Catastrophic Risk and Credit Markets

Mark J. Garmaise

and Tobias J. Moskowitz

∗

ABSTRACT

We provide a model of the eﬀects of catastrophic risk on real estate ﬁnancing and prices and

demonstrate that insurance market imperfections can restrict the supply of credit for catastrophe-

susceptible properties. Using unique micro-level data, we ﬁnd that earthquake risk decreased

commercial real estate loan provision by 22 percent in our California properties in the 1990’s. The

eﬀects are more severe in African-American neighborhoods. We show that the 1994 Northridge

earthquake had only a short-term disruptive eﬀect. Our basic ﬁndings are conﬁrmed for hurricane

risk, and our model and empirical work have implications for terrorism and political perils.

∗

UCLA Anderson School and Graduate School of Business, University of Chicago and NBER, respectively. We thank

Bing Han, Robert Novy-Marx and participants at the Vail Real Estate Research Conference for helpful comments.

We thank AIR Worldwide Corporation, Guy Carpenter and COMPS.com for providing data. Moskowitz thanks the

Center for Research in Security Prices and the Neubauer Family Faculty Fellowship for ﬁnancial support.

Correspondence to: Mark Garmaise, UCLA Anderson School, 110 Westwood Plaza, Los Angeles, CA 90095-1484.

E-mail: [email protected].

Catastrophic events can dramatically aﬀect the well-being of people throughout entire regions.

Episodes such as September 11, 2001, the December 26, 2004 tsunami in Asia, and Hurricane

Katrina in August 2005 highlight the risks borne by individuals, particularly those with limited

ﬁnancial resources. Financial markets can help to manage these risks by playing two crucial roles.

First, markets provide a mechanism through which risk is allocated eﬃciently. Second, ﬁnancial

markets can serve as a stable source of funding during post-catastrophe periods. Little is known,

however, about how well ﬁnancial markets perform these functions. In this paper, we provide

a model of the eﬀects of catastrophe risk on the ﬁnancing and pricing of properties and ﬁnd

corroborating evidence for the model using unique micro-level data on earthquake risk (the average

annual loss due to earthquake damage) and credit. We show that apparent ineﬃciencies in the

supply of catastrophe insurance have a substantial ongoing distortionary eﬀect on credit markets.

In particular, our results indicate that earthquake risk reduced the provision of bank ﬁnancing

by approximately 22 percent in California commercial real estate loan markets in the 1990’s. We

also ﬁnd, however, that the large 1994 Northridge earthquake aﬀected the market for only about

three months following the event. This suggests that while catastrophic risk may not be generally

allocated eﬃciently, the additional distortions caused by even signiﬁcant catastrophic events are

quite short-lived. Our work highlights general features of catastrophic risk markets that are shared

by a variety of perils including hurricane, terrorism and political risks. We extend our basic ﬁndings

to hurricane risk and argue that our results imply that, in the absence of well-functioning insurance

markets, terrorism risk is likely to discourage bank ﬁnancing of properties in high proﬁle U.S. cities

and political risk may impede the development of corporate debt markets in emerging economies.

Our model examines the potential distortionary eﬀect of catastrophe risk on credit markets.

We emphasize that bank ﬁnancing of catastrophe-susceptible properties is likely to be ineﬃcient

for two reasons. First, banks are not expert in assessing or repairing catastrophic damage to prop-

erties. As a result, banks are likely to ineﬃciently liquidate properties aﬀected by a catastrophe.

Second, banks do not specialize in monitoring whether owners are implementing all positive NPV

safety-enhancing investments. In the presence of a bank loan, owners may prefer not to make

these investments. Insurers, by contrast, are expert in b oth assessing catastrophic damages and in

monitoring the execution of safety-increasing improvements, so the presence of a well-functioning

The management of catastrophic risk has been the theme of a recent stream of research analyzing insurance

(Jaﬀee and Russell, 1997, Niehaus, 2002, Zanjani, 2002), reinsurance (Froot and O’Connell, 1997, Froot, 2001) and

catastrophic-loss derivatives (Cummins, Lalonde and Phillips, 2004).

insurance market can ameliorate the problems in bank ﬁnancing of properties at risk of a catastro-

phe. Froot (2001), however, contends that catastrophe insurance is over-priced and in relatively

short supply due to capital market imperfections and market power enjoyed by the relatively small

number of catastrophe reinsurers. We show that a p oorly functioning catastrophe insurance market

will lead to less bank ﬁnancing of catastrophe-susceptible properties, reduced market participation

by less-wealthy investors and incomplete insurance coverage. Ineﬃciencies in the catastrophe in-

surance market will also derail positive NPV investments by investors who require loans.

To test this theory, we perform an empirical analysis using unique data on catastrophic earth-

quake risk and commercial property loan contracts and prices in the U.S. in the 1990s. Using data

from Standard and Poor’s (S&P) ratings of commercial mortgage-backed securities (CMBS), we

show that only 35% of properties in earthquake zones carry earthquake insurance and the prob-

ability that insurance is purchased is increasing in earthquake risk. These ﬁndings are consistent

with our model and suggest that earthquake insurance is ineﬃciently supplied.

To analyze the impact of earthquake risk on the provision of ﬁnance, we match property-level

ﬁnancing and price information on commercial properties from COMPS.com with a unique data

set of micro-level earthquake risks, provided by AIR Worldwide Corporation (AIR). We ﬁnd that

increased earthquake risk dramatically reduces the likelihood that a property will be ﬁnanced

with bank debt, controlling for census tract ﬁxed eﬀects. This within-neighborhood identiﬁcation

is empirically feasible b ecause diﬀerences in soil conditions create highly localized variation in

the eﬀects of earthquakes; the AIR earthquake risks reﬂect both fault location and detailed soil

condition data. Our results suggest that in Los Angeles county, for example, the median quake

risk reduces the probability of bank ﬁnancing by over 20 percent. This evidence indicates that

imperfections in the allo cation of catastrophe risk disrupt credit markets in a manner consistent

with the theory.

We then examine the cross-section of properties to determine if these eﬀects are stronger for

diﬀerent groups of buyers or properties in ways predicted by the theory. We show that when insur-

ance ﬁrms (insurers or insurance brokers) purchase properties, earthquake risk has a signiﬁcantly

smaller aﬀect on the probability that a bank loan is used to ﬁnance the property than it does for

other buyers. This is consistent with the idea that insurance ﬁrms have better access to earthquake

insurance and are therefore not as severely aﬀected by the general lack of supply. As predicted

by the model, the use of bank credit by insurance ﬁrms is thus less distorted by the presence of

earthquake risk.

We also ﬁnd that older properties and properties in areas with large African-American popula-

tions are especially unlikely to be ﬁnanced with bank debt in the presence of quake risk, controlling

for the overall provision of bank loans. Earthquakes are likely to cause greater damage to older

properties, so quake risk is probably most important for these buildings. Earthquake insurance

may be particularly hard to obtain in African-American neighborhoods, which would create more

serious credit distortions linked to earthquake risk in these areas. Although it is not directly re-

lated to our theory, we further show that quake risk has an unusually strong negative eﬀect on the

provision of bank credit in areas slated for development.

Earthquake risk also inﬂuences the characteristics of buyers and ﬁnancing banks. We ﬁnd that

in the pool of non-corporate buyers, the purchasers of high quake risk properties come from zip

codes with higher median home values. This supports the implication of the model that properties

with higher catastrophic risks will be purchased by buyers who are wealthier on average. We

also show that local banks are relatively more likely to ﬁnance high quake risk properties, which

is consistent with the idea underlying the theory that damage assessment and monitoring (both

better performed by nearby, rather than distant, banks) are more important for properties with

greater catastrophe risk.

In addition to its disruptive inﬂuence on real estate ﬁnancing, our model shows that catastrophic

risk has a direct eﬀect on asset pricing: properties at risk for earthquake damage should have lower

prices, reﬂecting their increased potential for physical destruction. We ﬁnd, however, that there is

substantial variability in the manner in which quake risk is priced. In particular, we show that it is

only in larger deals that buyers consistently apply greater discounts to properties with higher quake

risk than others in the same census tract. Small buyers may not have access to highly localized

quake risk data due to either a lack of information or because acquisition of such information

involves paying some ﬁxed cost and they are of insuﬃcient scale.

We also examine the aftermath of one of the largest earthquakes in recent history, the January,

1994, Northridge earthquake, which caused an estimated $42 billion in damages ($14 billion of

which was insured). The Northridge quake caused a negative shock to the supply of earthquake

insurance, and we study the impact and longevity of this shock. Our analysis shows that, consistent

with the theory, properties with high quake risk were especially unlikely to be ﬁnanced with bank

loans in the period directly following the Northridge quake. The insurance supply shock generated

by the quake further exacerbated the reduced provision of bank loans to high catastrophe risk

properties. The duration of this eﬀect was approximately 3 months. We demonstrate that local

banks were less likely to make loans to high-quake-risk properties in the period after the event. We

also ﬁnd that, as the model predicts, bank-ﬁnanced transactions were concentrated in lower-risk

properties following the Northridge quake, while cash-ﬁnanced transaction displayed no such shift.

All these eﬀects were short-term and the earthquake appears to have had no signiﬁcant medium-

or long-term impact on the pricing and ﬁnancing of catastrophe risk.

Our model emphasizes general features of catastrophic risk markets: ineﬃciencies in bank

liquidation of catastrophically-damaged properties, lack of bank specialization in monitoring safety-

improving investments and restricted supply of catastrophe insurance. These features are shared

by an assortment of catastrophic perils including earthquake, hurricane, terrorism and political

risks. Earthquake data is particularly suitable for testing the theory for two reasons. First, risk

assessors can generate objective quantitative measures of earthquake risk. Second, earthquake risk

varies at the highly local level, which enables the use of census tract ﬁxed eﬀects to control for

unobservables. Our empirical results using the earthquake data show the relevance of the theory

and therefore support the application of the model’s predictions to other catastrophic risk markets.

To explore the broader implications of our ﬁndings, we extend the empirical work to an analysis

of hurricane risk, which shares many of the features of earthquake peril. While the properties in

our sample have relatively low exposure to hurricanes, we do ﬁnd evidence that properties with

higher hurricane risk than others in the same zip code tend to receive less bank ﬁnancing. This

eﬀect is magniﬁed when the price of catastrophe insurance is high. These results provide further

evidence for the theory in a second catastrophic risk setting.

Terrorism risk has recently become a subject of intense interest. We argue that the supply of

terrorism risk insurance is likely to be even more restricted than that of natural disaster insur-

ance for three reasons. First, terrorism risk is particularly diﬃcult to evaluate and this ambiguity

can hamper the supply of insurance (Hogarth and Kunreuther, 1985). Second, terrorism is en-

dogenous, so markets set up to allocate terrorism risk may be manipulated (Poteshman, 2006).

Third, the damages from a catastrophic act of terrorism may be may greater than those caused

by natural phenomena. As a result, the eﬀects we document for earthquake risk are likely to be

more severe for terrorism. This suggests that a decision by the government to decline to support

the terrorism insurance market (for example, by refusing to renew the Terrorism Risk Insurance

Act) can have important consequences. In particular, our work suggests that in the absence of

government-subsidized terrorism insurance, high proﬁle and high density areas such as the down-

towns of large U.S. cities would be likely to experience both a signiﬁcant shift away from bank

ﬁnancing of properties and market exit by less well-capitalized investors. Our ﬁndings also indicate

that reconstruction after a terrorism incident would likely be hampered by an especially limited

supply of credit in the immediate period following an event.

Political risk such as nationalization or currency controls (which are typically of greatest concern

in emerging economies) exhibits properties similar to those of other catastrophic perils: huge po-

tential losses and lack of bank skill in liquidating aﬀected investments. Political risk, like terrorism,

also faces the problem of uncertain hazard assessment, and the supply of political risk insurance

is quite limited (Hamdani et al., 2005). Our model and empirical work thus indicate that less

well-capitalized ﬁrms that require ﬁnancing will be signiﬁcantly more likely to invest in emerging

markets if there is a suﬃcient supply of fairly priced political risk insurance. Our results linking

catastrophe insurance to loan provision also suggest that an eﬃcient political risk insurance market

will encourage the issuance of emerging market corporate bonds.

The remainder of the paper is organized as follows. Section I describes our theoretical model

of the pricing and ﬁnancing of catastrophe risk. Section II details the commercial real estate and

earthquake data. Section III investigates the eﬀects of earthquake risk on real estate ﬁnancing and

prices. Section IV analyzes the impact of the Northridge quake. In Section V we consider the

implications of our ﬁndings for hurricane, terrorism and political risks. Section VI concludes.

I. Model

We develop a theory of the pricing and ﬁnancing of properties in the presence of catastrophic risk.

We begin with a simple model of identical investors all of whom are ﬁnancially unconstrained.

We then consider the presence of an investor with a higher valuation (or private beneﬁts), who is

potentially ﬁnancially constrained. We will argue that banks are ineﬃcient at ﬁnancing catastrophe-

susceptible properties, but that this ineﬃciency can be ameliorated by insurers. We conclude the

model by examining the implications of imperfections in the supply of catastrophic insurance for

the ﬁnancing of properties.

A. Unconstrained investors

For simplicity, consider a property generating cash ﬂow C

each period t with an associated dis-

count rate r for these cash ﬂows. We assume that C

t+1

= α

t+1

, where the {α

} are mutually

independent and E[α

] = 1 for all s. This implies that E[C

t+1

] = C

. Properties diﬀer in

their susceptibility to catastrophic (e.g., earthquake or hurricane) damage.

For each property, a

parameter p describes both the probability that the property will be aﬀected by a catastrophe and

the severity of any such catastrophe. Each period t with probability p there is no catastrophe.

With probability 1 − p a catastrophe occurs and leaves the property with a salvage value that is

a fraction R

of the undamaged property value V

, where R

∈ [0, 1] is a random variable.

also assume for simplicity that the current cash ﬂows are lost.

We presume that the {R

} are

identically and independently distributed, with common mean E[R] ∈ (0, 1). We refer to (1 − p) as

the catastrophic risk of the property.

In the event of a catastrophe, we assume that instantaneous repairs costing (1 − R

must

be undertaken before the property will generate future cash ﬂows.

The net present value of these

repairs when there is a catastrophe is N (V

, R

) = V

≥ 0. We assume that catastrophe risk

(both frequency and severity) is uncorrelated with economy-wide ﬁnancial wealth (as argued by

Froot, 2001) and that investors are well diversiﬁed and ﬁnancially unconstrained. Investors will

always undertake the repairs, since N ≥ 0.

The average annual loss q is deﬁned as q = (1 − p)(1 − E[R]), which is the expected fraction

of property value that is destroyed by the catastrophe in each period. The average annual loss

is sometimes described as the catastrophe premium (R¨uttener, Liechti and Eugster, 1999). Using

standard arguments, we show in the Appendix that V

r+q

. The cap rate (ratio of earnings to

price) is thus given by

r+q

The above straightforward discrete time model yields the basic intuitions necessary for our tests.

Duﬃe and Singleton (1999) provide a general theoretical treatment of the pricing of assets in the

The model applies to all cases of damage risk (e.g., ﬁre), but, as we will discuss, the insurance supply distortions

that play a role in the theory are most plausible for catastrophe risk.

The scientiﬁc literature on earthquakes has established the Gutenberg-Richter frequency magnitude law, which

states that magnitudes are governed by an exponential distribution, with relatively little variation across locations in

the exponential parameter (Kagan, 1997). While the frequency of earthquakes can vary substantially across regions,

the distribution of earthquake severities, given an event, is actually quite stable across areas. This result suggests

that modelling the severity of the damage as independent of its frequency is reasonable. Hurricanes, cyclones and

ﬂo ods also exhibit similar frequency-magnitude patterns (MacDonald, 2000).

The results hold for fractional cash ﬂow losses as well.

Assuming that a catastrophe also changes the future cash ﬂows generated by a property has no eﬀect on the

results.

presence of exogenous value-destruction risk.

Our ﬁrst result describes the pricing eﬀects of catastrophe risk: properties subject to catastrophe

risk sell for lower multiples of their current cash ﬂows.

Result 1. The derivative of the cap rate with respect to the average annual loss is at least one.

The proofs of all Results are given in the Appendix.

B. Constrained Investors and Diﬀerential Valuations

We now introduce two modiﬁcations to the base model. First, for a given property with current

value V

in period s, with probability ζ there is an investor who can generate an additional value

b ≥ 0 from purchasing the property and managing it for the next period. This additional valuation

might arise from a positive net present value investment known only to the investor or from a

private beneﬁt. For simplicity we will assume that b is a private beneﬁt, though our results are not

sensitive to that assumption. The realization of b is known to the investor.

