JUNE 2006 E W I N G E T A L . 395

Insurer Stock Price Responses to Hurricane Floyd: An Event Study Analysis Using Storm Characteristics

BRADLEY T. EWING Jerry S. Rawls College of Business, and Wind Science and Engineering Research Center, Texas Tech University, Lubbock, Texas

SCOTT E. HEIN Jerry S. Rawls College of Business, Texas Tech University, Lubbock, Texas

JAMIE BROWN KRUSE Natural Hazards Mitigation Research Center, East Carolina University, Greenville, North Carolina

(Manuscript received 20 January 2005, in final form 30 September 2005)

ABSTRACT

This research uses an event study methodology to examine the effect of Hurricane Floyd and the associated scientific and media releases on the market value of insurance firms. The research is unique in that information describing the development of the storm over time and space is incorporated to determine how the financial market reacted to changing news about a storm’s characteristics. Key empirical results can be summarized as follows. Overall, there was a negative effect on insurer stock price changes around the synoptic life cycle of the storm; however, this effect was neither constant nor was it always negative on each day of the cycle. Significant market reaction to the news concerning the path and strength of the storm prior to the storm landfall was found. The results herein suggest that markets find reliable time-sensitive reports provided by the , the National Hurricane Center, and other media outlets to be valuable information.

1. Introduction tops the list (the 11 September terrorist attack ranks second). Given the large amount of physical and eco- In September 1999 Hurricane Floyd hit the area nomic damage it should not be surprising that insurance around Wilmington/New Hanover County, North firms were materially affected by these windstorms. Carolina. Swiss Re ranks Floyd 23rd on its list of the 40 What is not so clear is exactly how and to what extent most costly insurance losses worldwide from 1970 the value of insurance firms would respond to the ex- through 2004. (The ranking uses property and business pected damage. The accuracy of and public access to interruption losses, excluding life and liability losses information concerning the expected magnitude of a and can be found online at http://www.swissre.com.) tropical system has expanded significantly over the 10 Insured property and business interruption losses of yr since Hurricane Andrew. Because the severity of a $2.548 billion (indexed in 2002 dollars) were attributed hurricane develops and evolves over both time and to Floyd. Although Floyd was devastating to millions of space, dating this type of event by date of landfall alone people, it ranks far below Hurricane Andrew, which is likely to give an incomplete description of the market moved across south of in August 1992. response. With $20.511 billion of insured property and business It is expected that the development of potentially interruption losses (indexed in 2002 dollars), Andrew catastrophic events plays an important role in deter- mining the value of firms in the insurance industry. As such, this study seeks to determine how the stock prices Corresponding author address: Dr. Jamie Brown Kruse, Center for Natural Hazards Research, East Carolina University, Brew- of insurance firms behaved before, during, and imme- ster A-438, Greenville, NC 27858. diately after Hurricane Floyd because financial market E-mail: [email protected] participants would take accurate and readily available

© 2006 American Meteorological Society

Unauthenticated | Downloaded 09/29/21 04:12 AM UTC

WAF917 396 WEATHER AND FORECASTING VOLUME 21 spatial and atmospheric characteristics of the storm into earlier studies. In contrast, this study examines market account. We also perform a comparison study analyzing reactions produced by Hurricane Floyd on the insur- the response of insurer stock prices to 1992’s Hurricane ance industry prior to hurricane landfall, taking into Andrew. account the actual development of the storm over time Economic research aimed at measuring the effect of and space. It is logical to expect that market analysts do an event on the value of firms and businesses is typi- not wait for all uncertainty to be resolved, but rather cally conducted in the event study framework. It has use all information as it becomes available to update become common in finance to measure an event’s eco- and refine their estimates of the financial impact of a nomic impact by using asset prices observed over a rela- hurricane. We conduct an event study that utilizes the tively short time period (Campbell et al. 1997). This synoptic life cycle of a hurricane as the event “date” methodology utilizes financial market data to estimate and describe the responses of insurance stock returns in the equilibrium or “normal” return for a stock price or light of storm characteristic information. In other index over a window of time (via a standard time series words, we see a multitude of different events that cor- model), and then isolate the effect of the event by con- respond with the release of information by the National trolling for when the event occurred. A normal return Hurricane Center (NHC) and National Weather Ser- for an industry is seen as capturing all quantifiable risk vice (NWS). Although investors may use a broad vari- in the value of the stock. In other words, the stock price ety of available information sources, including media takes into consideration the value of future insurance reports and expert opinion, we use the measurements premiums and payouts based on the first and second provided by the NHC because it is relatively unbiased, moments of the net income distribution. The first mo- publicly available, and considered credible. Johnson ment captures the statistical expectation. The second and Holt (1997) note that, generally speaking, weather moment captures the variance and therefore the re- data and forecasts are “public goods” that generate sig- maining risk, given the risk management strategies in nificant positive externalities in the economic sense. place for typical firms in the industry. Abnormal per- The input of meteorological and atmospheric param- formance is indicated by returns that are statistical out- eters such as pressure, wind speed, storm direction, and liers to the distribution, in other words, an event with location associated with the hurricane event can be very unexpected magnitude. Thus, normal performance may important information to financial markets and are be compared to abnormal performance that is, presum- used to construct the windstorm characteristics. Hurri- ably, driven by the event (Campbell et al. 1997). Using cane track and intensity as provided by the NHC were event study methodology, research has shown that the used to characterize the evolution of the hurricane market value of insurance firms is significantly affected event. This unique feature of our research provides a by catastrophic events, such as a hurricane. Lamb significant improvement over the standard event study (1998) found evidence that in 1992 when Hurricane An- analysis. To date, no other study has focused on the drew hit south Florida and Louisiana, property and ca- characteristics of a windstorm when analyzing or ex- sualty firms with exposures in these areas saw their plaining the market responses of insurer stock prices. stock returns adversely affected. This is not surprising The ability of an unfettered trading institution to given the amount of destruction these natural hazards quickly assimilate relevant information is summarized cause and the corresponding insured losses. Still others by the efficient markets hypothesis that states all rel- have argued that a natural disaster may have two op- evant and available information is incorporated into posing effects on the value of insurer stock prices current financial asset prices (Campbell et al. 1997). (Angbazo and Narayanan 1996)—a negative effect is The efficient markets hypothesis is one of the most hypothesized because of payments on claims and a influential and fundamental theories in finance. It sim- positive effect may be because of expectations of higher ply states that financial market prices should reflect all future premiums. available information. In particular, this means that MacKinlay (1997, p. 37) has stated in his review of market prices should reflect expectations of relevant “event studies in economics and finance” that “An im- events. For example, if market participants are con- portant characteristic of a successful event study is the vinced that a firm is going to increase its dividend, this ability to identify precisely the date of the event.” How- will already be priced in the stock price and the stock ever, previous research has not specifically considered price will not change with the news of a dividend in- the storm’s characteristics as the storm evolves over crease. Roll (1984), for example, provides evidence time, but instead the focus has been centered on the showing that financial markets, in the case of prices for date that the hurricane made landfall. That is, the date orange juice futures contracts, are very much linked to that a hurricane makes landfall has defined the event in weather forecasts, especially predictions of freezing

