Risk Selection and Moral Hazard in Natural Disaster Insurance Markets: Empirical evidence from Germany and the United States

Transcription

1 Risk Selection and Moral Hazard in Natural Disaster Insurance Markets: Empirical evidence from Germany and the United States Paul Hudson Department of Environmental Economics, Institute for Environmental Studies, VU University Amsterdam, the Netherland W.J. Wouter Botzen Department of Environmental Economics, Institute for Environmental Studies, VU University Amsterdam, the Netherlands. Jeffrey Czajkowski Wharton School Risk Center University of Pennsylvania Heidi Kreibich GFZ German Research Centre for Geosciences, Potsdam, Germany October 2014 Working Paper # Risk Management and Decision Processes Center The Wharton School, University of Pennsylvania 3730 Walnut Street, Jon Huntsman Hall, Suite 500 Philadelphia, PA, USA Phone: Fax:

2 THE WHARTON RISK MANAGEMENT AND DECISION PROCESSES CENTER Established in 1984, the Wharton Risk Management and Decision Processes Center develops and promotes effective corporate and public policies for low-probability events with potentially catastrophic consequences through the integration of risk assessment, and risk perception with risk management strategies. Natural disasters, technological hazards, and national and international security issues (e.g., terrorism risk insurance markets, protection of critical infrastructure, global security) are among the extreme events that are the focus of the Center s research. The Risk Center s neutrality allows it to undertake large-scale projects in conjunction with other researchers and organizations in the public and private sectors. Building on the disciplines of economics, decision sciences, finance, insurance, marketing and psychology, the Center supports and undertakes field and experimental studies of risk and uncertainty to better understand how individuals and organizations make choices under conditions of risk and uncertainty. Risk Center research also investigates the effectiveness of strategies such as risk communication, information sharing, incentive systems, insurance, regulation and public-private collaborations at a national and international scale. From these findings, the Wharton Risk Center s research team over 50 faculty, fellows and doctoral students is able to design new approaches to enable individuals and organizations to make better decisions regarding risk under various regulatory and market conditions. The Center is also concerned with training leading decision makers. It actively engages multiple viewpoints, including top-level representatives from industry, government, international organizations, interest groups and academics through its research and policy publications, and through sponsored seminars, roundtables and forums. More information is available at

3 Risk Selection and Moral Hazard in Natural Disaster Insurance Markets: Empirical evidence from Germany and the United States Paul Hudson 1, W.J. Wouter Botzen 1, Jeffrey Czajkowski 2, Heidi Kreibich 3 1 Department of Environmental Economics, Institute for Environmental Studies, VU University Amsterdam, the Netherlands. 2 Wharton Risk Management and Decision Processes Center, University of Pennsylvania, U.S. 3 GFZ German Research Centre for Geosciences, Potsdam, Germany. Abstract Adverse selection and moral hazard are commonly expected to cause market failures in natural disaster insurance markets. However, such problems may not arise if individuals mainly buy insurance based on risk preferences. Advantageous selection can occur if individuals with insurance are highly risk averse and seek to reduce risk. We offer a comprehensive empirical study of risk selection in natural disaster insurance markets and whether disaster preparedness activities differ when people have natural disaster coverage. Statistical analyses are based on survey data of individual disaster insurance purchases and risk mitigation activities in Germany and the United States. Consistent results are obtained in both countries supporting advantageous selection. This has significant potential public policy relevance regarding the effectiveness of their respective existing market structures for natural disaster insurance. Keywords: Deductible, Flood insurance, Moral hazard, Natural disaster insurance, Private marketization, Risk selection. JEL codes: G22, Q54. 1