Second, we assume that this investor has wealth w

< V

with probability l ∈ (0, 1). With

probability (1−l ), the investor has wealth w

> V

. That is, we now allow for ﬁnancially constrained

investors who do not have enough cash to purchase the property. The investor with a private beneﬁt

pays only the market value V

if he purchases the property. This is true, for example, if the property

is sold in a public auction.

B.1. Ineﬃciency of Bank Financing in the Absence of Insurance

We begin by assuming that no catastrophic insurance is available. Constrained investors will require

a bank loan. Banks have unlimited capital, and we assume that the banking sector is competitive;

our focus is on credit market distortions that are related to catastrophic risk, rather than general

credit market distortions. We presume that the bank ﬁnancing of catastrophe-susceptible properties

is ineﬃcient. This ineﬃciency may arise from two possible sources.

First, banks do not specialize in the evaluation or remediation of catastrophic damage. We

presume that a catastrophe creates an information asymmetry between the owner, who can gauge

the damage, and the bank, which does not specialize in damage assessment. Banks lack the or-

ganizational skill to evaluate or repair catastrophic damage.

While asymmetric information and

liquidation (i.e., foreclosure) costs are, in practice, to some degree present for every property, these

Anderson and Weinrobe (1986) and Shah and Rosenbaum (1996) ﬁnd that large earthquakes lead to mortgage

defaults. Ciochetti (1997) shows that the costs of foreclosure on commercial mortgages can be substantial.

considerations are greatly magniﬁed for properties damaged in an earthquake.

Second, banks do not specialize in monitoring whether a property owner is implementing all pos-

itive net present value (NPV) damage-reducing safety investments. In the presence of bank debt, an

owner may prefer to forgo these investments, since he b ears the costs and the bank enjoys a portion

of the beneﬁts (Smith and Warner, 1979). The bank ﬁnancing of catastrophe-susceptible properties

will thus lead to ineﬃcient underinvestment in positive NPV safety enhancement projects.

Consider a property with value V

that may experience a catastrophe with probability (1 − p).

Both of the two arguments just given suggest that if an investor purchases the property with w in

equity and a bank loan of V

−w ≥ 0, then the ineﬃciency associated with bank ﬁnancing will result

in value destruction, which we denote by loss(w, V

, p). We will assume that the value destruction

satisﬁes two properties:

Property 1: loss(w, V

, p) is decreasing in p and w

Property 2: loss(w, V

, p) > 0 if V

> w and p < 1, but loss(w, V

, p) = 0 if V

= w or if p = 1

The ﬁrst property captures the idea that the expected eﬃciency losses generated by either costly

bank liquidation of damaged properties or by poor bank monitoring of damage prevention projects

are more severe both when the catastrophe risk is high and when the equity provided by the investor

is low. The second property speciﬁes that if there is no catastrophe risk, or if the project is wholly

equity ﬁnanced, then there is no value destruction. The presence of some catastrophe risk combined

with bank ﬁnancing always leads to a measure of ineﬃciency, however small.

In the appendix we provide a formal model of the ﬁrst argument analyzing the eﬀects of a

bank’s inability to assess or repair catastrophic damage. We show that this formal model yields

the two properties detailed above. The intuition underlying the second argument suggests that

the importance of damage prevention projects will increase in the catastrophe risk. Investors who

invest more equity will capture more of the beneﬁts of the projects, and are thus more likely to

pursue them. The second argument thus quite naturally also gives rise to the two properties.

We deﬁne Π

(w, V

, p) to be the net present value received by this investor if he invests w ≤ V

in cash and borrows V

− w in order to purchase a property with catastrophe frequency (1 − p).

Thomas (1997), for example, describes a set of apartment buildings moderately or severely damaged in the

Northridge quake that were seized by a bank after mortgage default. Three and a half years after the quake, fewer

than half the buildings were habitable. Repair of non-defaulted properties was essentially complete by that time.

Damage assessment is diﬃcult even for insurers. Estimates by insurers of their losses in the January 1994 Northridge

earthquake grew from $5.5 billion in June 1994 to $12.5 billion in July 1995 (Roth, 1998).

The value of the investor’s equity claim is equal to the value V

of the property minus the value

of the bank’s debt claim minus any value destroyed in ineﬃcient liquidation. Credit markets are

competitive, so the net present value of the investor’s investment is therefore equal to his private

beneﬁt minus the value destroyed in ineﬃcient liquidation. Since not investing is always an option,

we have

(w, V

, p) = max{b − loss(w, V

, p), 0}.

B.2. Insurance Market

We now introduce the possibility of catastrophe insurance. We will assume that there are two

potential insurers and that they oﬀer full insurance contracts (covering all cash ﬂow and property

value losses), though our results are not sensitive to either of these assumptions. Insurers diﬀer

from banks in two respects. First, insurers are assumed to have the ability to swiftly evaluate the

extent of catastrophic damage for claims payment purposes. Second, insurers can monitor owners

and ensure that they make damage-preventing positive NPV investments. We presume that loan

contracts can be written that have a ﬁrst priority claim on insurance payouts. The provision

of insurance thus removes the two sources of ineﬃciency associated with the bank ﬁnancing of

properties subject to catastrophe risk. The insurer’s claim payment in the event of a catastrophe

reveals the extent of the damage to the bank and removes the information asymmetry between the

property owner and the bank. The insurer will also undertake the monitoring of safety projects.

We presume, however, that the catastrophe insurance market is potentially imperfectly com-

petitive. In particular, due to market power or capital constraints on the part of reinsurers (Froot,

2001), we assume that each insurer is able to make a bid on any given catastrophe insurance con-

tract with probability n ∈ [0, 1]. Whether an insurer can bid is independent of what happens to his

competitor. Each insurer who can bid proposes a price for full insurance, and the property owner

may select the lower bid or choose not to purchase insurance. We describe the insurance market as

perfectly competitive if n = 1 and imperfectly competitive otherwise. The fair value of insurance

is the market value of the insurance payout.

We describe the bid of an insurer by viewing it as a surplus, sur, over the fair value of the

insurance. If both insurers bid for the contract, the unique equilibrium is for each to set sur = 0. If

only one insurer bids, he will choose sur to maximize his exp ected proﬁts. Financially unconstrained

For simplicity, we ignore the costs associated with damage assessment and monitoring. Including these costs does

not aﬀect the results.

investors will never pay a positive premium for insurance, since they need not ﬁnance with a loan

and will therefore never suﬀer the ineﬃciencies of bank ﬁnancing, so the insurer maximizes the

proﬁts he realizes from selling to constrained investors. Any bid sur > w

will be rejected, since

the investor will have insuﬃcient resources to pay this amount. The bid will also be rejected if it is

less expensive for the investor to purchase the property without insurance: sur > l oss(w

, V

, p).

Lastly, if sur > b, the investor will prefer to forego the property purchase rather than buying

insurance.

Insurers do not know the realization of the private beneﬁt b, but they know its associated cdf

and pdf f

. We assume that F

satisﬁes the MHR property. We denote by b

∗

the solution to the

following maximization problem:

max

y≥0

y (1 − F

(y)) +

(t)tdt

. (1)

The MHR property guarantees that b

∗

is unique. An insurer bidding alone will set sur =

min{b

∗

, w

, loss(w

, V

, p)}. As a property’s catastrophic risk increases, Property 1 indicates that

the option of ﬁnancing with a bank loan becomes increasingly unattractive. This implies that as

catastrophe risk increases two things occur in an imperfectly competitive insurance market: A

single insurer bidding will charge a higher surplus and if no insurer bids the investor is less likely to

proceed with the purchase. Both of these eﬀects reduce the frequency of bank-ﬁnanced transactions.

If the insurance market is perfectly competitive, insurance is always supplied at zero surplus.

Result 2. If the insurance market is imperfectly competitive, the probability that a transaction

is ﬁnanced with bank debt is decreasing in the property’s catastrophic risk. If the insurance market

is perfectly competitive, the probability that a transaction is ﬁnanced with bank debt is independent

of the property’s catastrophic risk.

In an imperfectly competitive insurance market, expensive insurance premiums combined with

the ineﬃciency of bank ﬁnancing in the absence of insurance together discourage less wealthy

investors from purchasing properties with high catastrophic risk. Even though the credit mar-

ket itself is competitive, the insuﬃcient supply of insurance leads to distortions in the ﬁnancing

of catastrophe-susceptible properties. Investors with substantial wealth purchase the properties

without insurance or a loan.

Result 3. If the insurance market is imperfectly competitive, the average wealth of the pur-

chasing investor is increasing in the property’s catastrophic risk. If the insurance market is per-

fectly competitive, the average wealth of the purchasing investor is independent of the property’s

catastrophic risk.

As catastrophic risk increases, Property 1 shows that fewer ﬁnancially constrained investors

will buy the property without insurance. In an imperfect insurance market, if insurance is oﬀered,

the surplus demanded will increase with catastrophic risk, but in such a way that the investor will

still prefer to buy insurance rather than purchasing the property without it. The net eﬀect is that

as catastrophic risk increases, more of the bank-ﬁnanced transactions will carry insurance. In a

perfect insurance market, insurance will always b e purchased because it avoids the ineﬃciencies of

bank ﬁnancing.

Result 4. If the insurance market is imperfectly competitive, the probability that a bank-ﬁnanced

transaction is accompanied with insurance is increasing in the property’s catastrophic risk. If the

insurance market is perfectly competitive, all bank-ﬁnanced purchases of properties with positive

catastrophic risk will be accompanied with insurance.

As catastrophic risk increases, in an imperfect insurance market, two classes of investors elect

not to purchase the property. First, constrained investors who receive no insurance bid and ﬁnd the

cost of ﬁnancing the property without insurance too high. Second, constrained investors who receive

one bid and ﬁnd both the insurance bid and the cost of ﬁnancing the property without insurance

too high. Since it is economically eﬃcient for the investor to always purchase the property, this

suggests that there are social welfare costs arising from imperfections in the insurance market:

positive net present value projects must be abandoned.

Result 5. If the insurance market is imperfectly competitive, the probability of a transaction is

decreasing in the property’s catastrophic risk. If the insurance market is perfectly competitive, the

probability of a transaction is independent of the property’s catastrophic risk.

We also consider the eﬀects of a change in the competitiveness of the insurance market, such

as might arise following a large catastrophic event. As competitiveness increases, more ﬁnancially

constrained investors will be able to purchase aﬀordable insurance and buy properties.

Result 6. The probability that a transaction is ﬁnanced with debt is increasing in insurance

market competitiveness n.

An increase in competitiveness will particularly beneﬁt constrained investors considering pur-

chasing high-catastrophe-risk properties. Thus, as competitiveness increases, there should be an

especially large increase in bank-ﬁnanced purchases of high-risk properties. Unconstrained in-

vestors using cash to make their property purchases will be unaﬀected by the competitiveness of

the insurance market.

Result 7. The average catastrophic risk of bank-ﬁnanced transactions is increasing in insurance

market competitiveness n. The average catastrophic risk of all-cash transactions is independent of

insurance market competitiveness n.

In Section III we will test these predictions using empirical data on earthquakes. Clearly, in

practice, the insurance market will not be perfectly competitive. Our empirical results, however,

will examine the importance of this imperfection and quantify its spillover eﬀects on credit markets.

II. Data and Summary Statistics

We brieﬂy describe the variety of data sources used in the paper.

A. Transaction-level data from the U.S. commercial real estate market

Our transaction-level commercial real estate sample consists of 32,618 transactions drawn from

across the U.S. over the period January 1, 1992 to March 30, 1999 compiled by COMPS.com, a

leading provider of commercial real estate sales data. Garmaise and Moskowitz (2003, 2004) provide

an extensive description of the COMPS database and detailed summary statistics. The data span

11 states: California, Nevada, Oregon, Massachusetts, Maryland, Virginia, Texas, Georgia, New

York, Illinois, and Colorado, plus the District of Columbia.

Commercial properties are grouped into ten mutually exclusive types: retail, industrial, apart-

ment, oﬃce, hotel, commercial land, residential land, industrial land, mobile home park and special.

Panel A of Table I reports summary statistics on the properties in our sample. The average (me-

dian) sale price is $2.2 million ($590,000), and there are only 42 transactions involving REITS (less

than 0.2% of the sample). Capitalization rates, deﬁned as current net income on the property

divided by sale price, and property age are also reported.

The COMPS database provides detailed information about speciﬁc property transactions, in-

cluding property location, identity and location of market participants, and ﬁnancial structure.

In particular, COMPS provides eight digit latitude and longitude coordinates of the property’s

location (accurate to within 10 meters).

The COMPS data contain ﬁnancing information for each property transaction. We focus on the

terms of the loan contract, including interest rates, and the size and presence of loans. As Panel A

of Table I indicates, the average loan size (from bank and non-bank institutions) as a fraction of

sale price is over 75%. Bank loans are used in 53% of transactions, vendor-to-buyer (VTB) loans

(i.e., seller ﬁnancing) are used in 19% of deals and less than 5% of deals involve assumed debt. The

data also contain rich detail on loan terms including the annual interest rate, the maturity of the

loan, whether the loan rate is ﬂoating or ﬁxed, whether amortized and the length of amortization,

and whether the loan is subsidized by the Small Business Administration (only 1.3% of loans). The

COMPS data do not, however, include insurance information on properties.

B. Commercial Mortgage-Backed Securities Data

We make use of data on CMBS transactions provided by the S&P RatingsDirect database. This

data provides descriptive details on ﬁfty CMBS transactions over the period 1996 to 1999. Each

CMBS issuance covers multiple properties, as described in Panel B of Table I, and in many cases

information on earthquake risk and insurance is given.

C. Earthquake risk

AIR Worldwide Corporation (AIR) provides detailed data on the earthquake risks associated with

our COMPS properties’ locations. AIR is a highly regarded vendor of estimates for various types

of catastrophe risks. Using its proprietary CATStation Hazard Module, AIR generates location-

speciﬁc assessments of the expected average annual loss due to earthquake risk. (Cummins, Lalonde

and Phillips, 2004 describe the AIR catastrophe models.) The average annual loss denotes the

fraction of property value that is expected to be destroyed by an earthquake in any given year. It is

expressed as a percentage, and it reﬂects both the likelihood of an earthquake and the distribution

over potential severities. Prop erty characteristics will also have an eﬀect on the impact of an

earthquake, but the AIR estimates incorporate only location, not structure, characteristics.

use AIR’s estimate of average annual loss as our measure of quake risk. AIR provides location-

speciﬁc matches for each of our COMPS properties at eight digit latitude and longitude coordinates,

which are accurate to within 10 meters.

The AIR earthquake model uses both fault location and detailed soil condition data. Soil

characteristics have a large impact on the way seismic waves are transmitted. Using this data,

the AIR model makes highly localized predictions of average annual loss. For example, the AIR

soil database for the area around the San Francisco Bay has a horizontal resolution of 24 square

AIR does generate structure-speciﬁc estimates, but these were not provided to us.

meters.

Panel B of Table I presents summary statistics for AIR earthquake risks. For most properties

in our sample, the average annual loss is described as less than 0.1%, which we code as 0. There

are 9,785 properties with positive quake risks, all located in California, Oregon and a handful of

sites in Massachusetts. Our data include 12,288 properties in California and 9,386 properties in

Los Angeles county.

In many cases the S&P data supplies earthquake risk and insurance data on the properties

securing the CMBS issuance. For properties in earthquake zones (primarily California, but also

including parts of Nevada, Washington, Oregon, Utah and Missouri) the probable maximum loss

(PML) is often speciﬁed, along with information about whether the property has earthquake in-

surance. S&P deﬁnes the PML to be the expected earthquake damage (expressed as a fraction of

property replacement cost) that has a ten percent chance of being exceeded during a ﬁfty year pe-

riod (Standard and Poor’s, 2003). This measure of earthquake risk is closely related to the average

annual loss. In most cases, the S&P data will specify only whether the PML exceeds a threshold,

typically 20%.

D. The Northridge Earthquake

Our data and sample time period also allow us to consider the impact of an actual sizable earth-

quake, the Northridge earthquake of January 17, 1994. The Northridge earthquake measured 6.7

on the Richter scale, caused 57 deaths and was responsible for direct economic damages of approx-

imately $42 billion (of which $14 billion was insured), according to reliable estimates (Petak and

Elahi, 2000). The U.S. Geological Survey (USGS) provides data on the severity of ground mo-

tion during the Northridge quake.