Unauthenticated | Downloaded 09/29/21 04:12 AM UTC JUNE 2006 E W I N G E T A L . 397 weather, provided by the NWS. In a similar vein, this TABLE 1. Summary of descriptions of Hurricane Floyd. (Source: paper investigates whether the financial markets re- NHC) sponded to hurricane-track and intensity information as Synoptic life cycle 7–17 Sep 1999 provided by the NHC therefore indicating its value. Estimate of total U.S. damage $3–$6ϩ billion Number of deaths in United States 56 2. Background and related research Maximum wind speed (Saffir–Simpson 135 kt (category 4/5) The direct losses from hurricanes can total in the scale category) millions, even billions, of dollars (West and Lenze Wind speed at landfall 90 kt Saffir–Simpson scale category at landfall Category 2 1994). Hurricanes have immediate adverse effects on Location of landfall 33.8°N, 78.0°W an economy and these effects may persist for some time into the future (Ewing and Kruse 2001, 2002). Our fo- cus is on Hurricane Floyd for two reasons. First, the (Dasgupta 1999). The five states that suffered the larg- technological infrastructure for accurately predicting est insured losses to personal and commercial property and disseminating hurricane information to the public and vehicles were North Carolina ($1.25 billion), New was vastly different from that at the time of Andrew. Jersey ($95 million), Pennsylvania ($63 million), South Second, Floyd was a large storm that had the potential Carolina ($60 million), and Maryland ($45 million; Das- to cause losses even worse than were observed. Figure gupta 1999). 1 shows a comparison of satellite images of Floyd and The city of Wilmington, North Carolina, near Floyd’s Andrew. Floyd arrived 7 yr after Andrew making land- landfall, is a popular destination for vacationers and is fall near Cape Fear, North Carolina, on 16 September located about 30 miles from the Atlantic coast. The 1999. Table 1 gives vital statistics of Hurricane Floyd economy of Wilmington has a large commercial district compiled from a preliminary report prepared by the and depends on tourism, entertainment, manufactur- NHC (Pasch et al. 1999). Floyd began as a tropical ing, retail services, and a variety of white-collar busi- depression, progressed to category-4 hurricane within a nesses. Thus, when a storm such as Floyd devastates an few days, and then weakened to category-2 intensity area with a developed economic base, the value of in- before landfall. Figure 2 plots the storm path of Hurri- surance firms should be affected, because insured losses cane Floyd. Heavy rainfall amounts and severe flooding will be quite high and the opportunity to insure future were mainly responsible for the 56 deaths attributed to clients depends on business survival. Surprisingly, very Floyd, making it the deadliest hurricane in the United little research has examined the effect of hurricanes on States since Agnes in 1972 (Pasch et al. 1999). As a insurer stock prices. Lamb (1995, 1998) and Angbazo result of the storm, business activity was disrupted in a and Narayanan (1996) focused exclusively on property large area (Ewing and Kruse 2001). Homeowners and and liability firms and examined how a hurricane af- businesses filed 526 550 claims as a result of the storm fected insurer stock prices using event study methodol-

FIG. 1. Comparison of satellite images of Hurricanes Floyd and Andrew. (Source: National Oceanic and Atmospheric Administration)

Unauthenticated | Downloaded 09/29/21 04:12 AM UTC 398 WEATHER AND FORECASTING VOLUME 21

FIG. 2. Storm track of Hurricane Floyd. Points on the figure represent the path of the hurricane. Points on this figure correspond to the data presented in Table 2. The path can be differentiated by a variety of storm characteristics. (Source: Unisys) ogy, but did not take the storm characteristics into ac- the affected areas. This result contrasts the findings of count in the analysis. Rather, they focused only on the Lamb (1995) who also examined Andrew and con- timing of hurricane landfall. It is our conjecture that the cluded that the market discriminates based on risk ex- timely information provided by the NWS is an impor- posure. Based on this conflicting research, a significant tant determinant of the market reaction to windstorms. finding for the effects of Floyd using the broader in- Thus, our study has the potential to add significantly to surer portfolio is all the more compelling. However, the the research on catastrophic events and market re- lack of a significant finding would not rule out possible sponses. hurricane-related effects. To date, only the response of The market value of a firm is captured by its stock property and liability firms to hurricanes has been ex- price. Movements in the stock price reflect underlying amined. Moreover, a blending of lines of insurance and changes in firm value. To examine firms in the insur- geographic distribution is present in the S&P Insurer’s ance industry, we use the Standard & Poor’s (S&P) stock index, resulting in a potential loss of information. Insurer stock price index. This index then tracks a port- Thus, we view our investigation as being conservative folio of the stock prices of twenty-three major publicly in uncovering evidence of market anticipation of hur- traded insurance firms. These firms are engaged in the ricane landfall. provision of property and casualty (10 firms), multiline Insurer stock prices should be affected by expecta- (3 firms), broker (2 firms), and/or life and health insur- tions of damages associated with a windstorm that ex- ance (8 firms). It is specifically designed by S&P to ceed the expectation already priced into the market. capture general movements in the stock prices of firms Higher expected damages, associated with true catas- in the insurance industry. In contrast to Lamb (1998), trophes, imply greater expenditures by insurers and who looks exclusively at stock prices of property and should, therefore, result in lower earnings and returns casualty insurance firms, we chose this broader mea- for insurance firms. However, many times windstorms sure of insurer stock price because Angbazo and Naray- do not fully develop into hurricanes and may remain a anan (1996) have provided some evidence of hurricane tropical storm, or the storm may shift direction and “contagion effects” in the value of insurance firms. In remain at sea. Thus, we expect that expectations of this respect, contagion refers to a hurricane affecting damages will adapt as more detailed information about most insurers regardless of actual claims exposure in a storm becomes available and, based on the efficient