4 1. Introduction Over the last decades economic damages from natural disasters, and floods in particular, have been increasing and this trend is likely to continue (IPCC, 2012; Munich Re, 2013). Insurance could play an important role in managing natural hazard risks and promoting recovery from disasters, because insurance reduces financial risks for households by spreading risks over many policyholders, and helps people to get back on their feet after a disaster occurs (Botzen, 2013). Insurance can also provide incentives for risk reduction by charging premiums that act as a price signal of risk, or by providing premium discounts to policyholders who protect their property against disaster damage (Kunreuther, 1996). On the other hand, insurance coverage may result in moral hazard if insured individuals take fewer measures to limit risk if they expect that insurers will compensate their damage irrespective of their mitigation efforts (Ehrlich and Becker, 1972; Arnott and Stiglitz, 1988). This poses problems if the behavior leading to moral hazard cannot be observed by the insurer, meaning that increased risk taking is not completely reflected in a higher insurance premium (Chiappori and Salanie, 2000). Moreover, adverse selection may obstruct the adequate functioning of natural disaster insurance markets i if mainly individuals who face a high risk demand insurance, and insurers do not adequately factor such risks into higher prices because of information asymmetries between the insurer and the insured (Akerlof, 1970; Rothschild and Stiglitz, 1976) ii. Here we undertake an empirical analysis of field survey data to assess whether the proper functioning of natural disaster insurance markets in the United States and Germany is occurring through the avoidance of (adverse) risk selection and moral hazard behavior. Cohen and Siegelman (2010) provide a comprehensive review of empirical studies of adverse selection or moral hazard effects in the following insurance markets: automobile, mortality risk, long-term care, crop, and health insurance, finding mixed results across these markets. Cohen and Siegelman (2010) conclude that whether or not adverse selection and moral hazard effects arise importantly depends on characteristics of the individual insurance market. In particular, adverse selection may not be observed due to institutional and regulatory frameworks and when private information about risk is in the possession of some but not all policyholders, policyholders fail to use the private information they have, or insurers have a superior ability to assess risk. Moreover, adverse selection and moral hazard may not be an issue if insurance purchase decisions are mostly driven by risk aversion, and if highly risk averse people who purchase insurance also take other precautionary measures which limit risk (de Meza and Webb, 2001; Cohen and Siegelman, 2010). The examination of such effects have been extensively researched for health insurance markets. Several of these studies show that adverse selection is present (Sloan and Norton, 1997; Finkelstein et al. 2005; Courbage and Roudaut, 2008), although there are studies that argue the opposite (Cardon and Hendel, 2001; Finkelstein and McGarry, 2006; Cutler et al., 2008; Einav et al., 2013). For example, Finkelstein and McGarry (2006) arrive at an opposite finding to that of a moral hazard effect since individuals with health insurance in the U.S. take more measures to reduce health risks than uninsured individuals, which may be explained by risk aversion. Likewise, in their investigation of life, acute health, annuities, long-term care, and Medicare supplemental insurance markets, Cutler et al. (2008) find that those individuals who engage in less risk reducing behavior (i.e., moral hazard) are less likely to have each of these insurance types. Our focus on natural disaster insurance is of interest since to the best of our knowledge a systematic empirical analysis of the presence of (aspects) of adverse selection and moral hazard regarding natural hazard insurance has not been carried out. Moreover, as noted above it is not evident that adverse selection and moral hazard problems would arise in this market context. As 2