Speciﬁcally, we consider the peak ground acceleration (PGA),

which is the maximum acceleration experienced at a speciﬁed location on the earth’s surface during

the course of an earthquake. The PGA is a commonly used metric for earthquake severity, and

building codes often describe requirements for withstanding shaking in terms of horizontal force,

which is related to PGA. Data is provided for points on a grid system, with a distance between

grid p oints of approximately 1.15 miles on the north-south axis and 0.94 miles on the east-west

axis. We match our property locations to the nearest grid point in order to infer the extent of local

This description of the AIR model is drawn from http://www.air-worldwide.com. The January 1994 Northridge

earthquake caused some revisions to earthquake risk assessments. All the results in the paper are robust to using

only post-January 1994 data.

The data may b e found at http://earthquake.usgs.gov/shakemap.

PGA. This process generates PGA estimates for every COMPS property in Los Angeles county.

Summary statistics are given in Panel B of Table I.

E. Crime and Census Data

We also make use of local crime and census data. The crime risk data is provided by CAP Index,

Inc., who compute a crime score for a particular location by combining data from police reports,

the FBI’s

Uniform Crime Report

(UCR), client loss reports, and oﬀender and victim surveys with

geographic, economic, and population data. The crime risk estimates are property-speciﬁc for

the COMPS data and vary within census tracts (see Garmaise and Moskowitz (2005) for further

details). The census data come from the 1990 and 2000 U.S. censuses.

III. The Eﬀects of Earthquake Risk on Financing and Prices

Using the earthquake risk and commercial property loan data, we test the hypotheses outlined in

Section I by examining the impact of earthquake risk on ﬁnancing and prices. We also provide some

general descriptive statistics on the eﬀects of earthquake risk on commercial real estate markets.

The commercial real estate market is a useful laboratory to investigate the role of catastrophe

risks because the loans are typically secured and non-recourse, providing a set of project-speciﬁc

ﬁnancings for which the collateral value is of central importance.

A. Quake risk and insurance

The theory in Section I advanced the argument that an imperfectly competitive catastrophe in-

surance market can imp ede the supply of credit in the real estate market. Catastrophe insurance

reduces the costs of securing a loan because it avoids ineﬃcient foreclosures (liquidations) by banks

after a catastrophe. If catastrophe insurance is priced at a premium, as suggested by Froot (2001),

then buyers who require a loan will only purchase insurance when the costs of ineﬃcient liquidation

are very high, i.e. for high-quake-risk properties. This idea is captured in Result 4, which states

that in imperfectly competitive insurance markets the probability that a bank-ﬁnanced transac-

tion is accompanied with insurance is increasing in the property’s catastrophe risk. In perfectly

competitive insurance markets, all bank-ﬁnanced properties should carry catastrophe insurance.

Insurance data on individual properties is quite diﬃcult to secure (Squires, O’Connor and Silver,

2001), but the S&P CMBS database provides this information on properties in quake-susceptible

areas as part of its ratings process. The properties in this database have securitized loans, which

make up a signiﬁcant (and growing) fraction of total commercial mortgage lending during our

sample period.

Hence, these properties are representative of our larger database on commercial

real estate from COMPS, which does not contain information on quake insurance. Panel C of Table

I shows that for the 482 properties in the S&P CMBS transactions that reside in quake-prone areas,

only 169 or 35% purchased earthquake insurance. Among properties facing the highest quake risk

(those with probable maximum loss, PML, > 20%), the fraction obtaining earthquake insurance is

54%. These results are consistent with Result 4 from the model of Section I.

As a more formal test, Table II presents results from regressions relating quake risk to the

purchase of quake insurance in the S&P CMBS database. There are 482 properties in earthquake

zones with catastrophe insurance information supplied by CMBS.

For these properties, we regress

a binary variable for whether quake insurance was purchased on an indicator variable for High quake

risk (PML above 20%). Unfortunately, the S&P CMBS database does not provide information on

quake risk other than identifying properties as high or low quake risk based on having a PML

greater than 20%. All regressions include year dummies and transaction attributes and robust

standard errors are reported throughout. In the ﬁrst column of Table II, we display results from

a logit regression. The coeﬃcient on High quake risk is statistically signiﬁcant with a t-statistic of

7.41. In the second column of Table II, we describe the results of a ﬁxed eﬀects (conditional) logit

regression that includes CMBS issuer ﬁxed eﬀects. The coeﬃcient on High quake risk is statistically

signiﬁcant with a t-statistic of 2.55. The economic magnitude of this eﬀect is quite large. The point

estimate on High quake risk implies a 33.80 percentage point increase from 35.06% to 68.86% in the

probability of earthquake insurance being provided.

(The estimate from the speciﬁcation in the

ﬁrst column is even larger.) For this sample of debt-ﬁnanced earthquake-zone properties, those with

a PML of at least 20% (e.g., high quake risk) thus have a dramatically higher probability of carrying

earthquake insurance than other properties.

The third column of Table II uses transaction-level

rather than issuer-level ﬁxed eﬀects. The magnitude of the coeﬃcient on high quake risk diminishes

somewhat, but remains signiﬁcant in both statistical (t = 3.04) and economic terms.

Vandell (1998) shows that by the end of 1997, 15.1 percent of general commercial real estate mortgage credit was

securitized, while 25.5 percent of apartment debt was securitized.

These properties are primarily located in California, though some are also located in the states of Nevada,

Washington, Oregon, Utah, and Missouri (Standard and Poor’s, 2003).

The economic magnitude of a ﬁxed eﬀects logit is best considered in terms of its impact on the odds ratio.

Earthquake insurance is provided for 35.06% of the CMBS earthquake-zone properties, which gives an o dds ratio of

0.3506

1−0.3506

= 0 .5399. Since the estimated coeﬃcient on High quake risk is 1.41, moving from low to high quake risk

multiplies the odds ratio by exp(1.41) = 4.10, which yields an odds ratio of 2.21, which is equivalent to an earthquake

insurance probability of 68.86%.

In general, insurance is especially valuable for low frequency, highly risky, events.

The ﬁndings in Table II are consistent with other work showing that lenders require earthquake

insurance for high risk properties (Glickman and Stein, 2005) and that commercial real estate

investors purchase earthquake insurance primarily at the request of lenders (Porter et al., 2004).

Overall, the results are consistent with the imperfect-insurance equilibrium of Result 4. Result 4

indicates that in an imperfectly competitive insurance market not all loan-ﬁnanced properties will

carry insurance, and the fraction with insurance will rise with quake risk. The evidence in Table II

and the fact that only 35% of the properties in quake zones carry insurance strongly supports the

model. The fact that these loans are securitized in public markets makes clear that diversiﬁcation

of catastrophe risk is not the primary role played by insurance in the ﬁnancing of properties. The

purchasers of public CMBS are very well diversiﬁed and cannot credibly be argued to be over-

exposed to California earthquake risk. The model presented in Section I, however, shows that even

properties ﬁnanced by well-diversiﬁed lenders will carry earthquake insurance, so that lenders (and,

indirectly, borrowers) can avoid ineﬃcient liquidation.

B. Quake risk and commercial ﬁnancing terms

Table II provides evidence of an imperfectly competitive catastrophe insurance market. According

to Result 2, in an imperfectly competitive insurance market the probability of bank ﬁnancing will

decline with catastrophic risk, while in a perfect insurance market ﬁnancing and catastrophic risk

will be unrelated. Therefore, as an additional test of the competitiveness of earthquake insurance

markets, we examine the relation between bank ﬁnancing and earthquake risk. Table III considers

the eﬀect of quake risk on the commercial real estate ﬁnancing terms oﬀered by banks in the

COMPS data set. To isolate the impact of quake risk, it is important to control for neighborhood

features, since the lending environment can vary across diﬀerent districts of a city (Ross and Tootell

(2004), Garmaise and Moskowitz (2005)). We conduct our tests using census tract ﬁxed eﬀects to

diﬀerence out unobservables at the census tract level. A census tract typically covers between 2,500

and 8,000 persons or about a 4-8 square block area in most cities, and is designed to be homogeneous

with respect to population characteristics, economic status, and living conditions (source: United

States Census Bureau). Quake risk, however, is not uniform within a census tract due to highly

localized variation in soil conditions. There are 1,210 tracts in our data set that contain properties

with p ositive quake risks, and 202 tracts (with 2,235 properties) that have within-tract variation in

quake risk.

While earthquake risk is clearly highly variable across diﬀerent regions of California

and the U.S., we emphasize here that our use of census tract ﬁxed eﬀects means that we are

only comparing properties within the same tract. Our econometric identiﬁcation arises solely from

within-tract variation.

More formally, our econometric model considers the eﬀect of earthquake risk on the provision

of loans by banks, a binary variable indicating whether a bank loan was obtained. The equation

estimated is,

P rob(bank f inancing)

= F (quake risk

, price

, controls

, property type, year, tract) + ²

, (2)

where controls

is a vector of controls containing a set of property and neighborhood attributes for

asset i, property type, year, and tract represent property type, year, and census tract ﬁxed eﬀects

in which the property resides, and ²

is an error term. The sale price is included as a regressor to

control for value in current use, thereby isolating the component of quake risk related to secondary

or collateral value, since the theory we aim to test is about liquidation value in the event of a

catastrophic risk. We estimate a logistic functional form for the binary dependent variable.

In advance of our discussion of the empirical results, it is worthwhile to consider the econometric

issues raised by our speciﬁcation in equation (2). The ﬁrst point is that the sale price itself may

be a function of quake risk; we might expect high quake risk properties to realize lower prices

according to Result 1. We examine the evidence testing this result in the next section. This issue

presents no special econometric problem because the logistic model can be estimated consistently

in the presence of correlations between independent variables.

The second, and more serious, issue is that some unobservable variable (such as local ﬁnancial

and economic conditions or quality of borrower or property) may have a simultaneous eﬀect on

loan provision, sale prices, and quake risk, rendering all of our variables endogenous and diﬃcult

to interpret. (That is, this would result in a correlation between the independent variables and the

error term.) We will return to a discussion of this omitted variables problem in Section III C.1.

We are essentially estimating reduced form equations for the probability, price, quantity, and

terms of the debt supplied. Consistent with the model in Section I and as argued earlier, we believe

these eﬀects are closer to supply-side constraints.

There are 334 zip codes that contain positive quake risk properties, and 131 zip codes (with 4,666 properties)

that have within-zip-code variation in quake risk. The earthquake risk results in the paper are robust to using either

zip code or census tract ﬁxed eﬀects.

B.1. Bank loan provision

In column 1 of Table III, we report results from regressing a binary variable indicating whether

or not the property purchase is ﬁnanced with a bank loan on quake risk – the average annual loss

from the AIR data – and a set of control variables. The control variables include a set of property,

borrower, and local market characteristics which include the log of the sale price, an indicator for

whether the transaction is brokered,

an indicator for whether the buyer is a broker himself, an

indicator for corporate buyers, the 1990 property and personal crime risks, the age of the property,

the distances of the buyer and seller from the property, an indicator for development projects,

and ﬁxed eﬀects for property type, year, and census tract. The estimation method is via ﬁxed

eﬀects (conditional) logit. The regression shows that properties subject to greater quake risk are

signiﬁcantly less likely to be ﬁnanced with a bank loan. The coeﬃcient on quake risk is −2.63 with

a t-statistic of −2.83. To evaluate the economic magnitude of this eﬀect, consider the Los Angeles

county observed frequency of ﬁnancing of 58.5% and median quake risk of 0.2%. The point estimate

on quake risk in the regression implies a 13.1 percentage point reduction in the probability of loan

provision from 58.5 to 45.4%.

This reduction is 22.3% of the mean ﬁnancing frequency. (The

mean quake risk in Los Angeles county is 0.25%, which generates an even larger eﬀect.) Examining

all the California properties in our data set, for which the mean quake risk is 0.19%, the conditional

logit estimate implies a 12.4 percentage point reduction in the probability of ﬁnancing, which is

22.2% of the mean. The size of these eﬀects suggests that quake risk dramatically reduces the

provision of bank ﬁnance to properties at higher risk within the same census tract (and controlling

for price and all the other attributes).

The substantial reduction in loan provision for high quake risk properties lends support to Result

2 from our model: in a well-functioning insurance market there should be no relation between

earthquake risk and loan provision. If catastrophe risk is not eﬃciently supplied, then credit

markets may be distorted. This evidence corroborates Fro ot’s (2001) contention that catastrophe,

in particular earthquake, risk is not optimally allocated across market participants and may not be

correctly priced. Moreover, the magnitude of the eﬀect we document suggests insurance markets

can have substantial distortionary eﬀects on credit markets. A 22% reduction in the frequency

Garmaise and Moskowitz (2003) show that brokers have a signiﬁcant eﬀect on the ﬁnancing of commercial

prop erties through their relationships with banks. We therefore add broker presence as an additional control.

The economic magnitude is calculated using the odds ratio, as described earlier. If, instead of a conditional logit

mo del, we run a ﬁxed eﬀect linear probability model (OLS), the estimated eﬀect of a 0.2% increase in quake risk is

−9.2 percentage points, with a t-statistic of −2.51 (not reported in the tables).

of bank ﬁnancing can have signiﬁcant eﬀects on the real economy, as the ﬁnance and growth

literature emphasizes ((Peek and Rosengren (2000), Cetorelli and Gambera (2001), Klein, Peek,

and Rosengren (2002), Burgess and Pande (2003) and Garmaise and Moskowitz (2005)).

emphasized in the model in Section I (Result 5), ﬁnancially constrained investors are forced to

forego positive NPV projects when they cannot purchase fairly priced insurance.

B.2. Bank loan terms and Seller ﬁnancing

In columns 2 and 3 of Table III we analyze the eﬀect of quake risk on the terms of the bank loan

contract. Our theory does not provide clear predictions about these terms but we provide some

empirical evidence to oﬀer additional descriptive information. We ﬁrst consider interest rates. In

our theory, some constrained investors purchasing quake-susceptible properties obtain a mortgage

and acquire quake insurance. For these loans, there is no quake risk faced by the borrower and

the loans are secured by both the property and the quake insurance. For a given ratio of loan

amount to property value, these loans are actually safer than loans to properties with no quake

risk (because of the value of the insurance) and may therefore carry lower interest rates. Other

constrained buyers purchasing properties subject to quake risk obtain a mortgage without quake

insurance. These loans will be riskier than loans secured by properties without quake risk and will

therefore carry higher rates. The overall eﬀect is ambiguous.

Column 2 reports regression results of the interest rate of the loan on quake risk. In addition

to the previous controls, the ratio of loan size to property price (loan-to-value), the debt maturity,

an indicator for ﬂoating rate loans, an indicator for Small Business Administration-backed (SBA)

loans, and the log of bank assets are included as regressors. We ﬁnd that quake risk has no

statistically or economically signiﬁcant eﬀect on the interest rate. A 0.2% increase in quake risk is

associated with a 17 basis point increase in the annualized interest rate (the average annual rate

in the sample is 8.3%) and is statistically no diﬀerent from zero. The magnitude of this eﬀect

contrasts sharply with the ﬁndings on loan provision and the data do not permit clear statistical

inference.

We next consider leverage ratios. The relationship between leverage ratios and quake risk is also

The overall rate of bank loan provision in California is 55.95%, which is actually slightly above the average rate

in the whole sample. There are, however, macroeconomic, legal and regulatory reasons for why loan provision rates

may diﬀer across states (and indeed, the fraction of properties ﬁnanced with bank loans varies quite dramatically

across states in our data). Our census-tract ﬁxed eﬀects control for all such factors and allow us to isolate the eﬀects

of quake risk alone. The implication of our ﬁndings is that in the absence of quake risk, loan provision rates in

California would be roughly 12.4 percentage points higher than what we observe in the data.

ambiguous in the model because any feasible loan amount is optimal when insurance is acquired.

Nonetheless, to fully describe the eﬀect of catastrophe risk on ﬁnancing terms, we regress loan

size on quake risk. As is shown in column 3 of Table III, conditional on a loan being extended,

the size of the loan does not depend on quake risk. In column 4 of Table III we report results

from regressing a binary variable for the presence of seller ﬁnancing on quake risk and the usual

controls. In this regression as well, we ﬁnd an insigniﬁcant eﬀect from quake risk. In unreported

results, we also ﬁnd that other attributes of the loan, such as maturity, ﬂoating/ﬁxed rate status

and the presence of multiple lenders are also not aﬀected by quake risk. The central feature of the

relationship between quake risk and ﬁnancing is the one highlighted in column 1: high quake risk

properties are signiﬁcantly less likely to be ﬁnanced with bank debt. This is consistent with the

implication of Result 2 when insurance markets are not fully competitive.

