Tornado shelters and the housing market

Abstract

Mitigation against natural hazards often involves long-lived, immobile investments. Home owners must be able to capture the present value of future benefits to equate the private and societal return on mitigation. The capitalization of mitigation into home prices thus is crucial for home owners to have a proper incentive for mitigation. We investigate the existence of a premium for tornado shelters using home sales in Oklahoma City, where the deadly tornado outbreak of 3 May 1999 and the Oklahoma Saferoom Initiative increased public awareness of tornado shelters. We find that a shelter increases the sale price of a home by 3.5% to 4% or approximately $4200 given the mean price of homes sold in 2005. The magnitude of the premium is plausible given that shelters retail for $2500-$3000 installed.

Full text

This article was downloaded by:[Sutter, Daniel]
On: 21 November 2007
Access Details: [subscription number 7864457 19]
Publisher: Routledge
lnforma Ltd Registered in England and Wales Registered Number: 1072954
Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
Construction Management and
Economics
Publication details, including instructions for authors and subscription information:
http://www. informaworld. com/smpp/title-content=t7 1 3664979
Tornado shelters and the housing market
Kevin M. Simmons a; Daniel Sutter D
a Austin College, Economics and Business Administration, Sherman, USA
D University of Texas-Pan American, Economics and Finance, Edinburg
First Published on: 17 October 2007
To cite this Article: Simmons, Kevin M. and Sutter, Daniel (2007) 'Tornado shelters
and the housing market', Construction Management and Economics, 25:11,1117 -
1124
To link to this article: DOI: 10.1080101446190701618299
URL: http://dX.doi.oro/1 0. 1 080/0 1 4461 9070 1 6 1 8299
PLEASE SCROLL DOWN FOR ARTICLE
Full terms and conditions of use: http://wwr,,v.informaworld.com/terms-and-conditions-of-access.pdf
This article maybe used for research, teaching and private study purposes. Any substantial or systematic reproduction,
re-distribution, re-selling, loan or sub-licensing, systematic supply or distribution in any form to anyone is expressly
forbidden.
The publisher does not give any warranty express or implied or make any representation that the contents will be
complete or accurate or up to date. The accuracy of any instructions, formulae and drug doses should be
independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings,
demand or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or
arising out of the use of this material.

f-
o
o
N
a)
-o
E
q)
o
z
N
N
lt*
r
.c)
c
(u
o
f
0)
:f
9.
;
m
E
o)
E
G
o
c
=
o
o
Mitigation against natural hazards often involves long-lived, immobile investments. Home owners must be able
to capture the present value of future benefits to equate the private and societal return on mitigation. The
capitalization of mitigation into home prices thus is crucial for home owners to have a proper incentive for
mitigation. \We investigate the existence of a premium for tornado shelters using home sales in Oklahoma City,
where the deadly tornado outbreak of 3 May 1999 and the Oklahoma Saferoom Initiative increased public
awareness of tornado shelters. We find that a shelter increases the sale price of a home by 3.5% to 4oh or
approximately $4200 given the mean price of homes sold in 2005. The magnitude of the premium is plausible
given that shelters retail for $2500-$3000 installed.
Keywords: Natural disasters, prices, tornado shelters, mitigation, multiple regression
Construction Management and Economics (November 2007) 25, LLIT-I124
Tornado shelters and the housing market
fi{, R*rrtf*ciq*
USA
West Uniaersity Dr., Edinburg, 78541-2999,
most powerful tornadoes. In the 1990s FEMA included
tornado shelters in its National Mitigation Strategy and
issued design standards for shelters and safe rooms
(FEMA, 1999).
Tornado shelters, like other types of mitigation-
strengthened construction or elevating structures out of
a flood plain-require an upfront investment. Builders
and home owners will be reluctant to invest in
mitigation if unable to recoup some of the cost in the
form of a higher price for a home or building. But will
home buyers care enough about natural hazards to pay
extra for mitigation? $7e test for a house price premium
for tornado shelters in single family homes in
Oklahoma County. Oklahoma City affords an excellent
opportunity to test for such a premium, owing to high
tornado risk, high public awareness of the risk, and an
inventory of homes with shelters. The inventory of
shelters is due in part to the Oklahoma Saferoom
Initiative, sponsored by FEMA and the State of
Oklahoma after the 3 May 1999 tornadoes. The
programme offered $2000 rebates for shelter or safe
room installation and received more than l4 000
applications and paid out rebates for 6400 installed
shelters. If a price premium exists for tornado shelters
anywhere, it should be in Oklahoma.