5 Cohen and Siegelman, (2010) point out, these problems may not occur if a policyholder lacks an informational advantage about risk and/or does not use, or act, upon it. The latter, for example, occurs when individuals do not select into buying insurance on the basis of risk, but instead on risk attitudes, meaning that advantageous selection effects dominate. Such behavioral aspects are important since adverse selection may not occur if individuals are characterized by bounded rationality and misperceive the risk that they face (Kunreuther and Pauly, 2004). In other words, it is not necessarily the case that individuals facing a high objective natural disaster risk have a high risk perception (Botzen et al., 2009) or a high demand for natural disaster insurance coverage (Botzen and van den Bergh 2012a,b). In general, much evidence shows that individuals have difficulties with assessing low-probability/high-impact risks with which individuals have few experiences (Kunreuther et al., 2001) such as natural disasters. This may translate into poor decision making with respect to natural disaster insurance purchases (Kunreuther et al., 2013), which possibly prevents adverse selection from arising. It is an empirical question in markets for natural disaster insurance whether moral hazard and adverse selection effects dominate advantageous selection effects based on preferences, like risk aversion, and potential risk-reducing incentives provided by insurance. We examine the behavioral conditions that may lead to adverse selection and moral hazard by presenting an analysis of how flood insurance purchases relate with experiences of flood damage and flood risk mitigation behavior in Germany and in the U.S (Cutler et al., 2008). We further compliment this analysis by applying differing statistical models to Germany and the U.S. For Germany, we examine whether people who take out flood insurance face a higher flood risk. Moreover, we estimate whether after correcting for risk selection, individuals who have flood insurance experience more flood damage than uninsured individuals, which can be related to moral hazard when insured individuals take fewer flood damage mitigation measures. The absence or presence of moral hazard in Germany may be related to risk preferences of individuals, or the deductible on the flood insurance policy for which we cannot control directly. However, we expect the influence of the deductible to be small. This is confirmed by our complementary analysis for the U.S. which shows that the deductible has a mostly negligible impact on policyholders risk mitigation activities except at the highest deductible levels, levels typically higher than those in Germany. Furthermore for the U.S. it is examined whether people who have taken out separate homeowners insurance (which covers wind damage only) and flood insurance policies are more, or less likely, to take other measures that limit hurricane damage. The approach and data used in our study differs from most previous related research (i.e. those discussed in Cohen and Siegelman, 2010), in that we do not explicitly study information asymmetries between the policyholder and the insurance company. Instead we make use of field survey data of individual insurance purchases, risks, and risk mitigation activities to examine whether the behavior of policyholders is consistent with risk selection that may lead to adverse selection or moral hazard. Our study makes use of unique data sets about individual risk mitigation activities which are not available to insurance companies. This aspect of information asymmetry may lead to negative risk selection and behavioral changes if insurance companies do not charge fully risk based premiums, as is the reality in our case study areas. Understanding the behavior that may lead to moral hazard and adverse selection effects in the market for natural disaster insurance has important public policy relevance in Germany and the U.S. In Germany it has been argued that adverse selection is one of the reasons for the observed low (5-10%) market penetration of flood insurance, which has resulted in calls for introducing mandatory flood insurance coverage (Schwarze and Wagner, 2007; Seifert et al., 2013). In the 3

6 U.S., where flood insurance is primarily a publically underwritten insurance vehicle through the National Flood Insurance Program (NFIP), there have been recent calls for reform including more private market involvement (Michel-Kerjan and Kunreuther, 2011). Adverse selection would be a deterrent in this regard. Moreover, the movement toward risk-based premiums as a part of the recent flood insurance reform acts iii is aimed at providing incentives for mitigation for which it is relevant to know how far insurance acts as a mitigation disincentive (moral hazard). In the U.S. homeowners insurance market significant wind and hurricane deductibles are a part of a homeowner s insurance contract to help avoid potential moral hazard. However, such deductibles may substantially lower the attractiveness of the insurance for consumers, and whether these can be applied in recent major events, such has Hurricane Irene and Sandy, has been a contestable legal issue (Pomerantz and Suglia, 2013). It is, therefore, of interest to examine whether moral hazard is a major issue in the U.S. natural disaster insurance market, and whether deductibles are effective overall in stimulating policyholders to mitigate risks, as will be studied here. The remainder of this article is structured as follows. Section 2 provides information on the German and U.S. natural disaster insurance markets. Section 3 describes the methods and data used. Section 4 presents the results about adverse selection and moral hazard in Germany and the U.S. Section 5 concludes. 2. Natural disaster insurance markets in Germany and the U.S. Flood insurance market in Germany The German flood insurance market is based on free market provision (Keskitatalo et al., 2014). In addition to flood insurance payouts, government compensation and public donations play an important role in compensating flood damage; which may limit the insurance market by the presence of charity hazard. Flood insurance is provided as a bundled coverage with other natural hazard risks as a supplement to regular building or contents insurance (Keskitatalo et al., 2014; Seifert et al., 2013). The premium of the flood insurance supplement s contribution to the natural hazard bundle is to a certain extent differentiated on the basis of the flood probability. This is done using the Zürs flood zoning system (GDV, 2008). Zürs produces 4 zones of flood probabilities ranging from 1 (less than 1/200 years chance of flooding) to zone 4 (greater than 1/10 years chance of flooding). Moving from zones 1-4 entails an increase in premiums and deductibles (Seifert et al., 2013). The majority of households are located in zone 1, 10-12% are in zone 2, while 3% of households live in zones 3 and 4 (GDV, 2008). The penetration rate of flood insurance increased strongly in recent years. About 10 years ago it was estimated to be between 3-10% (GDV, 2003, Landtag Rheinland-Pfalz, 2005; Bogenrieder, 2004) and it is now estimated to be 19% and 33% for contents and residential buildings, respectively (GDV 2013). The national average hides large regional differences of penetration rates in Germany (Seifert et al., 2013). For instance, 95% of households are estimated to have flood insurance in Baden-Württemberg, while this is only 11% in Bremen (Keskitatalo et al., 2014). Overall, East Germany is estimated to have higher penetration rates than West Germany, due to a history of compulsory flood insurance in the East. Thieken et al. (2006) conducted surveys of German insurance companies and households in flood-prone areas, in 2002, in order to examine characteristics of flood insurance policy conditions in Germany, and whether flood insurance provides incentives for risk mitigation. This 4