Unauthenticated | Downloaded 09/29/21 04:12 AM UTC JUNE 2006 E W I N G E T A L . 399 markets hypothesis, insurer stock price responses are ering the standard risk-adjusted return for a portfolio as expected to respond accordingly. given by The incorporation of spatial and temporal character- R ϭ ␣ ϩ ␤R ϩ u , ͑1͒ istics of the storm makes this event study unique and t mt t allows us to determine how the financial market reacts where, in our case, Rt is the daily return on the S&P to changing news about a storm’s development. The Insurer stock price index (i.e., a portfolio of insurance results provided by such a study are particularly valu- firms’ stock prices), Rmt is the daily return on the mar- 1 able to the risk management strategies of firms in the ket portfolio (e.g., S&P 500 index) , ut is the zero mean insurance industry and other financial market partici- disturbance term, and ␣ and ␤ are parameters of the pants. These results are also important to state and market model. The Tropical Prediction Center of the national policymakers that deal with insurance issues, NWS defines the Atlantic hurricane season as 1 June and may also be of interest to financial market regula- through 30 November. The sample period for the daily tory agencies (e.g., Securities and Exchange Commis- data used in the analysis is from 1 October 1998 sion) and insurance industry regulators. through 31 July 2000. The sample period was chosen in order for us to obtain as long a sample period as pos- 3. Event study methodology sible without overlapping with other hurricanes. Thus, the sample period encompasses all of the 1999 hurri- The objective of this research is to determine wheth- cane season as well as the end of the 1998 and begin- er information about windstorm activity significantly ning of the 2000 hurricane seasons. Floyd was the only affects the stock prices of insurers. In particular, we hurricane to hit the Atlantic coast during this period. focus on Hurricane Floyd and define the event not sim- The sample size of 478 observations of daily returns for ply as the date of landfall, per se. Instead, we use the our insurance stock index is sufficient to capture the public information on the various measurements that underlying relationship between the insurance stock specifically describe the characteristics and develop- portfolio and the overall market portfolio. Equation (1) ment of the windstorm before it made landfall. Thus, is typically estimated using ordinary least squares to we conduct a series of event studies, in the sense of a obtain parameter estimates.2 However, it is not uncom- day-by-day analysis, and then compare and contrast the mon for stock returns to exhibit a statistical property results obtained from the various windstorm defini- known as autoregressive conditional heteroscedasticity tions. The study proceeds in two parts. In the first (ARCH), in which case an estimation procedure ca- analysis, we conduct a day-by-day event study and com- pable of handling this feature of the data is appropriate. pare the market response to the storm’s characteristics. Examination of the ordinary least squares regression In our event study terminology we vary the event date, revealed the presence of these ARCH effects. There- with one date for each day that the stock market is open fore, we estimated the model using the generalized au- for trading in the synoptic life cycle of Floyd. Moreover, toregressive conditional heteroscedasticity (GARCH) we analyze each separate (potential) event date with model of Bollerslev (1986).3 In effect, this model simul- the windstorm characteristics evolving around that day. taneously estimates both the mean of the series [i.e., An interesting feature of the storm data is that both Eq. (1)] and the (conditional) variance of the series maximum sustained wind speed and minimum central using the method of maximum likelihood. The main pressure are independent of time (Pasch et al. 1999). It advantage of this technique is that by taking into ac- is only when these characteristics are combined with count the additional information provided in the time- space/location that they have the potential to cause (in- varying nature of volatility, one can obtain more effi- sured) damage, and therefore affect the value of insur- ance firms. Thus, on any day in the synoptic life cycle, this information proxy is an index of the storm’s dam- 1 The S&P 500 Stock Price Index is the most widely used mea- age-causing potential. It is for this reason that we use a sure of overall stock market performance in the United States. 2 We computed the quasi-maximum likelihood covariances and number of possible event dates to capture the affects of standard errors as described in Bollerslev and Wooldridge (1992). the storm. The model is estimated under the assumption that the errors are The second analysis examines the cumulative effects conditionally normally distributed. of the storm. A brief description of the methodology 3 Below, we discuss the event study methodology as it pertains follows. Consider a portfolio of stocks and its return, to the returns. We also examined the variance equation in a simi- lar fashion, but found no evidence of changes in conditional vari- which is calculated by finding the daily change in the ance over the synoptic life cycle of either Hurricanes Floyd or index. The return is “risk adjusted” by controlling for Andrew. The variance results are thus not reported, but are avail- the returns of the overall market. We begin by consid- able upon request.

Unauthenticated | Downloaded 09/29/21 04:12 AM UTC 400 WEATHER AND FORECASTING VOLUME 21 cient parameter estimates. A detailed discussion of the the market portfolio during the event period of X days. (G)ARCH model can be found in Ewing and Kruse The result is the estimated risk-adjusted return for the (2002). Note also that, by design, Eq. (1) removes the insurance portfolio for each event date. Abnormal re- “commingling” effect of movements in the overall stock turn is calculated as the difference between realized market from movements in the insurance stock portfo- return and estimated return for each event day as lio. AR ϭ R Ϫ ͑␣ ϩ ␤R ͒. ͑4͒ Previous research has considered the arrival of a hur- t t mt ricane as a single event, defined by the hurricane’s date We cumulate abnormal returns in order to examine of landfall. This seems too limited to us. Rather, we see performance over a holding period. The cumulative ab- the arrival of a hurricane as a sequence of events normal return (CAR) for a 10-day event window (the 9 shaped by the release of information on the storm’s market days in the synoptic life cycle plus 1 day after location, projected path, wind speeds, intensity, etc. the life cycle ended) is given by