lVe analyse county tax assessor records of over
13 000 home sales in the Oklahoma Countv area in
KEVIN M. SIMMONS' and DANIEL SUTTER2*
I Austin College, Economics and Business Administration, Sherman,
2 (Jniaersity of Texas-Pan Ameican, Economics and Finance, 1201
USA
Received 20 October 2006; accepted 7 August 2007
Introduction
Natural hazards threaten people and property. tMhile
commonly described as 'Acts of God', the choices
people make about where and how to live-the built
environment-significantly affect natural hazards'
impact on society (Mileti, 1999). Numerous measures
can reduce the impact of hazards; from the standpoint
of economic efficiency, society should invest in the level
of mitigation which minimizes the sum of hazard costs
plus mitigation costs. Many scholars suggest that
society undertakes too little (or an inefficiently low
level) of mitigation. A recent analysis found that the
benefit-cost ratio of Federal Emergency Management
Agency (FEMA) mitigation exceeded four to one
(Multihazard Mitigation Council, 2005), suggesting
(but not proving) that additional investment in mitiga-
tion would have yielded net benefits.
\We examine one specific mitigation measure in this
paper, tornado shelters. Tornadoes are nature's most
violent storm, and particularly prevalent in the Great
Plains' 'Tornado Alley'. Storm cellars have a long
history in the area) but engineers over the past 30 years
have developed new underground tornado shelters and
above ground safe rooms capable of withstanding the
*Author for correspondence. E-mail: dssutter@utpa. edu
Construction Management qnd Economics
ISSN 0144-6193 print/ISSN 1466-433X online @ 2007 Taylor & Francis
http ://www.tandf. co. uk/j ournals
DOI: I 0. I 080/0 I 446 1907 0I 618299

f-.
C
C-
N
q)
-o
E
o
o
z
N
N
N
+J
.q
c
(g
o
f,
c)
J
g.
;
co
-o
q)
E
(U
o
c
3
o
o
1118
2005. Over 400 homes, almost 2.5%o of the sample,
have tornado shelters. Homes with shelters are
available in all size and price categories and not only
for the rich buying luxury homes. Home buyers should
be able to find a home with a shelter and other desired
features as well. A hedonic regression analysis reveals
that homes with shelters sell for 3.5% to 4o/o extra than
a comparable home without a shelter, a premium of
about $4200 for the average priced home.
The remainder of this paper is organized as follows.
Section 2 lays out the theoretical issues regarding the
market value of mitigation and reviews relevant
previous research. Section 3 provides variable defini-
tions and summary statistics. Section 4 examines the
characteristics and location of homes with shelters in
greater detail. Section 5 presents our regression
analysis. Section 6 offers a brief conclusion and
directions for future research.
Mitigation and housing rnarket: relevant
literature
Like many types of natural hazards mitigation, tomado
shelters are a long-lived, immobile form of self-protec-
tion.l A safe room is built into a home and cannot be
moved, while the cost of digging up and moving an
underground shelter is comparable to installing a new
shelter. The resident making the installation is unlikely
to personally consume all of the protection of the shelter
or saf,e room, which could have a useful life of over 30
years. Residents may consider only the safety benefits
during the time they plan to live in the home and ignore
the remaining benefits from the shelter. Owing to these
spillover benefits, society may experience an ineffrciently
low level of investment in tornado protection
(I(unreuther and I(leffner, 1992).
This time horizon problem can be overcome if the
shelter purchaser can sell the home at a premium
reflecting the value of remaining safety benefits. The
home owner can then capture the full benefit to society
of the shelter and can compare the benefits during the
period she expects to live in the home with the net cost
of the shelter. Premiums for shelters (and other
amenities or a well-maintained home) are crucial to
an efficient housing market.
Several problems can complicate the sale of homes
with tornado shelters at a premium. A first problem is
transactions cost. If only a small percentage of buyers
value shelters, matching buyers interested in shelters
with homes with shelters may be difficult. Ultimately
if almost nobody valued shelters, they would represent
vanity items with no market value. Also buyers
interested in a shelter may face a limited range
Simmons and Sutter
of choices and have to decide between a house
with all their desired features except a shelter and a
much poorer matched house with a shelter. A poor
match with other features reduces the equilibrium
premium.