7 survey revealed that deductibles were not dependent on the Zürs zoning system. Thieken et al. (2006) found that flood insurance deductibles in Germany ranged in between 500 and 5,000. These deductibles provide a small incentive for taking damage mitigation measures: namely between 5 and 50 in areas with a flood probability of 1/100. Deductibles and premiums were also not found to be dependent on flood risk mitigation measures implemented by policyholders. Windstorm and flood insurance in the U.S. In the U.S, a standard multi-peril homeowners insurance policy is normally required as a condition for a mortgage. These policies cover damage from fire, wind, hail, lightning and winter storms, among other common non-catastrophe perils (Czajkowski et al, 2012). Although catastrophe perils are covered in the standard homeowner insurance policy, in highly hazard prone areas of the U.S. some of these perils are subjected to separate deductibles that are not a typical nominal dollar deductible amount, but rather are a percentage of the insured value of the home. For example, both hurricane deductibles and more general windstorm deductibles are characteristic in hurricane and wind prone areas of the U.S. with percentage deductibles that generally vary from one to five percent of a home's insured value (Insurance Information Institute, 2014). However, in areas with especially high wind or hurricane risk, these deductible percentages may be as high as fifteen percent (Insurance Information Institute, 2014). Nineteen states in the U.S. have hurricane deductibles including the states of Alabama, Delaware, Louisiana, Maryland, Mississippi, New Jersey, New York, North Carolina, and Virginia where our U.S. survey respondents are situated (Insurance Information Institute, 2014). While standard U.S. homeowners insurance covers a number of catastrophe perils, coverage for flood damage resulting from rising water is explicitly excluded in homeowners insurance policies (Michel-Kerjan et al., 2014). Since 1968 the National Flood Insurance Program (NFIP), administered by the U.S. Federal Emergency Management Agency (FEMA), has been the primary source of residential flood insurance in the U.S. (Michel-Kerjan 2010, Michel-Kerjan and Kunreuther 2011). The NFIP was developed in 1968 because ever since the severe Mississippi floods of 1927, the private insurance industry believed flood risk was uninsurable due to adverse selection, the possibility of massive losses, and the inability to correctly price the product stemming from the level of sophistication in hazard assessment in the 1960s (Michel-Kerjan et al., 2014). As of January 1, 2014, there were 5.47 million NFIP policies-in-force nationwide which generated US$3.53 billion in premiums for a total of US$1.28 trillion under coverage. The average annual premium per policy is US$645 nationwide. The NFIP provides insurance up to a maximum limit for residential property damage, now set at US$250,000 for building coverage and US$100,000 on contents coverage. Some additional coverage is offered by private insurers above these limits for residential property owners, although this represents a small percentage of total residential coverage, namely less than 5 per cent approximately (Michel-Kerjan et al., 2014). To set premiums and support local governments, the NFIP maps participating communities, designating flood risks through different flood zones on the Flood Insurance Rate Maps (FIRMs) (Michel-Kerjan et al., 2014). A building that was in place before the mapping of flood risk was completed in that area is often given subsidized rates, while homes built after the risk mapping are charged premiums reflecting FEMA s flood maps. Around a quarter of properties are still subsidized today (Michel-Kerjan et al., 2014). Premiums are determined using the Actuarial Rate Formula which is focused on the high-risk A and V 100 year flood zones, i.e., areas with a 1 per cent or greater annual chance of flooding and coastal areas with a 1 per cent or greater 5