Our dating of the event then begins with any news sug- 10 gesting that the storm might pose a threat to land any- ϭ ͑ ͒ CAR ͚ ARt. 5 where in the United States. We then view each subse- tϭ1 quent day as a new event characterized by new infor- To test for statistical significance of abnormal returns mation provided to the public. We use the storm we follow Liu et al. (1992) and use Z statistics or t characteristics to summarize this daily information. statistics (distributed standard normal) to determine if To examine the effect of Hurricane Floyd as it de- the abnormal and cumulative abnormal returns are sta- veloped over time, we augment Eq. (1) with a set of tistically different from zero. If the event had no effect dichotomous variables defined by on insurer stock prices then these return measures ϭ would not be different from zero. Dst 1 for day s in the synoptic life cycle, and D ϭ 0 otherwise, s ϭ 1,2,...9, ͑2͒ An important feature of this study is the use of the st information regarding storm intensity and location as where s runs from Tuesday, 7 September through Fri- defining or distinguishing the event. The findings (i.e., day, 17 September, one day after the storm finally made the significance of abnormal returns) can be compared landfall. There are a total of nine trading days (i.e., days to determine if the market discriminates by the storm in which the stock market was open) during the synop- characteristics. It is expected that we may find distinct tic life cycle of Hurricane Floyd. Thus, we have the differences in the responses of insurer stock prices to event study model such things as the direction of the hurricane path, its intensity, location, and categorizations by wind speed, 9 ϭ ␣ ϩ ␤ ϩ ␦ ϩ ͑ ͒ etc. Rt Rmt ͚ sDst ut. 3 sϭ1 4. Discussion of results Examination of the estimated ␦ coefficients provides information as to the storm’s impact on insurance stock The results obtained from estimating the synoptic life returns on each day the market was open during cycle event study outlined in Eq. (3) are presented in Floyd’s synoptic life cycle. Previous research has fo- Table 2a. Several general comments are worth noting. ␦ ␦ cused attention on 8 and 9, ignoring the information First, the coefficient on the measure of market return provided prior to the storm’s arrival. Furthermore, by (S&P 500 Stock Index) is about 0.95. Because this co- relating these estimates to the storm’s characteristics efficient is less than one, it implies that, in general, the over this time we will be able to shed light on how the insurance firms’ stock prices are less volatile than the financial market responded to the storm’s anticipated overall stock market. This is consistent with the idea effect on insurance companies. that insurers are particularly adept at managing risk or To gain additional insight into the cumulative effect may reflect industry regulatory requirements that limit of Hurricane Floyd on insurance stock returns, we com- risky investments. Second, the model exhibits a good fit pute daily abnormal returns (AR) as the difference be- as measured by the adjusted R2 value of 0.41. Thus, the tween realized return and the risk-adjusted normal re- estimated model explains over 40% of the total varia- turn given in Eq. (1), where the latter represents a mea- tion in insurer stock returns. This compares favorably sure of superior performance (Rozeff and Zaman with many studies on stock market return behavior 1998). Equation (1) is estimated for observations prior (Campbell et al. 1997). Next, we turn our attention to a to the event period to obtain parameter estimates. discussion of the daily movements in insurer stock re- These estimates are then applied to realized returns on turns and the characteristics of the developing storm.

Unauthenticated | Downloaded 09/29/21 04:12 AM UTC JUNE 2006 E W I N G E T A L . 401

TABLE 2. Hurricane Floyd estimation results for S&P Insurer stock price return (Rt) model. (a) For the results presented the event study model was corrected for autoregressive conditional heteroscedasticity in order to obtain more efficient parameter estimates. Quasi-maximum likelihood covariances and standard errors are computed as described in Bollerslev and Wooldridge (1992). The model is estimated under the assumption that the errors are conditionally normally distributed. (b) The test statistic of Ϫ2.07 (standard normal distribution) that is reported is significant at the 5% level.

(a) Synoptic life cycle event study Coefficient Std error Z statistic Probability Central pressure (mb) Wind speed (kt) Constant Ϫ0.000 652 0.000 553 Ϫ1.180 728 0.2377

Rmt 0.946 757 0.053 125 17.821 32 0.0000

D1 (7 Sep 1999) 0.003 889 0.000 939 4.139 555 0.0000 1008 25 D2 (8 Sep 1999) 0.004 901 0.000 783 6.262 155 0.0000 1005 35 Ϫ Ϫ D3 (9 Sep 1999) 0.007 551 0.000 565 13.358 38 0.0000 1003 50 Ϫ Ϫ D4 (10 Sep 1999) 0.019 711 0.000 582 33.855 63 0.0000 985 60 D5 (13 Sep 1999) 0.005 089 0.000 577 8.820 300 0.0000 922 135 Ϫ Ϫ D6 (14 Sep 1999) 0.002 077 0.000 623 3.331 546 0.0009 927 135 Ϫ Ϫ D7 (15 Sep 1999) 0.013 474 0.000 630 21.390 21 0.0000 938 120 D8 (16 Sep 1999) landfall 0.007 282 0.000 912 7.988 196 0.0000 956 90 D9 (17 Sep 1999) 0.001 238 0.000 553 2.238 347 0.0252 984 50 R2 0.421 343 Adjusted R2 0.405 131 (b) Cumulative effects analysis 10-day CAR Ϫ0.021 69 t statistic Ϫ2.07

A day-by-day analysis not negative, but instead are positive and statistically significant. News reports over these 2 days focused on The reported D coefficients on the respective D vari- how far out to sea the storm was located and gave no ables in Table 2 provide estimates of the abnormal re- real indication of the impending damage (see the re- turns on that day observed for the S&P Insurer stock price index. Table 3 provides a summary of selected leases from The Associated Press, Miami Herald, and new reports concerning Hurricane Floyd. The reports Chicago Sun-Times in Table 3). It is possible that in- were compiled from a LexisNexis search using combi- surers would benefit from a storm being averted, if me- nations of the terms hurricane, tropical storm, insur- dia coverage highlighted the importance of being in- ance, and Floyd.4 The search was restricted to corre- sured, and insurers, in effect, received a form of low spond to the synoptic life cycle of the storm (7–17 Sep- cost, industry-wide advertising. Our LexisNexis search tember 1999). On Tuesday, 7 September, Floyd was found numerous news articles and reports focusing on classified as a tropical depression with winds in the 25– what types of insurance coverage homeowners require, 30-kt range.5 The storm was located far off in the At- how to file claims, precautions to take to reduce dam- lantic (14.6°N latitude, 46.2°W longitude) on Tuesday age, etc. However, on Thursday, 9 September, the and moved mostly westward toward and the Ca- maximum sustained wind speed of the storm rose to as ribbean, and then was upgraded to a tropical storm on high as 60 kt with central pressure beginning to fall. The Wednesday, 8 September. Thus, it is not surprising that storm also made a noticeable directional shift toward the coefficients on the event variables D and D are the United States by 1500 eastern daylight time (EDT), 1 2 although it was still far out at sea. This change in di- rection, as well as the expectation that tropical storm Floyd would intensify to hurricane status, was reported 4 As expected, a LexisNexis search on insurance brought up a number of news reports over this period of time. However, there in the Miami Herald (Table 3). The new information did not appear to be any single event that garnered much atten- corresponds to the negative and significant coefficient tion aside from the storm. The Gramm–Leach–Bliley Financial on D3. The storm’s location (17.2°N latitude, 55.5°W Modernization Act was signed into law 12 November 1999. A longitude) and the directional change (noting that the search for significant events associated with the movement of the stock market stays open until 1600 ET) might explain Act through Congress during our examination time interval for Hurricane Floyd yielded no evidence of any major events. this particular negative reaction in insurer stock price 5 One knot is approximately equivalent to 1.15 mph or returns. 0.5144 m sϪ1. By 1200 EDT on Friday, 10 September, tropical

Unauthenticated | Downloaded 09/29/21 04:12 AM UTC 402 WEATHER AND FORECASTING VOLUME 21

TABLE 3. Floyd-related news reports over the synoptic life cycle. Reports obtained from LexisNexis searches using combinations of the following words/phrases: hurricane, tropical storm, tropical depression, Floyd, and insurance. In total, the LexisNexis search engine retrieved more than 1000 documents.