A second problem is buyers ignoring tornado risk
altogether in purchasing a home. Many studies have
demonstrated the existence of a low probability, high
consequence event bias in decision making; people
treat the small probability of a tornado or other hazard
as zero (Camerer and I(unretuther, 1989). \ffhen
people ignore tornado risk, the perceived value of
protection (or mitigation) is zero, even though the same
persons might choose to buy a shelter if they perceived
the true probability of a tornado.2 In addition, some
home buyers will be new to tornado-prone areas and
consequently unaware of tornado risk in their home
purchase decision. Demand for homes with shelters
and the market premium falls as the percentage of
buyers who ignore tornado risk increases.
Third, potential buyers might exhibit myopia
(I(unreuther and I(leffner, 1992). Mitigation typically
involves upfront expenditure and produces benefits
over the useful life of the measure. Myopic individuals
focus on the initial cost (or apply too high a discount
rate) and choose not to invest in mitigation measures
which seem to have positive expected value. Thus
home buyers might see the higher price of a home with
a shelter and excessively discount the benefits.
I(unreuther (1998) argues that financing mitigation
through a mortgage or home improvement loan should
offset myopia since people can compare the annualized
cost with the annual benefit. Since a premium for an
existing shelter is included in the sale price of the home
and can be financed in the mortgage, myopia may not
be a severe problem for tornado shelters.
Economists have estimated hedonic models of home
prices for many decades. The technique has been
employed, for example, to confirm the impact of
environmental factors like air and noise pollution on
home prices. Most applications with respect to natural
hazards have investigated the impact of risk or risk
perception on home prices as a test for low probability
event bias. For example, Brookshire et al. (1985) and
Beron er al. (1997) found that homes in seismic zones
designated by the state of Califomia sell at a discount,
and that the size of the discount declined after the
1989 Loma Prieta earthquake. Shilling et al. (1985),
MacDonald et al. (1987) and Spreyer and Ragas
(1991) each found that homes located in flood plains
in three different Louisiana cities sold at approximately
a 60/o discount relative to homes outside the flood
plain. Hallstrom and Smith (2005) and Carb one et aI.
(2006) examined the impact of Hurricane Andrew on
home prices in two Florida counries, Miami-Dade and

t\
O
O
N
L
0)
-o
E
c)
o
z
r
N
N
lt-
:
.o)
c
(E
o
t'
o
f,
a
;
m
ro
C)
E
o
o
c
3
o
o
Tornado shehers and the housing market
Lee. The growth in house prices declined after Andrew
in areas of each county most susceptible to storm surge
flooding, consistent with a greater perceived risk of
hurricanes throughout the state. In all cases these
studies found that participants in the housing market
were aware of and responded to the natural hazard
risk. Relatedly, Ewing et al. (2007) found that powerful
tornadoes and hurricanes lowered a housing price
index in the affected metropolitan area by 0.5% to 2o/o
in the quarter of the event, but have no permanent
effect on house prices. Thus both natural hazards risk
and hazard events can affect home prices, although
none of these studies examined mitigation of homes
facing a hazatd risk directly.3
Two studies have applied hedonic price models to
examine whether housing markets value natural
hazards mitigation. Simmons et al. (2002) found a
statistically significafi- 5o/o price premium for houses
with hurricane blinds in a Texas Gulf coast city. The
price premium approximately covered the full cost of
hurricane blinds for the average home in their sample.
Simmons and Sutter (2007) found a 5%o premium in
lot rents for tornado shelters in manufactured
home parks in Oklahoma. For the average park in their
sample, the rent premium would approximately cover
the cost of an underground shelter. Their resuk,
however, just failed to attain statistical significance.
Several studies have examined the value of shelters in
tornado-prone areas, which affects the likelihood of
detecting a house price premium. Simmons and Sutter
(2006) find that the cost per life saved for tornado
shelters in Oklahoma exceeds $50 million,a while
market revealed values of a statistical life rarely exceed
$10 million ffiscusi et al., 2000). This suggests rhat
with full information about tornado risks relatively few
households will be willing to pay very much for shelters,
leading to the transactions costs problems discussed
above in matching the small proportion of home buyers
interested in shelters with the small proportion of
homes with shelters. Two studies have directly esti-
mated the value of shelters using the contingenr
valuation method. Ozdemir (2005) in a survey of
residents of Lubbock, Texas found that the mean rhat
residents were willing to pay for shelters was 92449,
while Ewing and I(ruse (2006) found a willingness to
pay mean of $2500 in a survey of Parade of Homes
visitors in Tulsa. $Tillingness to pay reflects residents'
subjective perception of tornado risk and preferences
toward risk, while Simmons and Sutrer's cost per life
saved uses an objective measure of risk-actual tornado
fatalities. Buyers' actual willingness to pay will drive the
market premium for shelters, so we may find a market
premium even though objective risk data suggest that
the premium is likely to be small.