8 annual chance of flooding and an additional hazard associated with storm waves respectively (Michel-Kerjan et al., 2014). Federal law requires property owners in these 100-year floodplains with a mortgage from a federally backed or regulated lender to purchase flood insurance. Despite the mandatory purchase requirement, due to a lack of proper enforcement take-up rates are typically low (50 percent or less), especially in non-coastal areas (Dixon et al, 2006; Czajkowski et al, 2012). Although take-up rates due vary substantially depending upon location (Dixon et al., 2006). FEMA rates also vary depending on the elevation of the first floor of the dwelling in relation to the 100-year return flood event. However, FEMA does not collect elevation information for many of the insured houses (Michel-Kerjan et al., 2014). Michel-Kerjan et al. (2014) show that the NFIP s overall pricing strategy leads to important divergences from the true risk for a number of residents covered by the program. Rates are not risk-based at the individual level (probabilistically defined), so prices might be too high in some areas and too low in others. The NFIP offers deductibles ranging between $500-$5000. Michel-Kerjan and Kousky (2010) find that 97 percent of NFIP policy-holders choose deductible levels of $1000 or less. Finally, to encourage mitigation, the NFIP operates the Community Rating System (CRS), a voluntary program that rewards communities that undertake mitigating activities with premiums discounts depending on the level of actions taken. Approximately 1200 of the over 21,000 NFIP communities actively participate in the CRS program. However, the mitigation emphasis of the CRS program is at the community level, not the individual policy-holder. All told, characteristics of both countries natural disaster insurance markets may or may not indicate the presence of adverse selection or moral hazard. For example, low market penetration rates combined with non-risk-based (subsidized) premiums may indicate an increased likelihood of adverse selection. Similarly, low deductible choices by the insured, or lack of knowledge pertaining to separate high wind deductible levels, may lead to moral hazard behavior. On the other hand, penetration rates vary significantly based upon location despite the lack of enforcement signaling varying risk attitudes and levels of risk aversion. Or rates may be risk-based enough, as well as deductible levels high enough, such that moral hazard behavior is muted. We aim to further investigate these possibilities through our field data. 3. Methods and data Survey data The German data are obtained from surveys carried out in the Elbe and Danube river catchment areas in response to flood events occurring in 2002, 2005 and The sample population was collated by using official data to collect all of the streets that suffered from a flood. The sample population was refined into the experimental sample by drawing a random sample of households from the identified addresses. The survey was conducted as a 30 minute telephone interview directed to the person with the best knowledge about the flood damage in a household. The surveys provide approximately 2,000 respondents in total (Kreibich et al., 2005; 2011; 2012) of which 42% have flood insurance. The high insurance penetration rate is the result of the majority of observations lying in the Elbe catchment area where the insurance penetration rate is traditionally high. The surveys were intended to explain both damage outcomes from the flood and if a respondent had undertaken precautionary flood risk mitigation measures; the overlapping variables are important for investing risk selection and possible moral hazard (and mainly consist of 6

9 objective risk variables). The flood risk mitigation measures taken from the German survey to examine moral hazard are defined as the following dummy variables: no water barriers (if no mobile barriers to prevent water entering the building are available); no adapted building use (if flood endangered floors are not used in a low value way); no flood-proofed home (if valuable fixed units are not avoided as interior fitting in the flood-endangered floors and if water-resistant materials for interior fitting are not used); no flood risk information (if the household had not collected any information about flood protection before or during the flood); no flood awareness (if the respondent did not know that the area the household is living in is at risk of flooding); not a member of a flood coping network (the household is not a member of a citizens initiative for the improvement of flood preparedness and protection). We have also developed an indicator for how likely a respondent feels that they will be affected by a flood to act as a proxy for subjective risk preferences. Another proxy for subjective risk preferences is a dummy for the river catchment area a respondent is located in due to different risk cultures across Germany. All variables used are described in Appendix A and more detailed information about the surveys can be found in Kreibich et al. (2005; 2011). The U.S. data are obtained from field surveys that measured the evolution of coastal residents risk perceptions and preparation plans as three hurricanes Irene (2011), Isaac (2012), and Sandy (2012) approached the U.S. during the 2011 and 2012 hurricane seasons. In these studies, perceptions and preparation decisions were notably measured in real time as they were being made by residents threatened by the storms. The surveys for these three storms provide 1,698 respondents in total iv and include information on whether respondents had a homeowner s insurance policy that would pay for damages to one s home resulting from the storm, if they had a separate flood insurance policy, and whether they knew the amount of their insurance policy deductible or would have to look it up. While 85 percent of total respondents indicated having a homeowner s insurance policy, only 32 percent answered having a separate flood insurance policy. Answers to these two questions serve as our indicator variables for whether a respondent has homeowner s insurance or flood insurance. We utilize four dummy variables for the behavioral moral hazard measures: no preparation (if have not undertaken any of the presented short-term preparation activities); no window protection (if answer no to whether their home has any sort of window protection); no mitigation (if answer no to whether ever modified their home to reduce the amount of hurricane wind damage other than having window protection); and no evacuation plans (if answer no to whether they plan to evacuate to someplace safer). Short-term preparation activities identified included whether purchased supplies for the home such as food, water and batteries; filled car with gas; filled generator with gas (or readied generator); put up storm shutters; took in furniture or other outside precautions; and made reservations or plans in case evacuation is needed. While only eight percent of total respondents indicated not doing any short-term preparation activities, 67, 78, and 71 percent did not undertake any window protection, long-term mitigation, or evacuation plans respectively v. In order to account for an individual s subjective risk perception of the event in relation to undertaking any risk reducing activities we include a measure of safety perception. Responses to the following question were given on a 0 to 100 scale: How safe did one feel about staying in your home through the storm, considering both wind and water? 0 indicated certain that it will not be safe and 100 indicated certain that it will be safe vi. The mean perception of safety values for any one storm were all above 75 indicating that survey respondents felt relatively safe concerning the impending hurricanes. Respondent location data also allowed for spatial geocoding 7