Date Source Summary of news release Tuesday, 7 September The Associated Press National Weather Service reports the U.S. Virgin Islands will see some pass- ing showers and thunderstorms from an approaching tropical wave. NWS expects winds of 20 kt or less in the northeastern Caribbean. Wednesday, 8 September Miami Herald A weather system in mid-Atlantic could become Tropical Storm Floyd and threaten the Virgin Islands. Chicago Sun-Times Floyd is about 755 miles east of the Leeward Islands moving west-northwest. Thursday, 9 September Miami Herald Tropical Storm Floyd is expected to become a hurricane, gaining strength and heading westward. Friday, 10 September The Associated Press Floyd may hit the U.S. mainland as a major hurricane next week. Director of National Hurricane Center says “It’s in a position where it could hit almost anywhere.” Saturday, 11 September The Times-Picayune Headline reads, “Florida residents told to watch Floyd closely, prepare over weekend.” Floyd about 1300 miles away on Friday. Expected to develop winds of 120 mph by Monday. Deputy director of National Hurricane Cen- ter forecasting the hurricane “to become much stronger” and “headed in the direction of the southeast US coast.” Sunday, 12 September Jupiter Courier Forecasters at National Hurricane Center say Floyd could threaten the U.S. some time late next week. Described geographic location of storm and winds of 80 mph expected to top 110 mph by Monday. However, a major trough approaching from the east may push Floyd north, away from the coast. Monday, 13 September The Times-Picayune Floyd became category-4 hurricane Sunday night. Florida State Emergency Management Director says “storm is going to be so large, that there could be a very, very big evacuation.” Floyd measures over 450 miles in diameter adhering “to a trajectory that would bring its central core to the central Bahamas” by Tuesday or Wednesday. Tuesday, 14 September CCNfn News anchor to president of Insurance Information Institute: “this hurricane could actually be worse than Hurricane Andrew. What did that cost insur- ers?” Response: “insured damage. . . was approximately $16 billion.” CNN Correspondent: “Hurricane Floyd is giving a 155 mph whip- ping. . . and it’s got Florida in its sights.” Located about 275 miles east- southeast of Miami and “roaring” for Florida at 14 mph. “Biggest concern” may be Georgia and Carolinas. Wednesday, 15 September CNNfn CNNfn anchor says, “Storms like Hurricane Floyd can generate some stag- gering insurance claims, and that can change the bottom line of insurance companies.” Several insurance analysts interviewed throughout the day dis- cussing reinsurance, homeowner claims, and estimates of insured damages ranging from $4 to $10 billion. Reports a sell off of insurance company stocks based on fears of Floyd. Analyst says that insurance industry cash flows are much weaker than they were around the time of Andrew (1992). FEMA director announces emergency declaration for South and North Carolina. Thursday, 16 September USA TODAY Storm lessens. Chief economist of SunTrust Banks discusses how hurricanes provide an economic boost. Friday, 17 September CNNfn Insurance analyst discusses impact on insurance company stock prices and notes that some “were not impacted that much.” Discusses possible future insurance rate increases.

storm Floyd had been officially upgraded to category-1 over, the storm continued its pronounced turn toward hurricane status. The event variable D4 is found to be the U.S. coast. Attempting to describe the likely path of negative and significant. The coefficient estimate is the storm, the director of the NHC said, “It’sinapo- rather large at about Ϫ1.97%, again indicating a strong sition where it could hit almost anywhere.” (Table 3) In response in insurer stock prices. The Associated Press addition, it is entirely feasible that the business calen- reported that Floyd might hit the U.S. mainland as a dar also played a role in influencing the market’s re- major hurricane in the week to follow (Table 3). More- sponse. Because long positions in common stock of in-