1119
Variable definitions
Our dataset consists of sales of single family homes in
2005 in Oklahoma County in the Oklahoma City metro
area. $7e omitted properties with areas of less than 700
square feet or a sale price of less than $ l0 per square
foot, and one home which sold for over $ I I million, or
more than three times the price of the next most
expensive home. We have a dataset of 13461 home
sales. The sales data for each county are from publicly
available Tax Assessor records. Our dataset is cross-
sectional and thus unfortunately does not allow analysis
of repeat sales of the same property which would
control for unmeasurable characteristics of the homes.
Our variable of interest is Shelter, a dummy variable
which equals one if a home has a tornado shelter or safe
room and zero otherwise. Local officials and emergency
managers in each county have over the past several years
created an inventory of tornado shelters. The inventory
combines building permits, rebates awarded as part of
the Oklahoma Saferoom Initiative, a property tax
exemption for shelters passed by voters in a referendum
in November 2002, and self-registration by home
owners with emergency managers. The inventories are
considered reasonably complete and include the date of
installation if a building permit was issued.5
The dependent variable is the natural logarithm of
Sales Price, the price in dollars. \We employ White's
robust standard errors in our regression. Table I reports
summary statistics for our dataset. The average sales
price was $119400, with a range of $8000 to $3.t
million. Just over of half the homes in the dataset sold for
less than $100000, while llo/o sold for over 9200000.
Figures I and 2 display the distribution of home sales
prices for all the homes and for homes with shelters.
We employ numerous control variables constructed
from the Tax Assessor's records. Our basic model for
the control variables follows Simmons et al. (2002),
limited by the exact home characteristics tracked by the
Oklahoma County Tax Assessor. Age is the age of the
home in years in 2005, or 2005 minus the year the
home was built. The oldest home in our sample was
built in 1860 while 548 homes were built in 2005. The
mean age of homes is 38.8, or the average home was
built in 1967. About 20o/" of homes are l0 years old or
less, while 28o/o of homes are over 50 years old. \7e
expect older homes to sell for less, everything else
equal.6 Square Feet is the size of the home. Our mean
home size is 1620, with a range from 700 to l3 395. \7e
expect larger homes to sell for a higher price. In
addition to Square Feet, t}:,e Tax Assessor records report
three attributes of home size, Rooms, Bedrooms and
Baths. Given the correlation between the variables, we
include only Bath.s in our regression.

tt20
Table 1 Summary statistics
Simmons and Sutter
Full sample Flomes with shelters
Mean Standard deviation Mean Standard deviation
f-
(
c
N
0)
-o
E
q)
o
z
N
N
It-
:
.q)
$
o
t-
o
f
@
;
co
E
C)
!o
Sale Price
Shelter
Square Feet
Age
Masonry
Asbestos
Brick Veneer
Hardboard
Siding
Vinyl
Other
Hip
Gable
Hip/Gable
Other Roof
Rooms
Bedrooms
Baths
rt9362
0.0297
1623
38.8
0.7535
0.0298
0.0292
0.0147
0.0705
0.0833
0.0191
0.24t0
0.4824
0.2696
0.0070
5.92
2.96
1.74
109780
0.r697
759
23.0
0.4310
0.1 699
0.1 683
0.1205
0.2560
0.2763
0.r367
0.4277
0.4997
0.4437
0.0832
r.3950
0.6670
0.6857
108791*
I
t484
4l.9{.*'1.
0.6L23+**
0.0617+*+
0.0272
0.0247*
0.0914
0. l679*{.{.
0.0148
0.2494
0.5012
0.2444
0.0049
5.72***
2.94***
l.5g+*+
96326
0
753
22.5
0.4878
0.2410
0.r628
0.r554
0.2885
0.3742
O,T2LO
0.4332
0.5006
0.4303
0.0702
1.36t6
0.6759
0.717 4
3
o
o.Atrotes: *, ** and *** indicate that the mean for homes with shelters differs from the overall sample mean at the l0%, 5Yo and l% levels
respectively in a two-tailed test.
Our dataset contains seven dummy variables describ-
ing a home's exterior and four dummy variables
describing the roof. The exterior variables are
Asbestos, Brick-Veneer, Hardboard, Masonry, Siding,
Vinyl and Other Exterior, while the roof variables are
Hip, Gable, HiplGable and Other Roof. All of these
variables were constructed by the authors from the Tax
Assessor records. \il7e omit Masonry and Hip in our
regressions, so the coefficients on the included variables
show the effect of each exterior or roof type relative to
these categories. \We have no expectations about the
signs of these variables, which we include as controls.