C. Cross-sectional Eﬀects of Quake risk on ﬁnancing terms

We further examine the role of quake risk on ﬁnancing across various property attributes as a

further test of our model and to help rule out alternative explanations. We ﬁrst consider the

purchase of properties by insurance ﬁrms (either insurers or insurance brokers). There are 102

insurance ﬁrms in our full sample. These insurance ﬁrms are likely to have strong relationships with

catastrophe insurance providers (perhaps within their own company) that should facilitate their

access to insurance. In essence, more catastrophe insurers are likely to bid on a property purchased

by an insurance ﬁrm, so the insurance market faced by insurance ﬁrms will be more competitive.

Result 2 thus indicates that catastrophe risk will have a smaller eﬀect on the likelihood that an

insurance ﬁrm purchases a property with a bank loan, relative to other buyers.

In the ﬁrst column of Table IV, we describe the results from regressing an indicator for the

provision of a bank loan on quake risk, quake risk interacted with an indicator for an insurance

ﬁrm buyer, an indicator for an insurance ﬁrm buyer and the usual controls. The coeﬃcient on

the interaction between quake risk and insurance ﬁrm buyer is positive and signiﬁcant, showing

that quake risk has a smaller eﬀect on the ﬁnancing of properties by insurers. Considering both

the coeﬃcient on quake risk and its interaction with the insurance ﬁrm buyer indicator, we ﬁnd

that quake risk has an insigniﬁcant eﬀect on the probability that an insurance ﬁrm will ﬁnance a

property with a bank loan, in contrast to its strong negative eﬀect for other buyers. This is further

evidence in favor of Result 2.

The eﬀect of the reduction in bank loan provision induced by earthquake risk is not uniform

across properties. Due to innovations in technology, improved building co des and structural deterio-

ration, older properties are likely more susceptible to earthquake damage than younger, seismically-

modernized properties (Schulze et al., 1987 and Otani, 2000).

In the second column of Table IV, we report results from regressing an indicator for the provision

of a bank loan on quake risk, quake risk interacted with property age, property age and the usual

controls. The coeﬃcient on the interaction between quake risk and age is highly signiﬁcant and

negative, showing that quake risk has a stronger eﬀect on the ﬁnancing of older properties.

In the third column of Table IV, we examine the interaction of quake risk with the percentage

of African-Americans in the property’s census tract on loan provision. We ﬁnd a negative and

signiﬁcant coeﬃcient on the interaction term: quake risk reduces the provision of bank loans es-

pecially in neighborhoods with more African-Americans. The regression includes the interaction

between quake risk and the log of median home value in the property’s census tract as a control.

This ﬁnding is consistent with the idea that property insurance is particularly diﬃcult to acquire

in African-American neighborhoo ds, even controlling for local home values (Squires, 2003). The

census tract ﬁxed eﬀects control for any reduction in the general availability of bank credit in

African-American neighborho ods, so this result suggests that distortions in the insurance market

aﬀect the supply of bank loans in these areas. An imperfect supply of catastrophe insurance thus

serves to especially harm areas with greater minority populations.

The fourth column of Table IV reports results from interacting quake risk with a development

indicator variable. We ﬁnd a negative and signiﬁcant coeﬃcient on the interaction, indicating that

the ﬁnancing of development deals is especially hampered by quake risk. While this ﬁnding is not

directly related to our model, it suggests that the ﬁnancing of potentially high NPV development

transactions is discouraged by the presence of quake risk. Overall, the results in Table IV indicate

that quake risk makes it especially diﬃcult to redevelop older buildings in areas with large minority

representation. This evidence suggests that neighborhood revitalization may be hindered by the

improper allocation of quake risk.

It may be argued that for older buildings, the land value is a larger proportion of total property value and land

is presumably less subject to earthquake risk. That may suggest that quake risk is less important for properties with

older buildings. We ﬁnd, however, that land is only infrequently used to secure a bank loan (only 26% of vacant land

is ﬁnanced with a bank mortgage compared to 58% of improved properties). This indicates that it is the structure

that is the primary collateral for bank loans and older structures present greater earthquake damage risks.

C.1. Omitted Variables

We now turn to the hypothesis that some unobserved variable is jointly determining quake risk

and loan provision. Earthquake risk itself is based on soil conditions and relative proximity to

fault lines (which are arguably exogenous), so the endogeneity concern here is one of matching

of types of properties to types of buyers or markets. Are local markets where quake risk is high

more ﬁnancially constrained? Are the types of owners of high quake risk properties diﬀerent

along other unobservable dimensions that might also aﬀect the probability of bank ﬁnancing?

Such unobservable eﬀects may create an omitted variable bias that could erroneously highlight a

signiﬁcant role for quake risk in predicting bank loan provision when no such role exists, or could

generate what appears to be no role for quake risk when in fact such risks are taken into account

by lenders. The sign of the potential bias is ambiguous.

To address the issue of omitted variable bias, we consider the following. First, we employ census

tract ﬁxed eﬀects to diﬀerence out unobservables at a level much ﬁner than the level at which local

ﬁnancial markets operate. In addition, since a census tract is designed to be homogeneous with

respect to population characteristics, economic status, and living conditions (source: United States

Census Bureau), the use of tract ﬁxed eﬀects helps to diﬀerence out other unobservables about the

local population. Local debt market conditions are clearly highly uniform within a census tract

(Kwast, Starr-McCluer, and Wolken, 1997), so the ﬁnancing environment is unlikely to be driving

the micro-level variation we study.

Second, the loans we consider are non-recourse, meaning that the lender may only seek the

collateral value and not any other assets of the borrower in the event of default. The non-recourse

feature of this market implies that borrower quality should be of far less importance than collateral

value.

Third, the bank loan term results also help address omitted variable concerns. For example,

if diﬀerent types of borrowers select properties (within a given census tract) with diﬀerent risk

characteristics (e.g., higher quality borrowers may avoid high quake risk properties), then we should

expect quake risk to not only aﬀect loan provision, but loan terms as well. Any unobservable quality

diﬀerences of borrowers or properties that might be related to quake risk and loan provision, would

almost certainly also be related to the ﬁnancing terms the bank is willing to supply. Omitted

variable problems of this type would predict greater loan provision and better loan terms. We ﬁnd,

The standard deﬁnition of the local banking market in the literature (e.g., Berger, Demsetz, and Strahan, 1999)

is the local Metropolitan Statistical Area (MSA) or non-MSA county.

however, that quake risk is only related to the probability of obtaining a loan, and is unrelated to

any loan term, so unobservable quality diﬀerences are unlikely to explain our ﬁndings.

Fourth, the lack of a signiﬁcant relation between quake risk and seller ﬁnancing also weighs

against an omitted variable explanation relating to unobserved borrower or property quality, which

would presumably be relevant to any lender (bank or seller). A direct role for quake risk is the only

plausible explanation for these diﬀerential eﬀects across lender types.

Last, we also control for the current sale price of the property in an attempt to isolate the

component of quake risk related to liquidation value. Variables related to market value and quake

risk simultaneously should be captured by the sale price and, in fact, may understate the eﬀect of

quake risk on loan provision. Potential omitted variables aﬀecting quake risk and ﬁnancing on a

speciﬁc property within a census tract, property type, and year and controlling for sale price and

other attributes, are diﬃcult to envision.

D. Earthquake risk and Selection of Buyers and Banks

In the model, we argued that the high cost of catastrophe insurance may prevent low wealth

investors from obtaining bank loans and thus may deter them from purchasing high catastrophe

risk insurance. This is formalized in Result 3, which states that in an imp erfectly competitive

catastrophe insurance market, the average wealth of the purchasing investor is increasing in the

catastrophe risk of the property. The COMPS data do not provide the wealth of property buyers,

but they do list the zip code of the purchaser. (The census tract of the buyer is not given.) For

non-corporate (i.e., individual) buyers, the median home value in the buyer’s zip code (from the

2000 census zip code tabulation area data) is a reasonable proxy for the buyer’s wealth. In column

1 of Table V, we display the results from regressing the median home value in the buyer’s zip code

on quake risk and the usual controls, in the subsample for which the buyers are non-corporate. We

ﬁnd a positive and signiﬁcant coeﬃcient on quake risk: within a given census tract, the buyers of

higher quake risk properties tend to come from wealthier zip codes. This is evidence in favor of

Result 3.

In column 2 of Table V we regress the fraction of the issuing bank’s deposits that are held within

the same county as the property on quake risk. We only include observations that include a loan

and for which we can identify the lending bank in this regression. We ﬁnd a marginally signiﬁcant

positive coeﬃcient on quake risk; local banks are more likely to make loans in high quake risk areas.

It is reasonable to suggest that liquidating a building damaged in an earthquake or monitoring

the implementation of safety-enhancing investment may more feasibly be done by a local bank.

While this ﬁnding does not indicate, as the model presumes, that banks are, in general, ineﬃcient

at lending to catastrophe-susceptible properties, it supports the idea that damage assessment and

monitoring (both better performed by closer banks) are important in the ﬁnancing of properties

subject to catastrophic risk, as the theory assumes.

The ﬁnding in column 2 also suggests that the need for diversiﬁcation is unlikely to serve as

an explanation for the importance of quake risk to ﬁnancing, for otherwise the risk would best be

borne by distant banks. As further evidence on this question, in column 3 of Table V we display

results from regressing the log of the assets of the bank making a loan on quake risk, the size of the

loan, and the previous set of controls. As the table indicates, banks making loans in high quake

risk areas are not signiﬁcantly larger in terms of asset size. Larger banks could presumably more

easily diversify their exposure to catastrophe risk. These results therefore do not provide support

to a risk diversiﬁcation explanation.

E. Quake risk and commercial real estate pricing

Result 1 states that property cap rates should increase with quake risk, on at least a one-to-one

basis. We test this hypothesis by regressing cap rates on quake risk and the full set of controls

from Table III. We report the results from this regression in column 1 of Table VI. The results

are inconclusive: the estimated coeﬃcient of 1.26 is consistent with Result 1, but the t-statistic of

0.85 indicates that the null hypothesis of a coeﬃcient of zero cannot be rejected. This test appears

to have too little power to provide evidence either in favor of or opposed to Result 1. Evidence in

other studies supports Result 1: Nakagawa, Saito and Yamaga (2004, 2005) ﬁnd that earthquake

risk reduces rents and land prices in Tokyo.

In a complementary test, we regress the log of sale price on quake risk, the log of earnings and

the controls. As reported in column 2 of Table VI, the coeﬃcient on quake risk is insigniﬁcant in

this speciﬁcation as well. If there is a ﬁxed cost in evaluating the highly localized quake risk that

we are studying, it may that quake risk is only determined for larger properties. Furthermore, the

result in column 2 of Table IV indicate that quake risk is more important for older properties. In

column 3 of Table VI, we display results from regressing the log of sale price on quake risk, the

log of earnings, age, the interaction between quake risk and both the log of earnings and age and

the usual controls. We ﬁnd that the interactions between quake risk and the log of earnings and

property age are both negative and signiﬁcant. This suggests that quake risk has an impact on

property prices primarily for larger and older properties.

IV. The Impact of the Northridge Earthquake

We now turn to the impact of a speciﬁc event, the Northridge, California earthquake of January

17, 1994, on local markets.

In addition to the direct eﬀects of the Northridge quake, press reports

document that earthquake insurance rates across California rose dramatically in the aftermath of

the Northridge earthquake.

. These indirect eﬀects of the earthquake suggest that quake risk may

have distorted both prices and ﬁnancing even in areas that experienced little physical damage.

In particular, the earthquake may have reduced the competitiveness of the catastrophe insurance

market. This shock to the supply of commercial earthquake insurance was, however, relatively

short-lived.

We ﬁrst consider the direct and indirect eﬀects of the earthquake on local cap rates. Earnings

are reported for the previous year, so the eﬀect of the Northridge quake on cap rates largely reﬂects

its eﬀect on prices. We regress cap rates on the peak ground acceleration (PGA), a measure of

quake intensity during the Northridge quake to measure the direct shaking eﬀect of the Northridge

earthquake. To capture the indirect eﬀects of the Northridge quake, we examine the interactions

of quake risk with the log of one plus the numb er of days following the Northridge quake on

which the transaction took place (for transactions within one year of the quake) and an indicator

for the year following the quake, and include the quake risk variable itself, which measures a

property’s susceptibility to future earthquake damage, irrespective of whether it was aﬀected by

the January, 1994 quake. The Northridge earthquake served as a signiﬁcant shock to the supply of

the catastrophe insurance, and our interaction variables measure the extent to which this shock had

a diminishing eﬀect over time. We also include the log of the number of days following the quake

(for transactions within one year of the quake) as an additional control.

The standard controls

from the previous regressions, including census tract ﬁxed eﬀects, are included in the regression.

The ﬁrst column of Table VII reports the results. Neither the PGA nor the interactions are

signiﬁcant. The indirect eﬀect of the earthquake seems to have been small, as quake risk did not

Comerio et al. (1996), Ong et al. (2003), Loukaitou-Sideris and Kamel (2004) discuss the eﬀects of the Northridge

earthquake.

See, for example, Business Insurance, July 4, 1994 and Journal of Commerce, April 19, 1994.

As documented in the accounts in Business Insurance, August 29, 1994 and Journal of Commerce, October 25,

1996.

An indicator for the year following the quake is almost identical to the year indicator for 1994. Including it has

no eﬀect on the regression results.

have a larger eﬀect on property prices in the year after the Northridge quake.

Result 6 states that the probability of bank ﬁnancing increases with insurance market compet-

itiveness. If the Northridge quake decreased the supply of catastrophe insurance, we should see

relatively less bank ﬁnancing for high catastrophe risk properties in its aftermath. We test this

prediction by regressing the probability of bank ﬁnancing on the previous set of variables for the

direct and indirect impact of the quake. We ﬁnd, as detailed in column two of Table VII, that the

interaction of quake risk with the log of one plus the number of days after the quake (for properties

sold within a year of the quake) is positive and signiﬁcant (t-statistic = 2.57). The interaction of

quake risk with an indicator for the year following the quake is negative and signiﬁcant (t-statistic

= −2.31). The second interaction term indicates that shortly after the Northridge earthquake

properties with high earthquake risk were particularly unlikely to be ﬁnanced with bank loans.

The ﬁrst interaction term indicates that this eﬀect moderated with time; as time passed the sup-

plementary post-Northridge negative eﬀect of quake risk on bank ﬁnancing subsided. Considering

the two interactions together, we ﬁnd that the impact of the Northridge quake had essentially

dissipated 3 months after the event. These results are consistent with Result 6: the catastrophic

insurance supply shock generated by the Northridge quake reduced the provision of bank ﬁnance

for high quake risk properties for approximately 3 months. The Northridge quake, however, had

no signiﬁcant long-term eﬀect on the pricing or ﬁnancing of quake risk. Consistent with this result,

Froot (2001) ﬁnds no long-term eﬀect of the Northridge earthquake on catastrophe premiums.

Column 3 of Table VII shows no signiﬁcant increase in the size of banks making loans to high-

quake-risk properties in Los Angeles county after January 1994. However, in column 4 we do ﬁnd

that local banks in Los Angeles county increased their lending to high-risk properties only gradually

in the ﬁrst two months following the Northridge earthquake.

Result 7 states that the average quake risk of bank-ﬁnanced transactions is increasing in insur-

ance market competitiveness, while the average quake risk of all-cash transactions is independent

of insurance market conditions. To test these predictions, we regress the quake risk of all properties

on the log of one plus the number of days following the Northridge quake on which the transaction

took place (for transactions within one year of the quake), the PGA, and the standard controls.

The results are displayed in the ﬁrst column of Table VI II. We ﬁnd a positive and signiﬁcant (t-

statistic = 1.93) coeﬃcient on the log of one plus the number of days following the quake, indicating

Bleich (2003) ﬁnds a 1-2 year eﬀect of the Northridge quake on prices, but he considers the impact of the quake

itself (i.e. PGA, which we control for), not the general quake risk that is our object of study.

that as the catastrophe insurance supply shock diminished, more high quake risk properties were

purchased. This is consistent with the theory, but Result 7 makes an even more precise prediction.

Speciﬁcally, the result states that the quake risk of bank ﬁnanced properties should increase as a

supply sho ck recedes, while the quake risk of all-cash transactions will be unaﬀected. In the second

column of Table VIII, we repeat the previous regression for the subsample of bank ﬁnanced prop-

erties. We ﬁnd that the average quake risk of bank-ﬁnanced transactions increased signiﬁcantly

(t-statistic = 2.61) in the days following the quake. As shown in the third column of Table VIII,

however, there is no eﬀect on the average quake risk of all-cash transactions. These results are

consistent with the predictions of Result 7.