Unauthenticated | Downloaded 09/29/21 04:12 AM UTC JUNE 2006 E W I N G E T A L . 403 surance companies on Friday could not be reversed un- Florida; Savannah, Georgia; Charleston and Myrtle til the following Monday, the threat was viewed as Beach South Carolina; Wilmington, Morehead City, being of a greater significance from a financial perspec- and Cape Hatteras, North Carolina; Norfolk, Virginia; tive. Ocean City, Maryland; Atlantic City, New Jersey; and The storm quickly grew in intensity, going from cat- New York City, New York. Floyd was now very close to egory-1 to category-3 status (with maximum sustainable the south Florida coast and heading up the Atlantic winds of around 110 kt (about 130 mph; 1 mph ϭ 0.4470 seaboard. Over these 2 days, CNNfn ran a number of msϪ1) over the weekend. The storm was near Cuba stories on Floyd. On Tuesday morning, a CNNfn news and getting closer to the United States. The Times- anchor asked the president of the Insurance Informa- Picayune (New Orleans, Louisiana) and the Jupiter tion Institute if “this hurricane could actually be worse Courier (Florida) carried stories warning Florida resi- than Hurricane Andrew. What did that cost insurers?” dents about the impending storm. The deputy director His response was that “insured damage. . . was approxi- of the NHC was forecasting the hurricane “to become mately $16 billion.” Later that morning, a CNN corre- stronger” and that it was “headed to the southeast US spondent reported that “Hurricane Floyd is giving the coast.” Winds were expected to reach 105 kt (120 mph; Bahamas a 155 mph whipping. . . and it’s got Florida in category 3). The NHC released statements saying that its sights.” The storm was reported to be about 275 Floyd could threaten the United States some time late miles east-southeast of Miami and “roaring” for Florida the following week, although a major trough approach- at 14 mph. CNN noted that the “biggest concern” might ing from the east might push Floyd north, away from be for Georgia and the Carolinas, because the storm the coast. (Table 3) By the market’s opening on Mon- would be turning up the coast. (Table 3) On Wednes- day, 13 September, Floyd had become a fierce cat- day, more reports about the storm aired on CNNfn. A egory-4 hurricane as reported in The Times-Picayune news anchor discussed how Floyd could lead to “stag- (Table 3). Interestingly, the event variable D5 was ac- gering insurance claims” that may damage the line of tually positive and significant, albeit with a relatively insurance companies. Several insurance analysts were small increase. The Times-Picayune noted that Floyd interviewed throughout the day, discussing the topics of was over 450 miles in diameter. In fact, Florida’s state reinsurance and homeowner claims, and giving esti- emergency management director remarked that this mates of insured damage ranging from $4 to $10 billion. “storm is going to be so large that there could be a very, It was also reported that there was a sell off of insur- very big evacuation” (Table 3). ance company stocks based on fears of Floyd. One ana- The event variables D6 and D7 (corresponding to lyst noted that insurance industry cash flows were much Tuesday, 14 September and Wednesday, 15 September) weaker than around the time of Andrew. The Federal 6 were both found to be negative and significant. The Emergency Management Agency (FEMA) declared NHC advisory provides the probability that a hurricane emergencies in South and North Carolina. (Table 3) will pass close to a location 24, 48, and 72 h out from the Finally, beginning late Wednesday afternoon (about time of the advisory. The purpose of issuing the hurri- 1500 EDT) the storm was downgraded to category 3, cane probabilities is to provide “guidance in hurricane with maximum sustainable winds of less than 110 kt. At protection planning by government and disaster offi- the market opening on Thursday, 16 September, Floyd cials” (National Hurricane Center 1999). Hurricane had lost much of its intensity and had become a cat- Probability Advisory 28 issued at 1100 EDT 14 Sep- egory-2 hurricane. By 1300 EDT Floyd had weakened tember 1999 indicated a probability of greater than to category-1 status, with winds in the range of 70–80 kt 10% that the center of the hurricane would pass within (80–95 mph). As the storm’s intensity diminished, the 65 nautical miles of the following cities within the next media turned its attention away from the immediate 72 h: Miami, West Palm Beach, Ft. Pierce, Cocoa damage and more to the longer-term impact of the Beach, Daytona Beach, Jacksonville, Marco Island, Ft. storm. USA TODAY ran a story in which the chief Myers, Venice, Tampa, Cedar Key, and St. Marks economist of SunTrust Banks discussed how hurricanes provide an economic boom (Table 3). During the night Floyd was again downgraded and was a tropical storm 6 We reran the analysis including a set of day-of-the-week by the beginning of Friday’s trading. Both D8 and D9 dummy variables, but none of these were statistically significant at were estimated to be positive and significant. Consis- the 5% level. The coefficient on D6 was only marginally signifi- ϭ tent with this finding, an insurance analyst interviewed cant (p 0.11) and the coefficient on D9 was insignificant in controlling for the day of the week. These results are available on on CNNfn about the impact of Floyd concluded that request. some insurance companies’ stock prices “were not im-

Unauthenticated | Downloaded 09/29/21 04:12 AM UTC 404 WEATHER AND FORECASTING VOLUME 21 pacted that much.” He then went on to discuss a pos- fall near New Orleans, Louisiana. Lamb (1998) and sible future rate increase. Interestingly, if one used the Angbazo and Narayanan (1996) also looked at the ef- traditional approach to analyze the financial impact of fects of Andrew, finding evidence that insurance com- Hurricane Floyd, by focusing on when it made landfall, pany stock prices fell when the storm made landfall. the D8 and D9 coefficients would have indicated that Lamb focused on property and casualty insurer stock the hurricane had a positive impact on stock returns for returns for firms with and without Florida and Louisi- insurance companies. ana exposure, finding the former adversely effected It is also interesting to note that Hurricane Floyd had whereas the latter showed no impact. While he found a major impact on insurer stock prices as evidenced by that Andrew had a significant adverse impact on ex- the negative and significant CAR shown in Table 2b. posed property and casualty insurance stock returns The CAR shows the estimate of the total effect of the when Andrew made landfall, he does not consider news regarding the hurricane as it approached the U.S. these stock returns prior to making landfall. Angbazo shoreline and made landfall. Our results indicate that and Narayanan (1996) similarly do not emphasize the the storm resulted in lower-than-normal returns, that is, days prior to the hurricane making landfall, but rather on a risk-adjusted basis, than would have been the case focus on the days after. had the storm not occurred. The 10-day CAR for Floyd Our evidence on Floyd indicated that information is estimated to be about 2%, a value that is very close describing the development of the storm over time and to the estimate obtained by Angbazo and Narayanan space could help explain how the financial market re- (1996) when examining the effects of Hurricane An- acts to changing news about a storm’s characteristics. drew. Thus, we conclude that Hurricane Floyd had a Thus, we conducted another event study, similar to that negative impact on insurer stock prices, not the positive of Floyd, but corresponding to the synoptic life cycle of effect that was seen when it finally made landfall. The Andrew (17–28 August 1992). In our examination of evidence is consistent with the hypothesis that the NWS Andrew we once again use the S&P Insurer Stock Price and the NHC were providing valuable information to Index, which includes insurance firms outside of the financial markets regarding both the path and the se- property and casualty insurance firms, so our findings verity of the storm. are not directly comparable to those of Lamb (1998) Generally speaking, the response of insurer stock and Angbazo and Narayanan (1996). The sample pe- prices to the development of Floyd in both time and riod was chosen so as to have the longest period pos- space, and in terms of characteristics describing the sible without overlapping into other hurricanes and storm intensity, is remarkably in line with models of contained 479 observations. Andrew was the only hur- investor behavior. Information, in all its varied forms, is acted on quickly and assimilated into daily price ricane to make landfall in the United States during this changes. Our results indicate that storm characteristics period. Equation (3) was estimated with the dichoto- can have statistical explanatory power with respect to mous variables corresponding to the synoptic life cycle the movement in insurer stock prices. To the extent that days in which the stock market was open, that is, 18–21 the evidence indicates a significant adverse market re- and 24–28 August. Otherwise, the estimation was un- action for insurance stock prices as the storm was ap- dertaken in the same fashion that was used for analyz- proaching landfall, the findings in this paper are con- ing the effects of Floyd. sistent with Cummins et al. (2004) who find evidence Table 4 presents the results obtained from estimating indicating that catastrophic loss index options are not the synoptic life cycle event study for Andrew. Com- fully effective in hedging insurance companies. paring the results from Floyd (occurred 7 yr after An- drew) with those of Andrew, the Andrew model does not explain nearly as much of the variance of insurer 5. A comparison to Hurricane Andrew stock returns as the Floyd model. The adjusted R2 is only 3% in the Andrew model compared with 41% in As the final task in our analysis we examined the the Floyd model. Another difference is that the esti- response of insurer returns to Hurricane Andrew and mated coefficient on market returns (beta) is 0.95 in the compare the results with those found for Floyd. In Au- Floyd model and just 0.31 in the Andrew model, sug- gust 1992, the United States was subject to perhaps the gesting that the systematic risk was much lower during most destructive hurricane to ever hit the mainland. the earlier time period associated with Hurricane An- Hurricane Andrew caused more than $25 billion in drew. These basic differences between the two models damages. The hurricane hit the south Florida coast be- may indicate that 1) returns of insurance firms have fore continuing into the Gulf where it made final land- become more sensitive to movements in the market as