\We include a number of dummy variables to control
for location. About 65%o of homes sold in 2005 were in
Oklahoma City, so we include a dummy variable OKC
equal to one for homes in Oklahoma City and zero
otherwise. $7e also define a set of zip code dummy
variables for each of the 5l zip codes included in our
dataset. These variables should control for differences
in local neighbourhood housing markets. In addition,
we include a set of dummy variables for the month of
sale of the home. Both the zip code and month dummy
variables were jointly significant, but to conserve space
we do not report them in our regression tables.
2(]
rt]
l6
i4
rl}
ra
Fsrcsnt lB
g
(l
4
g
0
$*les Frice Oistribution . All Hnmes
*s0- $T's-
t?5 38U
Ss*ss Frice DiEtvibution *Homes w:ith $heltens
LT Sfi 5&75 10u- 1is- 1ss-
12,5 15$ 1T5 1.?$- rs,s- t?$- BT
?0* t25 ?sCI ?sfl
f+-
lgs
?0& 2?5-
?39 ?5ff
l$. roCI- r3$-
lm 1:5 r50 'JI
!5CI Figure 2 Sales price
shelters
LT S* 5&?$
Figure 1 Sales price distributions of homes distribution of homes with tornado

N
o
o
N
0)
-o
E
q)
o
z
N
N
r
l-
u
:
.o
c
(0
o
t-
a)
#
J
9.
;
m
-o
o
E
G]
o
c
=
o
o
Tornado shehers and the housing market
A closer look at hornes with tornado shelters
Before turning to our econometric analysis, we first
examine the characteristics of homes with tornado
shelters in some detail. A problem for our empirical
analysis would arise if only large, expensive, new homes
had shelters. In this case a shelter premium could
actually be a premium for luxury homes, particularly if
our control variables fail to adequately capture the
features of upmarket homes. The distribution of
shelters across types of homes also affects societal
vulnerability (Peacock, 2003). If tornado shelters were
only available in half million dollar, 4000 square foot
homes, few people could afford a house with a shelter,
and policy makers might consider taking steps to make
shelters available for working class families.
Table 1 also presents the mean and standard devia-
tion of the variables for homes with shelters. The table
also indicates whether the mean significantly differs
from the population mean. A simple mean comparison
shows no evidence of a price premium for shelters, as
the mean sales price for homes with shelters is
$ 109 000, compared with $ I l9 000 for all 2005 home
sales, a difference which is statistically significant at rhe
l0% level. The comparison certainly indicates that
some average and below average priced homes have
shelters. Closer analysis shows that 2|o/o and 47% of
homes with shelters sold for under $50 000 and
between $50000 and $100000 respectively compared
with only l5oh and 37% of all home sales in these price
ranges. Clearly buyers in all price ranges can find
homes with tornado shelters.
\(/e might expect that only large, new homes would
have tornado shelters, but Table 2 shows this is not true.
Homes with shelters were on average three years older
and 1 40 f( smaller than all homes sold in 2005. Again
analysis of the distribution of homes by size and age
confirms that a wide variety of homes feature shelters.
Thirty per cent of homes with shelters were less than
1000 ft2 and only 17o/o over 2000 ft2, while 19o/o and,
23% of all homes sold fell into these categories. Forty-
three per cent of homes with shelters were over 50 years
old compared with only 32o/o of all homes sold.7
Table 1 also reports the mean values of the room,
exterior and roof variables for homes with shelters.
Comparison reveals that homes with shelters have
fewer Rooms, Bedrooms and Bath.s than the sample as a
whole, with the differences being statistically significant
at the 1% level. Homes with each category of interior
and roof have shelters, so buyers should not have to
sacrifice too many desired characteristics to buy a home
with a shelter.
As mentioned above, almost all the homes with
shelters in our sample were located in Oklahoma
ltzl
County. Overall about 3% of homes sold in
Oklahoma County had shelters. Our dataset includes
homes in 1l different cities within the county. The
percentage of homes sold with shelters ranged across
communities from 1.8% in Edmond to 6.9Yo in Del
City, which was affected by the 3 May 1999 tornado
and where many residents took advantage of the
Oklahoma Saferoom Initiative. Homes with shelters
are available throughout the county, so a buyer should
be able to find a home with a shelter near their desired
placed to live.
Results
Table 2 reports our regression results, The first column
contains the full model estimated on our entire dataset.