A. Commercial Earthquake Insurance Market after the Northridge Earthquake

We ﬁnd little additional eﬀect of the Northridge earthquake on credit markets after a period of three

months following the event. This ﬁnding is consistent with the rebound in the commercial earth-

quake insurance market after the Northridge quake. The Northridge earthquake had dramatically

diﬀerent impacts on the residential and commercial earthquake insurance markets in California.

In 1994 and 1995 insurers, who faced a state law requiring them to oﬀer earthquake insurance

along with any homeowner’s policy, in many cases elected to stop oﬀering any new homeowner’s

coverage rather than being forced to oﬀer earthquake insurance on terms they deemed unfavorable.

In response, the state government established the California Earthquake Authority as a publicly

managed, largely privately funded organization that provides catastrophic residential earthquake

insurance. No comparable state-managed organization was founded in the commercial earthquake

insurance market.

In the aftermath of the Northridge earthquake, commercial earthquake insurers have increased

their sophistication and made greater use of global reinsurance markets (Risk Management Solu-

tions, 2004). The California Department of Insurance (2003) reports that in the period 1994-1999,

commercial earthquake insurance coverage in the state increased, while residential coverage sharply

decreased. In Los Angeles and Orange counties, for example, commercial coverage grew from a

PML of $11.1 billion in 1994 to $13.6 billion in 1999. Residential coverage fell from $2.8 billion to

$1.1 billion during the same time period. This reasonably quick return to health for the commer-

cial earthquake market is conﬁrmed by the press accounts cited above and is consistent with the

short-term eﬀects of the Northridge earthquake that we ﬁnd in the data.

V. Catastrophic Risk and Finance: Broader Implications

The model in Section I highlights two aspects of the interaction between bank credit and catastrophic

risk insurance markets. First, banks are ineﬃcient ﬁnanciers of catastrophe-susceptible properties

because they are poor at liquidating devastatingly-damaged collateral and they do not specialize in

monitoring the implementation of safety-increasing investments. Second, due to insuﬃcient capital

and market power by a small number of reinsurers, the catastrophe insurance market is ineﬃcient.

These two features, as we will discuss below, are shared by a variety of catastrophic perils including

hurricane, terrorism and political risks. The theory implies that, as a consequence, in markets

aﬀected by any of these hazards increased catastrophic risk will be associated with reduced bank

credit provision, decreased market participation by less wealthy investors and missed investment

and development opportunities.

Our empirical work on earthquake risks broadly supports the predictions of the theory. Earth-

quake data oﬀer two advantages for testing the theory. First, earthquake risks have been quantiﬁed

clearly, which facilitates suitable tests. Second, earthquake risk exhibits signiﬁcant highly localized

variability, which enables the inclusion of census tract ﬁxed eﬀects in the tests, thereby minimiz-

ing the impact of unobservables. Our ﬁndings from the earthquake tests show the relevance of the

theory and therefore support the application of the general theoretical predictions to other settings.

A. Hurricane Risk

Hurricane risk closely resembles that of earthquakes. Hurricanes, like earthquakes, cause terrible

damage to properties that can be mitigated by appropriate preventative investments. The supply

of hurricane insurance also exhibits ineﬃciencies similar to that of earthquake insurance (Froot,

2001). We should therefore expect hurricane risk to have a similar eﬀects to that documented above

for earthquakes.

To expand the scope of our empirical tests, in this section we examine the impact of hurricane

risk on commercial property ﬁnancing. The properties in our sample all face relatively low hurricane

risk (all properties save one are described by AIR as having the same average annual loss from

hurricanes: less than 0.1%).

AIR, however, also provided us with each property’s percentile

hurricane risk within its county and properties within Massachusetts, Maryland, Virginia, Texas,

Georgia, New York and the District of Columbia exhibit variation in this relative hurricane risk.

The Texas properties in the sample are largely located in Dallas and Austin, away from the coasts.

As an additional test of Result 2 relating catastrophe risk to the provision of bank ﬁnancing,

we regress an indicator for whether a property was ﬁnanced with a bank loan on the property’s

percentile hurricane risk within its county and the previous controls, including census tract ﬁxed

eﬀects. The results, as reported in column 1 on Table IX, show that the coeﬃcient on hurricane

risk is insigniﬁcant. Given the generally low level of hurricane peril in our properties, this may

reﬂect a lack of localized variation in hurricane risk within our sample. We thus repeat the previous

regression, replacing the census tract ﬁxed eﬀects with zip code ﬁxed eﬀects. As displayed in column

2 of Table IX, in this speciﬁcation the coeﬃcient on hurricane risk is negative and signiﬁcant (t-

statistic=-2.26), as predicted by Result 2. A one-standard deviation increase in relative hurricane

risk reduces the probability of bank ﬁnancing by 2.3 percentage points.

Results 2 and 6 also link the eﬀects of catastrophe risk on ﬁnancing to the competitiveness of

the insurance market. Hurricane Andrew in 1992 had a dramatic eﬀect on the supply of hurricane

insurance (Froot, 2001), but only two properties in our sample with hurricane risk were sold in

1992. As an alternative measure of supply shocks to the hurricane insurance market, we consider

the Guy Carpenter Catastrophe Insurance Price Index. This variable is limited in two respects.

First, variations in a price index may reﬂect changes in demand rather than supply. Second, the

index is national and conﬂates the prices of multiple types of catastrophe insurance. Nonetheless,

the national price of catastrophe insurance is probably a reasonable proxy for the cost of all types

of catastrophe insurance outside of markets such as the hurricane insurance market in Florida and

the earthquake insurance market in California. Those two markets are excluded from this test, so

we will proceed with the analysis despite the limitations of the variable. We regress an indicator

for whether a property was purchased with a bank loan on hurricane risk, hurricane risk interacted

with the Guy Carpenter Catastrophe Insurance Price Index, the previous controls and zip code

ﬁxed eﬀects. (The Guy Carpenter Catastrophe Insurance Price Index is an annual index, so, given

the use of year ﬁxed eﬀects, we do not include it in the regression.) As described in column 3

of Table IX, the interaction between hurricane risk and the catastrophe insurance price index is

negative and signiﬁcant: hurricane risk reduces the probability of bank ﬁnance particularly in the

years during which catastrophe insurance is especially expensive. This is consistent with Results 2

and 6, and suggests that the theory helps to explain the eﬀects of hurricane risk on credit markets.

Recent research (Trenberth et al., 2007) proposes that there has been an increase in very severe

tropical cyclones, hurricanes, typhoons and other weather events that may be linked to human-

generated changes in the environment. As a result, understanding the role of optimal hurricane

risk allocation is increasingly relevant.

B. Terrorism Risk

Terrorism risk shares the central characteristics of other catastrophic risks: enormous potential

damages, specialized safety investments and ineﬃcient supply of insurance. After the Sept. 11, 2001

terrorist attacks on the U.S., insurers and reinsurers quickly began to exclude terrorism risk from

their policies (Hubbard, Deal and Hess, 2005). Consistent with the predictions of the theoretical

model, commercial lending and development in major cities were signiﬁcantly reduced (Serio, 2004).

Terrorism risk is more diﬃcult to optimally allocate than natural disaster risk for three reasons.

First, terrorism risk is much harder to quantify. This ambiguity may lead to reduced provision of

insurance (Hogarth and Kunreuther, 1985). Second, terrorism is endogenously determined by the

actions of human agents. The optimal allocation of risk usually involves redistribution and trading,

but in the case of terrorism risks this may lead to market manipulation (Poteshman, 2006). Third,

the scale of damages from an act of terror is potentially greater than from a natural disaster.

The U.S. government passed the Terrorism Risk Insurance Act (TRIA) in 2002 to address the

dislocations in the terrorism insurance market. TRIA requires insurers to oﬀer terrorism coverage

and institutes federal cost-sharing for terrorism damages. TRIA’s proponents argue that it has cre-

ated price stability in the terrorism risk insurance market and led to increased use of the insurance

(Hubbard, Deal and Hess, 2005). TRIA was renewed in 2005 and is up for renewal in 2007. The

long-term ineﬃciencies in the supply of natural disaster catastrophic insurance and the arguments

given above for why terrorism is even more diﬃcult to insure than earthquakes or hurricanes in-

dicate that the private market is unlikely to successfully supply suﬃcient terrorism insurance on

its own. The ﬁndings in this paper therefore suggest two implications of a failure to renew TRIA.

First, high proﬁle properties and those located in high density areas (possible terrorism targets)

are likely to suﬀer from reduced bank credit provision and market exit by less wealthy investors.

Downtown areas in large U.S. cities would be predicted to exhibit a substantial shift away from

debt ﬁnancing of properties and prices for larger properties may signiﬁcantly decline. Second, rede-

velopment after an act of terrorism will hampered by a particularly restricted supply of bank credit

in the immediate post-event period. These eﬀects will especially severe in minority neighborhoods.

C. Political Risk

The p olitical risk associated with an investment describes the probability of nationalization, cur-

rency controls or other government action that reduces the project value. This risk, which is

typically most salient in emerging markets, has the p otential to generate huge losses and the risk

assessment is uncertain, so the supply of political risk insurance, like that of other catastrophic

risks, is quite restricted (Hamdani et al., 2005). The providers of political risk insurance (particu-

larly the public agencies such as the Multilateral Investment Guarantee Agency of the World Bank

and the Overseas Private Development Corporation of the U.S. government) are typically skilled

at negotiating with the local authorities. In the terms of our model, these insurers are better at

liquidating investments adversely aﬀected by catastrophic political risk.

Our model and empirical work indicate that p olitical risk insurance can serve two important

functions. First, our results tying investment and market participation to the eﬃciency of insurance

markets imply that more ﬁnancially constrained and smaller ﬁrms will be signiﬁcantly more likely

to invest in developing foreign countries if fairly priced political risk insurance is available. Second,

there is growing interest in emerging market debt (Erb, Harvey, Viskanta, 2000). Our ﬁndings

linking loan provision to the availability of catastrophic insurance suggest that investors will be more

willing to provide debt ﬁnancing to at-risk ﬁrms when it is accompanied by political insurance.

This implies that a well-functioning political risk insurance market will be especially important in

promoting the issuance of emerging market corporate bonds.

VI. Conclusion

We present a model in which banks are ineﬃcient in ﬁnancing properties facing catastrophic risk

because they are poor at liquidating catastrophically-damaged properties and they do not special-

ize in monitoring the implementation of safety-improving investments. These functions are best

performed by insurers, but imperfections in the supply of catastrophe insurance can distort real

estate markets by limiting the provision of bank credit and preventing positive NPV projects from

being undertaken.

An empirical analysis of the eﬀects of earthquake risk provides evidence in support of the

theory. We ﬁnd that earthquake-prone properties are much less likely to be ﬁnanced with bank

As our model makes clear, the ineﬃcient liquidation problem faced by lenders may make them unwilling to supply

debt in the absence of insurance, irrespective of the promised interest rate. Alternatively, the lack of insurance may

increase the necessary promised rate to a level that borrowers regard as unacceptable.

debt, and we provide evidence that older properties, properties in African-American neighborhoods

and properties slated for development are all especially unlikely to receive bank ﬁnancing in the

presence of earthquake risk. The ﬁnancing of property purchases by insurance ﬁrms, that likely have

privileged access to earthquake coverage, is unaﬀected by earthquake risk. Non-corporate buyers

of high risk properties tend to come from wealthier areas. This evidence suggests that ineﬃciencies

in the catastrophe insurance market are reducing the provision of bank credit, limiting the market

participation of less wealthy investors and hampering neighborhood revitalization in disadvantaged

areas.

The 1994 Northridge earthquake led to a reduction in lending to high-risk properties, but this

eﬀect persisted for only three months. We ﬁnd no signiﬁcant longer-term ﬁnancing or pricing eﬀects

arising from the Northridge earthquake. Thus the large shock to the supply of earthquake insurance

arising from the earthquake exacerbated the degree of ineﬃciency in the ﬁnancing of earthquake

risk for only a short period of time.

Our model presents a framework for analyzing broad classes of catastrophic risks including

hurricane, terrorism and political perils, and our empirical work suggests implications for all these

markets. We show that hurricane risk, like that of earthquakes, reduces bank ﬁnancing. We argue

that terrorism and political risk share the central features of natural disaster risk and may be even

more subject to an insuﬃcient supply of insurance. Our ﬁndings suggest that a lack of terrorism

insurance is likely to cause a shift away from debt ﬁnancing in downtown U.S. cities and to lead to

the exit from these areas of less wealthy investors. A restricted supply of political risk insurance

will discourage foreign investments by ﬁnancially-constrained ﬁrms and depress emerging market

corporate bond issuance. Exposure to catastrophic risks, both natural and unnatural, continues

to grow due to population shifts to at-risk areas, global warming and changing political dynamics.

Continued ineﬃciencies in the sharing of catastrophic risks and their eﬀects on broader capital

markets may have implications for long-term growth in a wide variety of countries.

REFERENCES

1. Anderson, Dan R., and Maurice Weinrobe, 1986, Mortgage Default Risks and the 1971 San

Fernando Earthquake, AREUA Journal, 14, 110-135.

2. Benmelech, Efraim, Mark Garmaise, and Tobias Moskowitz, 2005, Do liquidation values

aﬀect ﬁnancial contracts? Evidence from commercial loan contracts and zoning regulation,

Quarterly Journal of Economics, August, 2005, 1121-1153.

3. Bethoux, Nicole, Guy Ouillon and Marc Nicolas, 1998, The instrumental seismicity of the

Western Alps: spatio-temporal patterns analysed with the wavelet transform, Geophysical

Journal Internation, 135, 177-194.

4. Bleich, Donald, 2003, The reaction of multifamily capitalization rates to natural disasters,

The Journal of Real Estate Research, 25, 133-144.

5. Bolton, Patrick, and David Scharfstein, 1996, Optimal Debt Structure and the Number of

Creditors, Journal of Political Economy, 104, 1-26.

6. Burgess, Robin, and Rohini Pande, 2003, Do Rural Banks Matter? Evidence from the Indian

Social Banking Experiment, Working paper, London School of Economics.

7. California Department of Insurance, 2003, California Earthquake Zoning and Probable Max-

imum Loss Evaluation Program.

8. Cetorelli, Nicola, and Michele Gambera, 2001, Banking Market Structure, Financial De-

pendence and Growth: International Evidence from Industry Data, Journal of Finance 56,

617-648.

9. Chen, Jun, and Yongheng Deng, 2003, Commercial Mortgage Workout Strategy and Con-

ditional Default Probability: Evidence from Special Serviced CMBS Loans, Working Paper,

University of Southern California.

10. Ciochetti, Brian A., 1997, Loss Characteristics of Commercial Mortgage Foreclosures, Real

Estate Finance 97, 53-69.

11. Comerio Mary C., John D. Landis, Catherine J. Firpo and Juan Pablo Monzon, 1996, Res-

idential Earthquake Recovery: Improving Californias Post-Disaster Rebuilding Policies and

Programs, Working Paper, California Policy Seminar Brief Series.

12. Cotterman, Robert F., and James E. Pearce, 1996, The Eﬀects of the Federal National

Mortgage Association and the Federal Home Loan Mortgage Corporation on Conventional

Fixed-Rate Mortgage Yields, in: U.S. Department of Housing and Urban Development, ed.,

Studies on Privatizing Fannie Mae and Freddie Mac, U.S. Department of Housing and Urban

Development. Washington, D.C.: U.S. Department of Housing and Urban Development.

13. Cummins, J. David, David Lalonde and Richard D. Phillips, 2004, The basis risk of catastrophic-

loss index securities, Journal of Financial Economics, 71, 77-111.

14. DiPasquale, D., and E. L. Glaeser, 1999, Incentives and Social Capital: Are Homeowners

Better Citizens?, Journal of Urban Economics 45, 35484.

15. Diamond, Douglas, 2004, Presidential Address, Committing to Commit: Short-term Debt

When Enforcement is Costly, Journal of Finance 59, 1447-1479.

16. Duﬃe, Darrell, and Kenneth J. Singleton, 1999, Modeling Term Structures of Defaultable

Bonds, Review of Financial Studies, 687-720.

17. Erb, Claude B., Campbell R. Harvey and Tadas E. Viskanta, 2000, Understanding Emerging

Market Bonds, Emerging Markets Quarterly 4, 1, 7-23.

18. Froot, Kenneth A., 2001, The market for catastrophe risk: a clinical examination, Journal of

Financial Economics, 60, 529-571.

19. Froot, Kenneth A., and Paul G.J. O’Connell, 1997. On the pricing of intermediated risks:

theory, and application to catastrophe reinsurance. NBER Working paper no. 6011.