Unauthenticated | Downloaded 09/29/21 04:12 AM UTC JUNE 2006 E W I N G E T A L . 405

TABLE 4. Hurricane Andrew estimation results for S&P Insurer Stock Price return Rt model. (a) The event study model was corrected for autoregressive conditional heteroscedasticity in order to obtain more efficient parameter estimates. Quasi-maximum likelihood covariances and standard errors are computed as described in Bollerslev and Wooldridge (1992). Day-of-the week control variables were included. (b) The test statistic of Ϫ2.60 (standard normal distribution) that is reported is significant at the 5% level.

(a) Synoptic life cycle event study Central Wind Coefficient Std error Z statistic Probability pressure (mb) speed (kt) Constant Ϫ0.000 794 0.000 913 Ϫ0.869 359 0.3847

Rmt 0.312 676 0.062 931 4.968 589 0.0000 Ϫ Ϫ D1 (17 Aug 1992) 0.006 046 0.001 016 5.948 308 0.0000 1006 35 Ϫ Ϫ D2 (18 Aug 1992) 0.006 071 0.000 890 6.823 119 0.0000 1000 45 Ϫ Ϫ D3 (19 Aug 1992) 0.020 249 0.000 810 24.991 33 0.0000 1005 45 Ϫ Ϫ D4 (20 Aug 1992) 0.000 577 0.001 078 0.535 615 0.5922 1015 40 D5 (21 Aug 1992) 0.000 224 0.000 913 0.244 859 0.8066 1007 50 Ϫ Ϫ D6 (24 Aug 1992) landfall in Florida 0.008 140 0.001 128 7.219 184 0.0000 951 115 Ϫ Ϫ D7 (25 Aug 1992) 0.006 278 0.001 021 6.150 707 0.0000 946 120 D8 (26 Aug 1992) in landfall Louisiana 0.003 580 0.000 821 4.361 289 0.0000 973 80 Ϫ Ϫ D9 (27 Aug 1992) 0.005 682 0.001 028 5.526 006 0.0000 998 30 D10 (28 Aug 1992) 0.013 542 0.000 913 14.826 77 0.0000 1000 20 R2 0.082 108 Adjusted R2 0.046 190 (b) Cumulative effects analysis. 10-day CAR Ϫ0.027 18 t statistic Ϫ2.60

a whole and 2) movements in these returns are now ticeable increase in the magnitude of the estimated co- 7 more readily explained than they used to be. efficient on D3 (Wednesday, 19 August) when Andrew Table 5 gives vital statistics of Hurricane Andrew turned more directly toward the United States. Stock compiled from a preliminary report prepared by the returns fell by almost 2% points with the arrival of this NHC (Rappaport 1993). Information about the devel- news. This impact is estimated to be much larger than opment of the storm may be used to explain the day- the impact of the storm actually making landfall. On to-day movements in insurer stock returns during the Thursday, the storm’s trajectory had turned nearly due synoptic life cycle of Andrew. Figure 3 shows the storm north (19.8°N latitude, 59.30°W longitude) as if it might path over time. The reported D coefficients in Table 4 go back out to sea. The corresponding coefficient on provide estimates of the abnormal returns on that day D4, while negative, was statistically insignificant. An- 8 observed for the S&P Insurer’s stock price index. By drew literally picked up speed (winds now were in the 1800 EDT Sunday, 16 August, Andrew was declared a 50-kt range) on Friday, 21 August, and turned back tropical depression (wind speeds around 25 kt; central toward the United States, although the coefficient on pressure around 1010 mb; 1 mb ϭ 1 hPa) located far out D5 was insignificant. Major storm changes occurred in the Atlantic and near the equator. On the 17 and 18 over the weekend as Andrew went from hurricane cat- August, the estimated coefficients on D1 and D2 were egory-1 status at 0600 EDT Saturday to category 5 by negative and statistically significant and Andrew was 1200 EDT on Sunday. On Monday, 24 August, Andrew classified as a tropical storm. However, there is a no- moved across the south part of Florida and then con- tinued toward the Gulf Coast states on Tuesday. Con- sequently, on both Monday and Tuesday the insurer 7 While these results neither answer the question of whether Andrew was a seminal event to investors nor tell us if investors stock returns reacted negatively and significantly (see pay more attention to storms in the post-Andrew period, they do coefficients on D6 and D7). By Wednesday, Andrew point to a possible fundamental change in the insurance stock made its second U.S. landfall just west of New Orleans market model. The reasons for this change are left to future re- and then dissipated to category 1. As with landfall in search. the case of Floyd, the insurer returns actually re- 8 The reported results are for the analysis that included day-of- the-week dummy variables because both Monday and Wednesday sponded positively (see coefficient on D8). The next 2 (compared with the omitted day, Friday) were statistically signifi- days exhibited some volatility and cycling with the ab- cant. normal returns estimated to be negative on Thursday,

Unauthenticated | Downloaded 09/29/21 04:12 AM UTC 406 WEATHER AND FORECASTING VOLUME 21

TABLE 5. Summary descriptions of Hurricane Andrew. (Source: times the insured losses. However, the CAR estimates NHC) are for the period leading up to the storm making land- Synoptic life cycle 17–28 Aug 1992 fall and are not comparable to an after-the-event as- sessment. Remember, for example, at one point Floyd Estimate of total U.S. damage $25ϩ billion appeared to be headed directly for Miami and also the Number of deaths in United States 40 Maximum wind speed (Saffir–Simpson 150 kt (category 5) CNNfn comment that “this hurricane could actually be scale category) worse than Hurricane Andrew.” In addition, another Wind speed at landfall 120 kt reason that the impact of Floyd was so disproportion- Saffir–Simpson scale category at landfall Category 4 ately large could stem from the experience of Andrew Location of landfall* 29.2°N, 91.3°W itself, because state insurance commissions made it * Indicates first landfall. clear that Andrews’ losses were not an excuse to raise insurance premiums (see Angbazo and Narayanan 1996). With the increasing concentration of population