Our variable of interest, Shelter, has a positive point
estimate, indicating a 3.5Yo price premium, and is
statistically significant at the 10% ievel in a two-tailed
test. The 3.5% price premium translates into a $4200
price differential for our average priced home. The
magnitude is very reasonable given that underground
shelters currently retail for $2500-$3000 installed,
while above ground safe rooms cost $5000 or more.
Note that this amount exceeds by about 67%o the mean
willingness to pay for a shelter or safe room of 92500
found by Ozdemir (2005) and Ewing and I(ruse
(2006).
Square Feet and Age perform as expected, with larger
and newer homes selling for a higher price and both
variables significant at better than the 1% level. Both of
these variables enter in logs in our specification, and
thus the coefficients are elasticities. The first column
shows that a 10% increase in area increases the sales
price by about 9%, while a l00o/o increase in Age, say
from one to two years, reduces price by about 2o/o. An
extra Bath increases the price by about 16%. Homes in
Oklahoma City sell at a statistically significant 13o/o
discount relative to the rest of Oklahoma County. Four
of the exterior variables are significant at the 10% level
or better, while none of the roof variables are significant
at the 10% level, so buyers seem to place greater weight
on the exterior than the roof.
The majority of homes in our sample are in
Oklahoma City, and as a robustness check on our
results, we report in column two a specification of the
model using only homes in Oklahoma Ciry. This
specification eliminates differences in local government
services across jurisdictions, which may be imperfecdy
controlled for with the zip code dummy variables, as a
possible explanation of our results. Note that the model
also includes zip code dummy variables to control for

TT22
Table 2 Regression analysis of home sales price
Simmons and Sutter
Full sample OI(C only Old houses excluded
f"
(
(
N
t-
c)
-o
E
C)
o
z
N
N
lt*
F
:
.c)
c
N
o
t-
0)
=
Q
;
co
io
c)
E
(U
o
=
o
o
Constant
Shelter
LN(Square Feet)
LN(Aee)
OKC
Asbestos
Brickveneer
Hardboard
Masonry
Siding
Vinyl
Hip
Gable
HipGable
Baths
Adjusted R2
No. of observations
4.64***
(34.3)
0.0342*
(1.80)
0.887l"t'.l'<
(4e.e)
_ 0.0179*+*
(8.14)
- 0. I 3g**
(11.6)
- 0. l7g*:e*
(3.e8)
0.304+{<1,
(7 .7 r)
-0.0529
(0.ee)
0.L42+**
(3.er)
- 0.0705*
(r.77)
0.0133
(0.34)
0.0384
(0.73)
-0.0t26
(0.24)
0.0486
(0.e2)
0.159:r'+*
(14.6)
0.732
13641
4.49*:r':r'
(27.6)
0.0409
(1.52)
0.9911.++
(40.2)
- 0.0205*.*'r.
(6.15)
_0.204***
(3.4e)
0.402*+*
(6.e4)
-0.0760
(1.08)
0.2L3***
(4.3r)
-0.0729
(r.37)
0.0163
(0.31)
0.02tr
(0.37)
-0.0391
(0.68)
0.0t32
(0.023)
0.L7 4**+
(r2.2)
0.684
8679
4.37+++
(2e.r)
0.0393*
(1.6e)
0.950+++
(4e.7)
-0.0121*:|<*
(5.52)
_0.0434*tF+
(3.44)
_0.426+*+
(4.48)
0.0821*
(1.86)
- 0. 179{'.**
(3.04)
-0.0650
(r.54)
- 0.0957**
(2.03)
-0.126**
(2.44)
0.0817
(r.44)
0.0453
(0.80)
0.llg**
(2.0e)
0.126***
(1 1.0)
0.769
9219
Nores: Absolute t-statistics based on robust standard errors in parentheses. *, ** and :r'*'r' indicate significance at the 0.10,0.05 and 0.01 levels
respectively in a two-tailed test. The OKC model includes only homes within Oklahoma City, while the 'old houses excluded' model omits
homes that are more than 50 years old. Each model also includes zip code and sales month dummy variables.
neighbourhood effects within Oklahoma City. The
point estimate of Sheher is a slightly larger 4.2% price
premium than in column one. The point estimate fails
to attain significance at the 10% level in a two-tailed
test, although the estimate is significant at the l0%
level in a one-tailed test. Overall the estimated
coefficients of the other variables are virtually
unchanged, with the only difference in significance
rhat Siding is no longer significant, although its point
estimate is slightly larger.
We noted earlier that 43%o of homes with shelters
were over 50 years old, and that some of these older
homes might have storm cellars instead of modern
shelters. Home buyers might not consider these
equivalent to new shelters and safe rooms) and this
might affect the estimated shelter premium. \(/e do not
know the age of the shelter, and so cannot control for
old shelters directly. Consequently column three
presents a specification omitting homes over 50 years
of age, which should exclude all vintage storm cellars.