20. Fudenberg, Drew, and Jean Tirole, 1993, Game Theory, MIT Press, Cambridge.

21. Garmaise, Mark J., and Tobias J. Moskowitz, 2003, “Informal Financial Networks: Theory

and Evidence,” Review of Financial Studies, 16, 1007-1040.

22. Garmaise, Mark J., and Tobias J. Moskowitz, 2004, “Confronting Information Asymmetries:

Evidence from Real Estate Markets,” Review of Financial Studies, 17, 405-437.

23. Garmaise, Mark J., and Tobias J. Moskowitz, 2005, Bank Mergers and Crime: The Real and

Social Eﬀects of Credit Market Competition, Journal of Finance, forthcoming.

24. Glaeser, Edward L., and Hedi D. Kallal, 1997, Thin Markets, Asymmetric Information, and

Mortgage-Backed Securities, Journal of Financial Intermediation, 6, 64-86.

25. Glickman, Andrea, and Joshua Stein, 2005, Best practices in commercial real estate ﬁnancing:

Insurance and commercial mortgages in the new world of risk, Brieﬁngs in Real Estate Finance

4, 309-317.

26. Hall, Brian J., and Kevin J. Murphy, 2000, Stock Options for Undiversiﬁed Executives, NBER

Working paper.

27. Hamdani, Kausar, 2005, Elise Liebers and George Zanjani, An Overview of Political Risk

Insurance, Federal Reserve Bank of New York Working Paper.

28. Hogath, Robin, and Howard Kunreuther, 1985, Ambiguity and Insurance Decisions, American

Economic Review 75, 386-390.

29. Hubbard, Glenn, Bruce Deal and Peter Hess, 2005, The Economic Eﬀects of Federal Partici-

pation in Terrorism Risk, Risk Management and Insurance Review 8, 177-209.

30. Jaﬀee, Dwight M., and Thomas Russell, 1997, Catastrophe insurance, capital markets, and

uninsurable risks, Journal of Risk and Insurance 64, 205-230.

31. Loukaitou-Sideris, Anastasia, and Nabil M. Kamel, 2004, Residential Recovery from the

Northridge Earthquake: An Evaluation of Federal Assistance Programs, University of Cali-

fornia Working paper.

32. Kagan, Y. Y., 1997, Seismic moment-frequency relation for shallow earthquakes: Regional

comparison, Journal of Geophysical Research 102, 28352852.

33. Kane, Edward J., 1999, Housing Finance GSEs: Who Gets the Subsidy?, Journal of Financial

Services Research, 15, 197-209.

34. Klein, Michael, Joe Peek and Eric Rosengren, 2002, Troubled Banks, Impaired Foreign In-

vestment: The Role of Relative Access to Credit, American Economic Review 92, 664-682.

35. Kwast, Myron L., Martha Starr-McCluer, and John D. Wolken, 1997, Market Deﬁnition and

the Analysis of Antitrust in Banking, Antitrust Bulletin 42, 973-95.

36. MacDonald, Gordon J., 2000, A Note on Order Statistics and Property Losses from Catastrophic

Hurricanes and Floods in the USA, Interim Report 00-047, International Institute for Applied

Systems Analysis.

37. Munnell, Alicia H., Geoﬀrey M. B. Tootell, Lynn E. Browne and James McEneaney, 1996,

Mortgage Lending in Boston: Interpreting HMDA Data, American Economic Review, 86,

25-53.

38. Nakagawa, Masayuki, Makoto Saito and Hisaki Yamaga, 2004, Earthquake Risks and Land

Prices: Evidence from the Tokyo Metropolitan Area, Nihon University Working paper.

39. Nakagawa, Masayuki, Makoto Saito and Hisaki Yamaga, 2005, Earthquake Risk and Housing

Rents: Evidence from the Tokyo Metropolitan Area, Univesity of Tsukuba Working paper.

40. Niehaus, Greg, 2002, The allocation of catastrophe risk, Journal of Banking and Finance, 26,

585-596.

41. Ong, Paul, James Spencer, Michela Zonta, Todd Nelson, Douglas Miller and Julia Heintz-

Mackoﬀ, 2003, The Economic Cycle and Los Angeles Neighborhoods; 1987-2001, University

of California Working paper.

42. Otani, Shunsuke, 2000, Seismic Vulnerability Assessment of Reinforced Concrete Buildings,

The Journal of the Faculty of Engineering, University of Tokyo, Series B 47, 5 - 28.

43. Peek, Joe, and Eric Rosengren, 2000, Collateral Damage: Eﬀects of the Japanese Bank Crisis

on Real Activity in the United States, American Economic Review 90, 30-45.

44. Petak, William J., and Shirin Elahi, 2000, The Northridge Earthquake, USA and Its Economic

and Social Impacts, Euroconference on Global Change and Catastrophe Risk Management:

Earthquake Risks in Europe, International Institute for Applied Systems Analysis (IIASA),

128.

45. Porter, Keith A., James L. Beck, Rustem V. Shaikhutdinov, Siu Kui Au, Kaoru Mizukoshi,

Masamitsu Miyamura, Hiroshi Ishida, Takafumi Moroi, Yasu Tsukada and Manabu Masudai,

2004, Earthquake Spectra 20, 12111237.

46. Poteshman, Allen M., 2006, Unusual Option Market Activity and the Terrorist Attacks of

September 11, 2001, Journal of Business 79, 1703-1726.

47. Risk Management Solutions, 2004, The Northridge, California Earthquake: RMS 10-Year

Retrospective.

48. Rohe, William M., Shannon Van Zandt and George McCarthy, 2000, The Social Beneﬁts and

Costs of Homeownership: a Critical Assessment of the Research, Working paper, Research

Institute for Housing America.

49. Roll, Richard, 2003, Beneﬁts to Homeowners from Mortgage Portfolios Retained by Fannie

Mae and Freddie Mac, Journal of Financial Services Research, 23, 29-42.

50. Ross, Stephen L., and Geoﬀrey M.B. Tootell, 2004, Redlining, the Community Reinvestment

Act, and Private Mortgage Insurance, Journal of Urban Economics, 55, 278-297.

51. Roth, Richard J., 1998, Earthquake insurance in California, in: Howard Kunreuther and

Richard J. Roth, eds., Paying the Price: The Status and Role of Insurance Against Natural

Disasters in the United States. Washington, D.C.: Joseph Henry Press.

52. R¨uttener, Erik, Dany Liechti and Stephan Eugster, 1999, The Risk Premium Distribution

(Average Annual Loss) with Respect to Earthquake Magnitude, XXIV European Geophysical

Society General Assembly.

53. Schulze, William D., David S. Brookshire, Ronda K. Hageman and John Tschirhart, 1987,

Beneﬁts and Costs of Earthquake Resistant Buildings, Southern Economic Journal, 53, 934-

951.

54. Serio, Gregory, 2004, Statement of New York State Insurance Department before U.S. House

of Representatives Subcommittee on Capital Markets, Insurnace, and Government Sponsored

Enterprises and the Subcommittee on Oversight and Investigations.

55. Shah, Haresh, and Daniel Rosenbaum, 1996, Earthquake Risk Shakes Mortgage Industry,

Secondary Mortgage Markets 13, 12-19.

56. Smith, Cliﬀord, and Jerold Warner, 1979, On Financial Contracting - An Analysis of Bond

Covenants, Journal of Financial Economics 7, 117-161.

57. Squires, Gregory D., Sally O’Connor and Josh Silver, 2001, The Unavailability of Information

on Insurance Unavailability: Insurance Redlining and the Absence of Geocoded Disclosure

Data, Housing Policy Debate, 12, 347-372.

58. Squires, Gregory D., 2003, Racial Proﬁling, Insurance Style: Insurance Redlining and the

Uneven Development of Metropolitan Areas, Journal of Urban Aﬀairs, 25, 391-410.

59. Standand and Poor’s, 2003, U.S. CMBS Legal and Structured Finance Criteria.

60. Thomas, Jerry L., 1997, Testimony before the Subcommiteee on Housing and Community

Oppurtunity, Committee on Banking and Financial Services, U.S. House of Representatives.

61. Trenberth, K.E., P.D. Jones, P. Ambenje, R. Bojariu, D. Easterling, A. Klein Tank, D.

Parker, F. Rahimzadeh, J.A. Renwick, M. Rusticucci, B. Soden and P. Zhai, 2007, Observa-

tions: Surface and Atmospheric Climate Change. In: Climate Change 2007: The Physical

Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the

Intergovernmental Panel on Climate Change [Solomon, S., D. Qin, M. Manning, Z. Chen,

M. Marquis, K.B. Averyt, M. Tignor and H.L. Miller (eds.)]. Cambridge University Press,

Cambridge, United Kingdom and New York, NY, USA.

62. Vandell, Kerry D., 1998, Strategic Management of the Apartment Business in a “Big REIT”

World, National Multi Housing Council Working paper.

63. Zanjani, George, 2002, Pricing and capital allocation in catastrophe insurance, Journal of

Financial Economics, 65, 283-305.

Appendix

We propose (and later conﬁrm) that V

= kC

for some constant k. Under this assumption V

has

the same risk as C

and may also be discounted at the rate r. The value of the property at any

given time s must satisfy:

= p

E[C

s+1

+ V

s+1

]

1 + r

+ (1 − p)

E [R

s+1

]

1 + r

. (3)

The equations V

= kC

and E[C

x+1

] = C

for all x then imply that

= p

(1 + k)C

1 + r

+ (1 − p)

E[R]kC

1 + r

which simpliﬁes to

k(1 + r − p − (1 − p)E[R]) = p. (4)

This equation shows that k is a constant, so the valuation equation (3) is indeed satisﬁed by our

proposed solution V

= kC

Proof of Result 1:

The cap rate is

r+q

. We calculate

− (r + q)

≥ 1.

Model of Ineﬃcient Liquidation by Banks of Catastrophe-damaged properties

We presume that banks do not observe R

and they are incapable of undertaking repairs on

their own. Consequently, if banks seize a property in default that is damaged in a catastrophe,

a fraction δ ∈ (0, 1) of its value is destroyed in liquidation. For example, we may assume that

there is limited liquidity for damaged properties (particularly after a catastrophe) and banks must

hold these impaired (and non-cash-ﬂow-producing) properties for several periods before they can

be sold.

We assume that when a bank provides a mortgage in perio d s with promised repayment m

in the following period, it receives min{m, C

s+1

+ V

s+1

} if there is no catastrophe. That is, we

assume that loans are non-recourse (as is the case for our sample of commercial real estate loans),

and for simplicity, we presume that there are no liquidation costs for the bank in the absence of

a catastrophe; the bank can simply sell a defaulted property on the open market. In the event of

a catastrophe, the bank does not observe the current value of the property, because it does not

observe R

. We assume that the bank makes a take-it-or-leave-it oﬀer to the property owner,

demanding m

≤ m in repayment. If m

< m, the bank negotiates terms by forgiving part of its

loan.

We presume that the banking sector is competitive; our focus is on credit market distortions

that are related to catastrophic risk, rather than general credit market distortions.

The property owner does observe R

s+1

. If the property owner remains in possession of the

property, he can repair it at a cost of (1 − R

s+1

and then sell the property for V

s+1

The

property owner will only accept the bank’s oﬀer if his net proceeds after repairing the property

exceed m

. (If a portion of the repair costs are eﬀort costs borne by the owner, then the owner

will strictly prefer to reject bank oﬀers that are too high.) That is, the bank’s oﬀer will only be

accepted if

Chen and Deng (2003) show that loan workouts with modiﬁed loan terms are common in commercial real estate.

The repair can be ﬁnanced by a bank that makes a top priority loan that is veriﬁably invested in the property.

Since the repair is positive net present value, the bank will be repaid.

s+1

≥ m

If the owner rejects the bank’s oﬀer, then the bank receives δV

s+1

. We set f

and F

be the probability density function (pdf) and cumulative distribution function (cdf) of the {R

respectively. The bank chooses m

to maximize

max

≤m

1 − F

s+1

¶¶

+ (1 − δ)

s+1

dr (5)

Equivalently, the bank solves

max

x≤

s+1

x (1 − F

(x)) + (1 − δ)

(r)rdr

(6)

We will presume that the {R

} have support [0, 1] and satisfy the monotone-hazard-rate property

(i.e.,

(x)

1−F

(x)

≥ 0) to guarantee that there is a unique interior solution to (6). The monotone-

hazard-rate (MHR) condition is satisﬁed by the uniform, truncated normal, truncated exponential

and many other distribution functions (Fudenberg and Tirole, p.267).

We deﬁne L(m, V, p) to be the current market value of a one-period mortgage with promised

repayment m secured against a prop erty with value V and catastrophe frequency (1 − p). We

ﬁrst show that catastrophe risk reduces the value of a loan. As the frequency of the catastrophe

increases, conditions A and B guarantee that the property’s payoﬀ next period will be more risky.

This risk is, of course, incorporated in the current property value, for a given value today of V , an

increase in catastrophe risk leads to a higher payoﬀ if no catastrophe occurs, but a lower payoﬀ in

the event of a catastrophe. The net impact is to reduce the value of a given mortgage claim.

Lemma 1. The market value of a mortgage with a given promised repayment is decreasing in

the underlying asset’s catastrophe risk. Formally, L(m, V, p) is increasing in p.

Proof of Lemma 1. For a given current (i.e., period s ) property market value V

, we denote

by ² the risk-neutral probability associated with next period’s market value V

s+1

, and we deﬁne r

to be the risk-free rate. We set d(m, V

s+1

, p) to be equal to the payoﬀ received by the debtholder

in perio d s + 1 when the promised repayment is m and the property value is V

s+1

. We will show

that for any outcome of V

s+1

, d(m, V

s+1

, p) is increasing in p. We let p

≥ p

be given. We have

(p) = V

r + (1 − p)(1 − E[R])

The function C

is decreasing in p. We ﬁrst suppose that m ≥ V

s+1

) ≥ V

s+1

The derivative of the objective function in (6) with respect to x is

s+1

(1 − F

(x) − δf

(x)x) .

The MHR property guarantees that there is a unique value x

∗

∈ (0, 1) that maximizes the

unconstrained objective function in (6). Since m ≥ V

s+1

, for both p

and p

, in the event of a

catastrophe the bank will solve the unconstrained problem and set x = x

∗

. We deﬁne

λ =

∗

(1 − F

∗

)) + (1 − δ)

∗

(r)rdr

E[R]

∈ (0, 1).

In the non-catastrophe states, the bank will receive the full cash ﬂows and property value. We

thus have for i = 1, 2

d(m, V

s+1

, p

) = p

s+1

+ C

s+1

)) + (1 − p)V

s+1

λE[R]

= V

s+1

(r + q(p

) + p

+ (1 − p

)λE[R])

= V

s+1

(r + 1 − (1 − p

)E[R](1 − λ)) .

It is clear that d(m, V

s+1

, p

) ≥ d(m, V

s+1

, p

). For m ∈ [V

s+1

+ C

s+1

), V

s+1

+ C

s+1

)],

d(m, V

s+1

, p

) is constant, while d(m, V

s+1

, p

) = p

m + (1 − p)V

s+1

λE[R] is increasing, so the

result just shown also demonstrates that d(m, V

s+1

, p

) ≥ d(m, V

s+1

, p

) on this range. Last, we

consider m < V

s+1

+ C

s+1

). If there is no catastrophe, the lender receives m under both p

and

. Now suppose there is a catastrophe. We have

s+1

≤ F

s+1

so V

s+1

dominates V

s+1

in the sense of ﬁrst-order stochastic dominance. For each

choice of m

, the objective function in (5) is the expectation of an increasing function of V

s+1

The ﬁrst-order stochastic dominance relationship shows that for any choice m, the payoﬀ to the

lender under p

exceeds that under p

. The payoﬀ to the lender under the optimal choice of m

must therefore be higher under p

. We denote the payoﬀ to the lender under the optimal choice by

y(p

) for i ∈ {1, 2}. We have y(p

) ≤ y(p

) ≤ m. We conclude that

m(p

− p

) ≥ y(p

)(p

− p

) ≥ (1 − p

)y(p

) − (1 − p

)y(p

)

⇒ p

m + (1 − p

)y(p

) ≥ p

m + (1 − p

)y(p

) ⇒ d(m, V

s+1

, p

) ≥ d(m, V

s+1

, p

We therefore have d(m, V

s+1

, p

) ≥ d(m, V

s+1

, p

) for all V

s+1

, so

L(m, V

, p

) =

d(m, V

s+1

, p

)²(V

s+1

)dV

s+1

1 + r

≥

d(m, V

s+1

, p

)²(V

s+1

)dV

s+1

1 + r

= L(m, V

, p

We deﬁne the debt capacity DC of a property to be the maximum loan value that can be

secured by the property. That is, DC(V

, p) = L(∞, V

, p). We now consider the payoﬀ of the

prospective buyer with a private beneﬁt. The investor will receive his private beneﬁt and an equity

claim on the property in exchange for his cash investment. We deﬁne Π

(w, V

, p) to be the net

present value received by this investor if he invests w ≤ V

in cash and borrows V

− w in order to

purchase a property with catastrophe frequency (1 − p).

