\psi A_t)$$

(7)

Then in period t + 2,

$$L_{t+2} =\psi L_{t+1} +A_{t+2}$$

(8)

Substituting gives:

$$\eqalign{ L_{t+2} =\psi ^3L_{t-1} +\psi ^2\delta D_t +(A_{t+2} +\psi A_{t+1} +\psi ^2A_t)\cr {\ldots}\cr }<!endaligned>$$

(9)

So that in period t + n,

$$L_{t+n}=\psi L_{t+n-1} +A_{t+n}$$

(10)

$$L_{t+n}=\psi^{n+1}L_{t-1}+\psi ^n\delta D_t +(A_{t+n} +\psi A_{t+n-1}+\cdots+\psi ^nA_t)$$

(11)

Therefore, the coefficient on the dummy variable in the nth period after the disaster is ψⁿδ and which measures the effect of the disaster on the local housing market in the nth period after the disaster.