Omission of old homes has a modest effect on Shelter,
which is now statistically significant at the 10% level in
a two-tailed test 4.0% price premium. The price
premium for the average priced home is now
approximately equal to the cost of an above ground
safe room. There are a few modest differences between
the specifications with the full dataset and with old
homes excluded. The coefficient on Square Feet is
slightly larger with an elasticity now just less than one,
while the coefficient on Age is reduced, and newer
homes in Oklahoma Ciry sell at a discount of just over
4%. Vinyl and HipGable now attain significance at the
5%o level, while the overall fit of the model is slightly
reduced.8

a
f-
o
o
N
0)
-o
E
q)
o
z
r
N
N
r
l'-
r
-
.c)
c
(U
o
t'
a)
f
9.
;
co
E
o)
E
(o
o
C
=
o
o
Tornado shehers and the housing market
Conclusion
The 2004 and 2005 hurricane seasons increased public
interest in mitigation measures capable of reducing
losses from natural hazards. Mitigation often involves a
long-lived, immobile investment. Home owners who
invest in mitigation may not capture all of the benefits
of mitigation if they do not live in the home throughout
the measure's useful life. House price premiums for
mitigation must exist for the market to provide proper
incentives for investment in mitigation.
\We have investigated the existence of a premium for
tornado shelters in the prices of homes in the Oklahoma
City metro area in 2005. Consistent with Simmons er
al. (2002) for hurricane shelters, we found a premium
for tornado shelters. The Sheher variable is statistically
significant at only the 10olo level, although this would be
at the 5% level in a one-tailed test, which would be
appropriate if we did not expect that Sheher could
reduce the sale price. The point estimate indicates a
fi420} shelter premium for the average priced home in
our sample, which exceeds the cost of underground
shelters but is somewhat less than the cost of above
ground safe rooms.
One direction for future research would be a long-
itudinal study involving repeat sales of homes. Repeat
sales analysis focuses on the change in price, and this
avoids problems with unchanging but hard to measure
factors affecting home prices (Hallstrom and Smith,
2005; Carbone et al., 2006). If repeat sales of homes
before and after shelter installation can be identified,
this would allow determination of whether a price
increase can be attributed to the shelter.
Future research could also search for shelter premia
in other real estate markets. Several factors made
Oklahoma City particularly likely to exhibit a premium
for tornado shelters: a high risk of damaging tornadoes,
high public awareness of tornado risk, and a sufficient
inventory of homes with shelters. It may be unlikely to
find a premium for shelters in other cities because the
cost per iife saved for tornado shelters in single family
homes increases quickly even when comparing tor-
nado-prone states like Oklahoma and Texas (Simmons
and Sutter, 2006). But along the Gulf coast or in
Florida, shelters and safe rooms also provide protection
against hurricane winds as well, so safe rooms might
carry a premium due to dual use in these areas.
Notes
1. Shelters reduce the probability residents are killed or
injured in a tornado, but do little to reduce damage and
thus are self-protection instead of self-insurance (Ehrlich
and Becker, 1972).
tt23
2. Note though that Peacock (2003) points out that home
buyers are often quite concerned with hazard risk when
looking to buy a home but often lack extra cash after
purchase to invest in mitigation.
3. Restricting building in or relocating homes out of a flood
plain is a type of mitigation.
4. The calculation assumes a 3o/o discount rate, $2000 cost
per shelter, and a 5O-year useful life of a shelter.
5. For shelters installed during 2005 the date of installation
was compared with the sales date to determine if shelter
installation preceded the sale. The invenrories of shelters
distinguish between shelters and safe rooms, but there
were too few safe rooms to investigate a separate effect of
shelters and safe rooms. Data were also obtained for
Cleveland County in the Oklahoma City metro area, but
the inventory of tornado shelters appeared quite incom-
plete (only six out of over 5000 homes sold in 2005 were
reported to have shelters), and thus we estimated the
models using only the Oklahoma County data.
6. We replace zero with 0.001 when taking the natural log of
Age.
7. The age of many of the homes with shelters raises the
possibility that these older homes have unfinished base-
ments or storm cellars and not modern shelters or safe
rooms. The inventory of shelters does not control for
quality or include only shelters built to FEMA specifica-
tions. Ewing and Kruse (2006) find that people would be
willing to pay 9600 extra for shelters certified by the
National Storm Shelter Association; we will control for
old homes in our econometric analysis.