For w ≤ V

, the investor with a private beneﬁt requires a loan with value dval = V

− w. The

value of the investor’s equity claim is equal to the value V

of the property minus the value of

the bank’s debt claim minus any value destroyed in ineﬃcient liquidation. If DC(V

, p) + w ≥ V

the debt ﬁnancing is possible, and we denote the value of ineﬃcient liquidation by loss(w, V

, p).

For notational convenience we set loss(w, V

, p) = ∞ when debt ﬁnancing is not possible. Credit

markets are competitive, so the bank’s debt claim has a value of dval. When debt ﬁnancing is

feasible, the net present value of the investor’s investment is therefore equal to his private beneﬁt

minus the value destroyed in ineﬃcient liquidation. Since not investing is always an option, we

have

(w, V

, p) = max{b + (V

− L(m, V

, p) − loss(w, V

, p)) − w, 0},

where m is such that L(m, V

, p) = V

− w. Thus

(w, V

, p) = max{b − loss(w, V

, p), 0}.

Lemma 2. Debt capacity decreases with catastrophe risk: DC(V

, p) is increasing in p. The

prospective buyer’s net present value is decreasing in catastrophe risk and increasing in his cash

investment. Formally, Π

(w, V

, p) is increasing in p and w.

Proof of Lemma 2. We have m = ∞, so m ≥ V

s+1

. Following the proof of Lemma 1, for all

s+1

d(∞, V

s+1

, p) = V

s+1

(r + 1 − (1 − p)E[R](1 − λ)) ,

which is increasing in p. This shows that debt capacity is increasing in p.

For w ≤ V

, the investor with a private beneﬁt requires a loan with value k = V

− w. We deﬁne

m(w, V

, p) to be the m such that L(m, V

, p) = V

−w, if it exists. Our result on debt capacity shows

that as p increases it is more likely that the project can be ﬁnanced. Thus, Π

(w, V

, p) = 0 for p

and w so low that DC(V

, p)+w < V

. Consider p and w such that the project can be ﬁnanced. It is

clear from (5) that L(m, V

, p) is increasing in m and Lemma 1 shows that L(m, V

, p) is increasing

in p. This implies that m(w, V

, p) is decreasing in p and w.

We now consider the optimal loan forgiveness strategy of the bank. We denote by ˆx(w, V

, V

s+1

, p)

the value of x that solves the problem (6) for a given V

s+1

,p and m = m(w, V

, p). We have that

m(w,V

,p)

s+1

is decreasing in both p and w. That shows that ˆx(w, V

, V

s+1

, p) is decreasing in p and w.

loss(w, V

, p) =

δ(1 − p)

s+1

ˆx(w,V

s+1

,p)

(r)rdr²(V

s+1

)dV

s+1

1 + r

The fact that ˆx(w, V

, V

s+1

, p) is decreasing in p and w shows that loss(w, V

, p) is decreasing

in p and w (Property 1).

For debt with any face value m > 0 and p < 1, the liquidation costs may be written as

δ(1 − p)

R R

min{

s+1

∗

}

(r)rdrV

s+1

²(V

s+1

)dV

s+1

1 + r

, (7)

which is strictly positive since f

has support [0, 1] (Property 2).

Proof of Result 2. If the investor is unconstrained, the transaction is consummated in all

cases in an all-cash deal. If the investor is constrained, he will only purchase the property with a

bank loan. There are three possibilities. With probability (1−n)

, no insurers bid and the investor

purchases the property only if b ≥ loss(w, V

, p). In this case the property is purchased without

insurance. With probability 2n(1−n) one insurer bids, the bid is sur = min{b

∗

, w

, loss(w

, V

, p)},

and the investor purchases the property only if b ≥ min{b

∗

, w

, loss(w

, V

, p)}. In this case the

property is purchased with insurance. With probability n

, two insurers bid, sur = 0 and the

investor always purchases the property with insurance. Thus probability that a transaction is

ﬁnanced with bank debt is given by

(1 − l)

(1 − n)

(1 − F

(loss(w, V

, p))) + 2n(1 − n)(1 − F

(min{b

∗

, w

, loss(w

, V

, p)})) + n

(1 − l) ((1 − n)

(1 − F

(loss(w, V

, p))) + 2n(1 − n)(1 − F

(min{b

∗

, w

, loss(w

, V

, p)})) + n

) + l

(8)

The result follows Property 1 that loss(w

, V

, p) is decreasing in p.

Proof of Result 3.

The average wealth of the purchasing investor is given by

(1 − l)

(1 − n)

(1 − F

(loss(w, V

, p))) + 2n(1 − n)(1 − F

(min{b

∗

, w

, loss(w

, V

, p)})) + n

+ w

(1 − l) ((1 − n)

(1 − F

(loss(w, V

, p))) + 2n(1 − n)(1 − F

(min{b

∗

, w

, loss(w

, V

, p)})) + n

) + l

(9)

The result follows Property 1.

Proof of Result 4. If the insurance market is perfectly competitive, insurance is provided at

fair value, and purchasing insurance allows the constrained investor to avoid all liquidation costs.

For debt with any face value m > 0, if (1 − p) > 0 Property 2 shows that these liquidation costs

are strictly positive, so all constrained investors will purchase insurance. If the insurance market

is imperfectly competitive, the proof of Result 2 showed that the probability that a bank-ﬁnanced

purchase carries insurance is given by

2n(1 − n)(1 − F

(min{b

∗

, w

, loss(w

, V

, p)})) + n

((1 − n)

(1 − F

(loss(w, V

, p))) + 2n(1 − n)(1 − F

(min{b

∗

, w

, loss(w

, V

, p)})) + n

)

. (10)

The result follows from the fact that loss(w

, V

, p) is decreasing in p.

Proof of Result 5. The probability of a transaction is given by

(1−l)

(1 − n)

(1 − F

(loss(w, V

, p))) + 2n(1 − n)(1 − F

(min{b

∗

, w

, loss(w

, V

, p)})) + n

+l.

The result follows from the fact that loss(w

, V

, p) is decreasing in p.

We now assess the impact on a change in the competitiveness of the insurance market in the

composition of properties sold in the market. We suppose that p is a random variable with cdf F

and pdf f

Proof of Result 6. The result follows from (8) and from the fact that for 1 ≥ b ≥ a,

(a(1 − n)

+ 2bn(1 − n) + n

) = 2(n − nb + b(1 − n) − a(1 − n)) ≥ 0.

Proof of Result 7. Unconstrained investors pay all cash and do not suﬀer liquidation costs,

so the average catastrophic risk of all-cash transactions is E[1 − p]. Constrained investors require

bank ﬁnancing. The pdf h

(p, n) of bank-ﬁnanced sales for a given n is

(p, n)

(p)

(1 − n)

(1 − F

(loss(w

, V

, p))) + 2n(1 − n)(1 − F

(min{b

∗

, w

, loss(w

, V

, p)})) + n

(x) ((1 − n)

(1 − F

(loss(w

, V

, p))) + 2n(1 − n)(1 − F

(min{b

∗

, w

, loss(w

, V

, p)})) + n

) dx

We will show that for n

≥ n

, h

(p, n

) dominates h

(p, n

) in the sense of the monotone like-

lihood ratio property (MLRP). That is suﬃcient to show

(1 − p)h

(p, n

)dp ≥

(1 − p)h

(p, n

)dp.

We require that

r(p) =

(1 − n

)

(1 − F

(loss(w

, V

, p))) + 2n

(1 − n

)(1 − F

(min{b

∗

, w

, loss(w

, V

, p)})) + n

(1 − n

)

(1 − F

(loss(w

, V

, p))) + 2n

(1 − n

)(1 − F

(min{b

∗

, w

, loss(w

, V

, p)})) + n

be decreasing in p. We ﬁrst note that

(p) =

(p) + B

(p) + D

(11)

is decreasing in p if j

(p) is increasing in p and

≥

. For p such that loss(w

, V

, p) ≤

min{b

∗

, w

}, r(p) =

(p)+B

(p)+D

, where j

(p) = 1−F

(loss(w

, V

, p)) is increasing in p and

≥

1 ≥

1−n

. For p such that loss(w

, V

, p) ≥ min{b

∗

, w

} but such that loss(w

, V

, p) < ∞

(i.e. such that loss(w

, V

, p) ≤ w

), r(p) =

(p)+B

(p)+D

, where j

(p) = 1 − F

(loss(w

, V

, p)) and

+2n

(1−n

)(1−F

(min{b

∗

}))

+2n

(1−n

)(1−F

(min{b

∗

}))

≥ 1 ≥

(1−n

)

(1−n

)

. The function r is discontinuous at p

∗

such

that loss(w

, V

, p

∗

) = w

. We note, however, that for p < p

∗

r(p) =

(1 − n

)(1 − F

(min{b

∗

, w

})) + n

(1 − n

)(1 − F

(min{b

∗

, w

})) + n

≥

(1 − n

)

(1 − F

)) + 2n

(1 − n

)(1 − F

(min{b

∗

, w

})) + n

(1 − n

)

(1 − F

)) + 2n

(1 − n

)(1 − F

(min{b

∗

, w

})) + n

= r(p

∗

)

since

+2n

(1−n

)(1−F

(min{b

∗

}))

+2n

(1−n

)(1−F

(min{b

∗

}))

≥ 1 ≥

(1−n

)

(1−n

)

Table I:

Summary Statistics of Commercial Real Estate Property Transactions,

Quake Risk, the Eﬀects of the Northridge 1994 Earthquake,

and CMBS Issuances

Panel A: Summary statistics of COMPS sale and loan transactions

Standard

Mean Median deviation 1

% 99

Sale price ($US) 2,204,878 590,000 10,609,900 112,000 30,047,860

Capitalization rate (%) 10.02 9.75 2.81 4.57 18.35

Property age (years) 37.74 31 32.95 1 109

Loan size (% of price) 75.49 77.27 16.28 17.24 100

Interest rate (%) 8.28 8.25 1.41 5 12

Maturity (years) 16.06 15 10.79 0.50 30

Panel B: Summary statistics of COMPS quake risk and Northridge PGA

Standard

Mean Median Deviation

Quake risk - all properties 0.07 0 0.12

Quake risk - CA properties 0.19 0.20 0.12

Quake risk - LA county properties 0.25 0.20 0.07

Northridge PGA - all properties 7.71 0 14.13

Northridge PGA - CA properties 20.46 18.72 16.40

Northridge PGA - LA county properties 26.78 23.08 13.52

Panel C: Summary statistics of S&P CMBS Transactions

Number

CMBS issuers 24

CMBS transactions 50

Individual properties 482

High PML properties 183

Insured properties 169

Insured High PML properties 98

Panel A reports the distributional characteristics of the property transactions in the COMPS database over the

period January 1, 1992 to March 30, 1999. The mean, median, standard deviation, and one and 99 percentiles

of sale price, capitalization rate (net operating income divided by sales price), property age, loan size (loan-

to-value), loan interest rate, and loan maturity are reported. Panel B reports the mean, median and standard

deviation of average annual loss due to earthquake risk (quake risk) from the AIR database and peak ground

acceleration (PGA) during the Northridge (1994) earthquake from the USGS database across all properties in

the COMPS database. Panel C reports the number of commercial mortgage-backed securities (CMBS) issuers,

total CMBS transactions, total number of earthquake zone properties included in the CMBS transactions, the

number of earthquake zone properties with probable maximum loss above 20% (High PML), the number of

earthquake zone properties with earthquake insurance and the number of insured High PML properties.

Table II:

Earthquake Risk and Insurance for Securitized Loans

Dependent variable Quake insurance Quake insurance Quake insurance

provided? provided? provided?

N 482 482 482

High quake risk 1.51 1.41 1.00

(7.41) (2.55) (3.04)

[35.9%] [33.8%] [24.4%]

Fixed eﬀects?

Year Yes Yes Yes

CMBS issuer No Yes No

CMBS transaction No No Yes

Estimation method Logit Logit Logit

0.09 0.09 0.02

Results from the regressions of an indicator for whether earthquake insurance was pro-

vided for the property securing the loan on an indicator for properties with probable

maximum loss above 20% (High PML), year dummies and transaction attributes. The

data is drawn from the S&P CMBS transactions database. The regressions are estimated

via binary ﬁxed eﬀects (conditional) logistic regression (Logit), as described, with robust

t-statistics reported in parentheses. The increase in probability of earthquake insurance

being provided for High quake risk properties (relative to the mean probability) in given

in square brackets. The reported R

is McFadden’s pseudo R

Table III:

Earthquake Risk and Commercial Real Estate Financing Terms

Dep endent variable Bank loan Interest Leverage Seller ﬁnancing

provided? rate provided?

Sample All Bank loans Bank loans All

N 32,618 3,943 11,478 32,618

Quake Risk -2.6313 0.8337 -0.0279 0.7689

(-2.83) (0.64) (-0.33) (0.74)

Brokered 0.5620 -0.0207 0.0001 -0.2947

(18.62) (-0.35) (0.04) (-8.16)

Broker Buyer 0.1520 -0.0395 -0.0009 0.3684

(1.76) (-0.38) (-0.10) (3.87)

Log (Price) -0.0094 -0.0144 0.0142 -0.2294

(-0.61) (-0.42) (3.39) (-11.07)

Corporate Buyer -0.1878 0.0415 0.0001 -0.2287

(-5.59) (0.63) (1.15) (-5.46)

Property Crime -0.0002 -0.0015 0.0000 -0.0020

(-0.27) (-1.02) (-0.57) (-2.56)

Personal Crime -0.0002 0.0015 0.0001 0.0016

(-0.44) (1.35) (1.08) (2.20)

Age -0.0017 -0.0001 -0.0027 0.0066

(-2.93) (-0.20) (-2.76) (7.90)

Log (Buyer Distance) -0.0908 -0.0103 0.0013 -0.1056

(-12.70) (-0.74) (1.66) (-11.52)

Log (Seller Distance) -0.0002 0.0187 0.0160 -0.0291

(-0.02) (1.62) (1.57) (-3.67)

Development 0.0541 0.1852 -0.0001 -0.0350

(0.80) (1.10) (-0.73) (-0.41)

Maturity -0.0101 0.0024

(-4.18) (1.00)

Floating? -0.3463 0.0841

(-5.89) (6.91)

SBA? -0.3900

(-1.19)

Loan Size/Price -0.3771

(-1.65)

Log(Bank Assets) -0.0605 -0.0009

(-6.14) (-1.39)

Fixed eﬀects?

Census tract Yes Yes Yes Yes

Year Yes Yes Yes Yes

Property type Yes Yes Yes Yes

Estimation method Logit OLS OLS Logit

0.19 0.72 0.32 0.12

Results from the regressions of an indicator for whether a bank loan was provided (ﬁrst column),

the interest rate on an extended bank loan (second column), the leverage (loan size divided by

sale price) on an extended bank loan (third column) and an indicator for whether seller ﬁnancing

was provided (fourth column) on quake risk and property and transaction attributes. The data

is drawn from the COMPS database. The regressors with reported coeﬃcients are the average

annual loss due to earthquake risk (obtained from AIR), indicators for whether a broker arranged

the transaction and for whether the buyer was a broker, the log of the sale price (excluded from

the leverage regression), an indicator for corporate buyers, the 1990 property and personal crime

risks (obtained from CAP Index), the age of the property, the log of buyer and seller distances

from the property, an indicator for development projects, loan maturity in years, indicators for

ﬂoating rate and Small-Business-Administration guaranteed loans, leverage and log of bank assets.

All regressions include ﬁxed eﬀects for property type, year and census tract, with coeﬃcients unre-

ported for brevity. The regressions are estimated via binary logistic regression (Logit) or ordinary

least squares (OLS), as described, with robust t-statistics reported in parentheses. Reported R

for Logit speciﬁcations is McFadden’s pseudo R

Table IV:

Cross-sectional Eﬀects of Earthquake Risk on Provision of Financing

Dependent variable Bank loan Bank loan Bank loan Bank loan

provided? provided? provided? provided?