27 August (see the coefficient on D9), and positive on and property along vulnerable coastal regions, Andrew

Friday, 28 August (see the coefficient on D10)atthe was not necessarily seen as an extreme outlier but end of the synoptic life cycle. rather as a harbinger of future large-loss hurricanes. The results of using the storm and weather informa- Nine property casualty insurance companies became in- tion in the event study analysis of Andrew yielded re- solvent as a direct result of Hurricane Andrew, thus sults that were remarkably in line and consistent with raising the question of whether insurance policies those of Floyd. Both analyses imply that storm devel- would be available at all to homeowners in some areas opment and characteristics have explanatory power as (Orth 1998). predictors of movements in insurer stock prices around the synoptic life cycle of a hurricane prior to the hur- 6. Concluding remarks ricane making landfall. The CAR for Andrew is 25% larger than the CAR for Floyd. This might strike some This study has examined how information about hur- as relatively small because Andrew resulted in about 3 ricane Floyd affected insurer stock prices. For compari-

FIG. 3. Storm track of Hurricane Andrew. Points on the figure represent the pathofthe hurricane. Points on this figure correspond to the data presented in Table 5. The path can be differentiated by a variety of storm characteristics. (Source: Unisys)

Unauthenticated | Downloaded 09/29/21 04:12 AM UTC JUNE 2006 E W I N G E T A L . 407 son purposes, we also examined the effect of Hurricane mation and inference in dynamic models with time-varying Andrew on insurer stock prices. A unique feature of covariances. Econ. Rev., 11, 143–172. our analysis is the use of public information about the Campbell, J. Y., A. W. Lo, and A. C. MacKinlay, 1997: The Econometrics of Financial Markets. Princeton University storms’ characteristics that are available from the NWS, Press, 632 pp. NHC, and other media sources. We use this informa- Cummins, J., D. LaLonde, and R. D. Phillips, 2004: The basis risk tion as explanatory variables to predict changes in in- of catastrophic-loss index securities. J. Financ. Econ., 71, 77– surer stock prices. It is argued that the development of 111. the storm over time and space is an important feature Dasgupta, D., 1999: Hurricane Floyd’s toll rises to $1.8 billion, up $500 million from initial estimate, in ISO’s resurvey of insur- that has been left out of previous hurricane event stud- ers’ losses. Insurance Services Office, Inc., Press Release, 6 ies. Each storm has unique characteristics, therefore December. knowledge on how a storm is evolving, in terms of vari- Ewing, B. T., and J. B. Kruse, 2001: Hurricane Bertha and unem- ables such as wind speed, storm direction, and time, ployment: A case study of Wilmington, NC. Proceedings of provides unique insight into how the market perceives the Americas Conference on Wind Engineering, CD-ROM. the market value of insurers. Overall, there is a nega- ——, and ——, 2002: The impact of Project Impact on the Wil- mington, NC labor market. Public Finance Rev., 30, 296–309. tive effect on insurer stock price changes around the Johnson, S. R., and M. T. Holt, 1997: The value of weather infor- synoptic life cycle of the storm; however, this effect is mation. Economic Value of Weather and Climate Forecasts, neither constant nor is it always negative on each day of R. W. Katz and A. H. Murphy, Eds., Cambridge University the cycle. The market response is as dynamic as the Press, 75–103. synoptic conditions associated with the storm. We find Lamb, R. P., 1995: An exposure-based analysis of property- liability insurer stock values around Hurricane Andrew. J. significant market reaction to the news concerning the Risk Insur., 62, 111–123. path and strength of the storm, which indicates the ——, 1998: An examination of market efficiency around hurri- value of information such as that provided by the NWS canes. Financ. Rev., 33, 163–172. and NHC to be valuable. The findings add to our un- Liu, P., S. Smith, and A. Syed, 1992: The impact of the insider derstanding of how and to what extent windstorm trading scandal on the information content of the Wall Street events affect the market performance of insurance Journal’s “Heard on the Street” column. J. Financ. Res., 15, 181–193. firms and may lead to improvements in risk manage- MacKinlay, A. C., 1997: Event studies in economics and finance. ment strategies. J. Econ. Lit., 35, 13–39. National Hurricane Center, 1999: Hurricane Floyd probabilities Acknowledgments. This work was performed under number 28 (11 am EDT September 14, 1999). National the Department of Commerce NIST/TTU Cooperative Weather Service, Miami, FL. Agreement Award 70NANB8H0059. We thank Robert Orth, R. J., 1998: Hurricane insurance protection in Florida. Pay- ing the Price: The Status and Role of Insurance against Natu- E. Chapman, Economist, Office of Applied Economics, ral Disasters in the United States, E. Lecomte and K. Ga- BFRL, NIST for his valuable input. The authors thank hagan, Eds., John Henry Press, 97–125. participants of the Texas Tech University Finance Pasch, R. J., and T. B. Kimberlain, and S. R. Stewart, cited 1999: Workshop for comments on an earlier draft of the pa- Hurricane Floyd: 7–17 September, 1999. National Hurricane per. We especially acknowledge the helpful comments Center Preliminary Rep. [Available online at http:// www.nhc.noaa.gov/1999floyd_text.htm.] of Harold Brooks, and three anonymous referees, John Rappaport, E. N., cited 1993: Hurricane Andrew. National Hur- Adams (who also provided research assistance), Don ricane Center Preliminary Rep. [Available online at ftp:// Bowlin, Jack Cooney, Ken Cyree, Phil English, Steve ftp.nhc.noaa.gov/pub/storm_archives/atlantic/prelimat/ Sears, Ray Spudeck, and Jian Yang. atl1992/andrew/.] Roll, R., 1984: Orange juice and weather. Amer. Econ. Rev., 74, 861–880. REFERENCES Rozeff, M., and M. Zaman, 1998: Overreaction and insider trad- Angbazo, L. A., and R. Narayanan, 1996: Catastrophic shocks in ing: Evidence from growth and value portfolios. J. Finance, the property-liability insurance industry: Evidence on regu- 53, 701–716. latory and contagion effects. J. Risk Insur., 63, 619–637. West, C. T., and D. G. Lenze, 1994: Modeling the regional impact Bollerslev, T., 1986: Generalized autoregressive conditional het- of natural disaster and recovery: A general framework and an eroskedasticity. J. Econ., 31, 307–327. application to Hurricane Andrew. Int. Reg. Sci. Rev., 17, 121– ——, and J. Wooldridge, 1992: Quasi-maximum likelihood esti- 150.

Unauthenticated | Downloaded 09/29/21 04:12 AM UTC