8. We also estimated the model excluding the 20 homes that
sold for more than g I million, but the results were very
similar to column one and so we do not report this
specification here.
References
Beron, K.J., Murdoch, J.C., Thayer, M.A. and
Vijverberg, \f.P.M. (L997) An analysis of the housing
market before and after the 1989 Loma Prieta earthquake.
Land Economics, 73, l0I-13.
Brookshire, D.S., Thayer, M.A., Tschirhart, J. and
Schulze, \7.D. (1985) A test of the expected utility model:
evidence from earthquake fisks. Journal of Political Economy,
93, 369-g9.
Camerer, C. and I{unreuther, H. (1989) Decision processes
for low probability events: policy implications. Joumal of
Policy Analysis and Management,8, 565-92.
Carbone, J.C., Hallstrom, D.G. and Smith, V.K. (2006) Can
natural experiments measure behavioral responses to
environmental risks? Environmental and Resource Econo-
mics, 33(3), 273-92.
Ehrlich, I. and Becker, G.S. (1972) Market insurance, self-
insurance, and self-protection. Journal of Pol,itical Economy,
80(4), 623-48.
Ewing, B.T. and I{ruse, J.B. (2006) Valuing self protection:
income and certification effects for safe rooms. Construction
Management and Economics, 24(10), I 057-68.

a
1.'
I
(
c\
o
-o
E
0)
o
z
N
N
l'*
:
.9
c
(u
o
t'
c)
f
9-
;
co
E
c.)
ro
(u
o
c
3
o
o
tL24
Ewing, B.T., Kruse, J.B. and Wang, Y. (2007) Local housing
price index analysis in wind-disaster-prone areas. Natural
Hazards, 40(2), 463-83.
FEMA (1999) Taking Sheher from the Storm: Building a Safe
Room Inside Your House, 2nd edn, FEMA publication 320,
Washington, DC.
Hallstrom, D.G. and Smith, V.K. (2005) Market responses ro
hurricanes. Joumal of Enaironmental Economics and
Managemenr, 50(3), 541*6L
I(unreuther, H. (1998) A program for reducing disaster
losses through insurance, in Kunreuther, H. and Roth
Jr, R.J. Paying the Price, Joseph Henry Press, \Tashington,
DC, pp. 209-28.
Kunreuther, H. and Kleffner, A. (1992) Should earthquake
mitigation measures be voluntary or required? Journal of
Regulatory Economics, 4, 321-35.
MacDonald, D.N., Murdock, J.C. and \White, H.L. (1987)
IJncertain hazards, insurance, and consumer choice:
evidence from housing markets. Land Economics, 63,
361-7 L.
Mileti, D.S. (1999) Disasters by Design, Joseph Henry Press,
Washington, DC.
Multihazard Mitigation Council (2005) Narural Hazard
Mitigation Saves Lives: An Independent Study to Assess the
Future Saztings from Mitigation Actiaities, 2 vols, National
Institute of Building Sciences, Washington, DC.
Simmons and Sutter
Ozdemir, O. (2005) Risk perception and the value of safe
rooms as a protective measure from tornadoes: a survey
method, in Ewing, B.T. and Kruse, J.B. (eds) Economics
and Wind. Nova Science Publishing, Hauppauge, NY,
pp. 89-104.
Peacock, $7.G. (2003) Hurricane mitigation status and factors
influencing mitigation status among Florida's single-family
homeowners. Natural Hazards Reaieut, 4(3), 149-58.
Shilling, J.D., Benjamin, J.D. and Sirmans, C.F. (1985)
Adjusting comparable sales for floodplain location.
Appraisal Joumal, 53, 429-36.
Simmons, K.M. and Sutter, D. (2006) Direct estimation of
the cost effectiveness of tornado shelters. Risk Analysis,
26(4),945-54.
Simmons, K.M. and Sutter, D. (2007) Tomado shelters andthe
manufactured home market. Natural Hazards, forthcoming.
Simmons, K.M., I(ruse, J.B. and Smith, D.A. (2002) Valuing
mitigation: real estate market response to hurricane loss
reduction measures. Southern Economic Journal, 68(3),
660-71.
Spreyer, J.F. and Ragas, \7.R. (1991) Housing prices and
flood risk: an examination using spline regression. Journal
of Real Estate Finance and Economics, 4, 395-407.
Viscusi, W.K., Vernon, J.M. and Harrington Jr, J.E. (2000)
Economics of Regulation and Antitrust,3rd edn, MIT Press,
Cambridge, MA.