Sample All Age Tract value All

available available

N 32,618 25,012 25,752 32,618

Quake Risk -2.6626 -0.8694 -0.4747 -2.5873

(-2.86) (-0.86) (-0.21) (-2.78)

(Quake Risk) * (Insurance Firm) 5.6787

(2.20)

(Quake Risk) * Age -0.0540

(-8.62)

(Quake Risk) *(% African-American) -54.0184

(-2.30)

(Quake Risk)*Log (Tract Median Income) 0.0000

(-0.11)

(Quake Risk) * Development -2.1183

(-3.20)

Insurance Firm -1.4721

(-4.21)

Brokered 0.5615 0.5900 0.6260 0.5605

(18.60) (16.92) (17.98) (18.57)

Broker Buyer 0.1502 0.1682 0.1195 0.1507

(1.74) (1.64) (1.12) (1.74)

Log (Price) -0.0067 -0.0416 0.0137 -0.0095

(-0.43) (-2.31) (0.73) (-0.61)

Corporate Buyer -0.1903 -0.2187 -0.2400 -0.1856

(-5.66) (-5.55) (-6.10) (-5.52)

Property Crime -0.0002 -0.0003 -0.0003 -0.0002

(-0.26) (-0.38) (-0.36) (-0.28)

Personal Crime -0.0002 -0.0004 0.0000 -0.0002

(-0.44) (-0.61) (-0.01) (-0.42)

Age -0.0017 -0.0006 -0.0014 -0.0017

(-2.92) (-1.00) (-2.44) (-2.90)

Log (Buyer Distance) -0.0903 -0.0884 -0.0849 -0.0908

(-12.61) (-10.83) (-10.07) (-12.68)

Log (Seller Distance) -0.0005 -0.0010 -0.0046 -0.0001

(-0.07) (-0.13) (-0.62) (-0.01)

Development 0.0539 -0.0025 0.0344 0.1633

(0.80) (-0.03) (0.42) (2.16)

Fixed eﬀects?

Census tract Yes Yes Yes Yes

Year Yes Yes Yes Yes

Property type Yes Yes Yes Yes

Estimation method Logit Logit Logit Logit

0.19 0.16 0.20 0.19

Results from the regressions of an indicator for whether a bank loan was provided on quake risk and property and

transaction attributes. The data is drawn from the COMPS database. The regressors with reported coeﬃcients are the

average annual loss due to earthquake risk (obtained from AIR), indicators for whether an insurance ﬁrm (insurer or

insurance broker) purchased the property, for whether a broker arranged the transaction and for whether the buyer was a

broker, the log of the sale price, an indicator for corporate buyers, the 1990 property and personal crime risks (obtained

from CAP Index), the age of the property, the log of buyer and seller distances from the property, an indicator for

development projects and interactions between the average annual loss due to earthquake risk and the following variables:

insurance ﬁrm purchaser, property age, the fraction of the property’s census tract population that is African-American

(obtained from the 2000 census), the log of median home value in the prop erty’s census tract (obtained from the 2000

census) and an indicator for development projects. All regressions include ﬁxed eﬀects for property type, year and census

tract, with coeﬃcients unreported for brevity. The regressions are estimated via binary logistic regression (Logit) with

t-statistics reported in parentheses. Reported R

for Logit speciﬁcations is McFadden’s pseudo R

Table V:

Earthquake Risk and Selection of Buyers and Banks

Dep endent variable Log of Median Home Value % Deposits Log of

in Buyer’s Zip Code in county bank assets

Sample Non-corporate Bank deposits Bank assets

buyers available available

N 17,967 7,487 11,478

Quake Risk 0.4346 0.4360 0.6033

(2.06) (1.65) (0.34)

Brokered 0.0322 -0.0218 0.0777

(4.15) (-1.74) (1.22)

Broker Buyer 0.0047 0.0038 -0.1009

(0.23) (0.14) (-0.61)

Log (Price) 0.0292 -0.0010 0.4168

(5.36) (-0.05) (3.30)

Corporate Buyer -0.0063 -0.0168

(-0.47) (-0.23)

Property Crime 0.0000 0.0000 -0.0004

(-0.10) (-0.12) (-0.25)

Personal Crime -0.0001 0.0001 -0.0008

(-0.40) (0.66) (-0.67)

Age 0.0000 0.0000 -0.0004

(0.16) (0.11) (-0.35)

Log (Buyer Distance) 0.0813 -0.0045 0.0175

(28.26) (-1.55) (1.06)

Log (Seller Distance) 0.0017 -0.0055 0.0016

(0.93) (-2.27) (0.12)

Development -0.0334 -0.0051 -0.3017

(-1.56) (-0.15) (-2.01)

Log(Loan Size) -0.0265 -0.0488

(-1.28) (-0.40)

Fixed eﬀects?

Census tract Yes Yes Yes

Year Yes Yes Yes

Property type Yes Yes Yes

0.47 0.47 0.37

Results from the regressions of the median home value in the buyer’s zip code (from the 2000

census), the ratio of in-county deposits to total deposits for the bank extending the loan and the

log of the total assets of the bank extending the loan on quake risk and property and transaction

attributes. The regressors with reported coeﬃcients are the average annual loss due to earthquake

risk (obtained from AIR), indicators for whether a broker arranged the transaction and for whether

the buyer was a broker, the log of the sale price (excluded from the cap rate and sale price regres-

sions), an indicator for corporate buyers, the 1990 property and personal crime risks (obtained

from CAP Index), the age of the property, the log of buyer and seller distances from the property

and an indicator for development projects. In the second and third columns, the log of loan size is

included as an additional control. All regressions include ﬁxed eﬀects for property type, year and

census tract, with coeﬃcients not reported for brevity. All regressions are estimated via ordinary

least squares (OLS) with robust t-statistics rep orted in parentheses.

Table VI:

Earthquake Risk and Commercial Real Estate Prices

Dep endent variable Cap Log of Log of

rate price price

N 12,444 12,444 12,444

Quake Risk 1.2609 -0.1342 5.3954

(0.86) (-1.00) (12.96)

Log (Earnings) 0.9255 0.9503

(286.92) (273.58)

(Quake Risk) * Log (Earnings) -0.3468

(-14.12)

(Quake Risk) * Age -0.0035

(-3.13)

Brokered 0.3786 -0.0366 -0.0336

(5.71) (-5.91) (-5.52)

Broker Buyer 0.1068 -0.0079 -0.0073

(0.80) (-0.61) (-0.58)

Corporate Buyer 0.2281 0.0255 0.0270

(3.27) (3.85) (4.12)

Property Crime -0.0023 0.0002 0.0002

(-1.62) (1.50) (1.33)

Personal Crime 0.0037 -0.0004 -0.0004

(2.84) (-3.27) (-3.14)

Age 0.0053 -0.0011 -0.0009

(3.21) (-7.13) (-4.68)

Log (Buyer Distance) 0.0424 0.0035 0.0025

(3.42) (2.82) (2.05)

Log (Seller Distance) 0.0430 -0.0016 -0.0019

(3.59) (-1.37) (-1.67)

Development 0.0938 0.0098 0.0090

(0.55) (0.60) (0.56)

Fixed eﬀects?

Census tract Yes Yes Yes

Year Yes Yes Yes

Property type Yes Yes Yes

0.43 0.97 0.98

Results from the regressions of capitalization rate (current earnings divided by sale price) and

log of sale price on quake risk and property and transaction attributes. The regressors with

reported coeﬃcients are the average annual loss due to earthquake risk (obtained from AIR), the

log earnings (second and third columns), indicators for whether a broker arranged the transaction

and for whether the buyer was a broker, an indicator for corporate buyers, the 1990 property and

personal crime risks (obtained from CAP Index), the age of the property, the log of buyer and

seller distances from the prop erty, an indicator for development projects and interactions between

earthquake risk and the following variables: log of earnings and property age. In the fourth and

ﬁfth columns, the log of loan size is included as an additional control. All regressions include

ﬁxed eﬀects for property type, year and census tract, with coeﬃcients not reported for brevity.

All regressions are estimated via ordinary least squares (OLS) with robust t-statistics reported in

parentheses.

Table VII:

The Eﬀect of the Northridge Earthquake on

Commercial Real Estate Prices and Financing

Dep endent variable Cap Loan Log of % Deposits

rate provided? bank assets in county

Sample All All L.A. County L.A. County

N 12,444 32,618 3,989 3,596

Quake Risk 1.5618 -2.7836 0.2840 0.3651

(1.08) (-2.96) (0.16) (1.38)

(Quake Risk) * Log (1+Days Post-quake) 0.1889 0.5732 0.1896 0.1204

(0.58) (2.57) (0.38) (1.77)

(Quake Risk) * (Year Post-quake) -1.7086 -2.5987 0.3925 -0.4682

(-1.03) (-2.31) (0.17) (-1.53)

Log (1+Days Post-quake) 0.0205 0.0000 -0.1043 -0.0230

(0.57) (0.00) (-1.09) (-1.86)

PGA 0.0314 0.0114 -0.0304 0.0033

(1.54) (0.87) (-1.22) (0.88)

Log (Price) -0.0090 0.8729 0.0460

(-0.58) (3.97) (1.33)

Brokered 0.3767 0.5625 -0.0556 -0.0220

(5.67) (18.63) (-0.48) (-1.33)

Broker Buyer 0.1097 0.1501 -0.1380 0.0249

(0.82) (1.74) (-0.54) (0.87)

Corporate Buyer 0.2268 -0.1885 0.0178 0.0271

(3.25) (-5.61) (0.13) (1.43)

Property Crime -0.0024 -0.0002 0.0010 0.0000

(-1.64) (-0.24) (0.40) (-0.06)

Personal Crime 0.0037 -0.0003 -0.0019 0.0001

(2.88) (-0.45) (-0.79) (0.19)

Log (Buyer Distance) 0.0423 -0.0909 -0.0160 0.0058

(3.42) (-12.70) (-0.54) (1.43)

Log (Seller Distance) 0.0433 -0.0004 0.0225 -0.0071

(3.62) (-0.06) (0.98) (-2.20)

Development 0.0938 0.0533 -0.3497 -0.0591

(0.55) (0.79) (-1.05) (-1.08)

Age 0.0053 -0.0017 0.0001 0.0003

(3.21) (-2.95) (0.06) (0.87)

Log (Loan Size) -0.4557 -0.0699

(-2.20) (-2.09)

Fixed eﬀects?

Census tract Yes Yes Yes Yes

Property type Yes Yes Yes Yes

Year Fixed Eﬀects? Yes Yes Yes Yes

Estimation method OLS Logit OLS OLS

0.43 0.19 0.37 0.35

Results from the regressions of capitalization (current earnings divided by sale price) rate, an indicator for whether a loan was

provided, the log of the total assets of the bank extending the loan and the ratio of in-county deposits to total deposits for

the bank extending the loan on quake risk, local shaking from the Northridge (1994) earthquake and property and transaction

attributes. The data is drawn from the COMPS database. The regressors with reported coeﬃcients are the average annual

loss due to earthquake risk (obtained from AIR), the interaction of earthquake risk with the log of one plus the number of

days following the Northridge quake on which the transaction took place (for transactions within one year of the quake), the

interaction of earthquake risk with a dummy for the year following the Northridge quake, the log of one plus the number

of days following the Northridge quake on which the transaction took place, the peak ground acceleration (PGA), a shaking

intensity measure, of the Northridge earthquake at the property’s location (provided by the USGS), the log of the sale price

(excluded from the cap rate regression), indicators for whether a broker arranged the transaction and for whether the buyer

was a broker, an indicator for corporate buyers, the 1990 property and personal crime risks (obtained from CAP Index), the

age of the property, the log of buyer and seller distances from the property, an indicator for development projects, and, in

the third and fourth columns, the log of loan amount. All regressions include ﬁxed eﬀects for property type, year and census

tract, with coeﬃcients not reported for brevity. The regressions are estimated via binary logistic regression (Logit) or ordinary

least squares (OLS), as described, with robust t-statistics reported in parentheses. Rep orted R

for Logit speciﬁcations is

McFadden’s pseudo R

Table VII I:

The Eﬀect of the Northridge Earthquake on

Transaction Volumes for Properties

with Varying Quake Risks

Dep endent variable Quake risk Quake risk Quake risk

Sample All Bank Loans No Bank Loans

N 32,618 17,141 15,451

Log (1+Days Post-quake) 0.0003 0.0008 0.0000

(1.93) (2.61) (0.11)

PGA 0.0002 0.0002 0.0006

(0.78) (0.60) (1.07)

Log (Price) 0.0000 0.0000 -0.0001

(-0.20) (-0.10) (-0.64)

Brokered 0.0000 0.0003 -0.0003

(-0.13) (1.01) (-0.94)

Broker Buyer 0.0000 0.0004 -0.0001

(0.04) (0.51) (-0.12)

Corporate Buyer 0.0000 0.0000 -0.0002

(-0.10) (-0.05) (-0.88)

Property Crime 0.0000 0.0000 0.0000

(0.71) (-0.44) (0.82)

Personal Crime 0.0000 0.0000 0.0000

(-0.73) (0.32) (-0.79)

Log (Buyer Distance) 0.0000 0.0000 0.0001

(0.43) (-0.45) (1.26)

Log (Seller Distance) 0.0000 0.0000 0.0000

(-0.22) (-0.25) (0.54)

Development -0.0002 -0.0006 0.0006

(-0.45) (-1.33) (1.38)

Age 0.0000 0.0000 0.0000

(-0.48) (-1.11) (-0.40)

Fixed eﬀects?

Census tract Yes Yes Yes

Year Yes Yes Yes

Property type Yes Yes Yes

Estimation method OLS OLS OLS

0.99 0.99 0.99

Results from the regressions of the average annual loss due to earthquake risk on the log

of one plus the number of days following the Northridge quake on which the transaction

took place (for transactions within one year of the quake), local shaking from the North-

ridge earthquake, the log of the sale price, indicators for whether a broker arranged the

transaction and for whether the buyer was a broker, an indicator for corporate buyers,

the 1990 property and personal crime risks (obtained from CAP Index), the age of the

property, the log of buyer and seller distances from the property, an indicator for devel-

opment projects and property age. The data is drawn from the AIR, COMPS and USGS

databases. All regressions include ﬁxed eﬀects for property type, year and census tract,

with coeﬃcients not reported for brevity. The regressions are estimated via ordinary

least squares (OLS), with robust t-statistics rep orted in parentheses.

Table IX:

Hurricane Risk and Commercial Real Estate Financing

Dep endent variable Bank loan provided? Bank loan provided? Bank loan provided?

N 32,618 32,618 32,618

Hurricane Risk -0.0029 -0.0043 0.0048

(-1.09) (-2.26) (1.25)

(Hurricane Risk)*(Cat. Ins. Price Index) -0.0001

(-2.73)

Log (Price) -0.0094 -0.0072 -0.0063

(-0.61) (-0.51) (-0.44)

Brokered 0.5620 0.5389 0.5405

(18.62) (19.37) (19.42)

Broker Buyer 0.1525 0.2381 0.2445

(1.76) (2.91) (2.99)

Corporate Buyer -0.1874 -0.1990 -0.1990

(-5.58) (-6.43) (-6.43)

Property Crime -0.0002 -0.0006 -0.0006

(-0.32) (-1.83) (-1.87)

Personal Crime -0.0002 0.0001 0.0001

(-0.39) (0.21) (0.25)

Log (Buyer Distance) -0.0909 -0.0848 -0.0850

(-12.70) (-12.74) (-12.76)

Log (Seller Distance) -0.0001 -0.0005 -0.0005

(-0.01) (-0.08) (-0.08)

Development 0.0540 -0.0027 -0.0032

(0.80) (-0.04) (-0.05)

Age -0.0017 -0.0020 -0.0020

(-2.94) (-3.76) (-3.72)

Fixed eﬀects?

Geographical Tract Zip Code Zip Code

Year Yes Yes Yes

Property type Yes Yes Yes

Estimation method Logit Logit Logit

0.19 0.17 0.17

Results from the regressions of an indicator for whether a bank loan was provided on hurricane risk and

property and transaction attributes. The data is drawn from the COMPS database. The regressors with

reported coeﬃcients are the property’s percentile hurricane risk within its county (obtained from AIR), an

interaction of this hurricane risk with Guy Carpenter’s Catastrophe Insurance Price Index (obtained from

Guy Carp enter), the log of the sale price, indicators for whether a broker arranged the transaction and for

whether the buyer was a broker or a corporation, the 1990 property and personal crime risks (obtained from

CAP Index), the log of buyer and seller distances from the property, an indicator for development projects

and the age of the prop erty. All regressions include ﬁxed eﬀects for property type, year and census tract,

with coeﬃcients unrep orted for brevity. The regressions are estimated via binary logistic regression (Logit)

with t-statistics reported in parentheses. Reported R

is McFadden’s pseudo R