10 in GIS where respondents were determined to be located in or out of the 100 year floodplain, as well as the distance in miles from the nearest coastline. 21 percent of survey respondents were located in a 100 year floodplain and the mean distance to the nearest coast for all respondents was 0.99 miles (0.54 miles for those in the 100 year floodplain, 1.11 for those outside), and these two measures serve as objective measures of risk in our estimations (correlation of and 0.11 vs. feeling of safety respectively). To control for any previous damage suffered from a hurricane we have a categorical variable of damage = 1 if have ever experienced damage from a hurricane, either while living in their present home or a different home, 0 otherwise. More detailed information on the real-time hurricane survey methodology, data, and specific questions can be found in Meyer et al. (2014). Statistical methods In the initial statistical analysis for both the German and U.S. samples we apply probit models to estimate the relation between risk reducing behavior and insurance purchasing (Cutler et al., 2008). Specifically, for the U.S. data sample the likelihood of an individual having either a homeowners or flood insurance policy is estimated as a function of the implementation of any behavioral risk reducing short-term preparation or long-term mitigation activities that were undertaken prior to the arrival of an impending hurricane. Undertaking these behavioral measures are hurricane risk reducing in that they could reduce damage to one s property (putting up storm shutters, taking in furniture, permanent modifications to one s home, etc.) or oneself (purchase of food and water supplies, made reservations in case evacuation is needed, plan to evacuate, etc.). We additionally control for an individual s subjective risk perception through a variable that indicates how safe one feels in staying in their home throughout the hurricane event. We also control for the objective risk of survey respondents by indication of their location in or out of a 100 year floodplain as well as how far they are located from the coast. Lastly, we control for their previous experience of hurricane damage. A similar approach is taken for Germany. We estimate the likelihood of having a flood insurance policy as a function of several variable groupings. The first set consists of behavioral risk reducing measures, such as deploying mobile water barriers, as these measures are likely to reduce the damage suffered in case of a flood. We control for subjective risk preferences through a dummy variable for the catchment area, and if the individual feels they will not be flooded again. Objective risk measures are if the respondent has been flooded before and is located within a 100 year floodplain vii. The second set of empirical models differs between Germany and the U.S. Propensity score matching (PSM) is applied to the German data (Rosenbaum and Rubin, 1983). Our PSM approach is in line with other studies that use matching methods to investigate moral hazard in insurance markets (e.g., Barros et al., 2008). This is similar to the studies reported in Cohen and Siegelman, (2010) that exploit natural experiments in order to detect moral hazard and adverse selection in damage outcomes. In order to detect adverse selection and moral hazard in this set-up the information used by an insurer must be known and conditioned upon. However, the German data is based on a series of surveys of households collected after the major flood events, and includes data on risk and household mitigation activities of which it is in principle unknown whether these are observed by the insurer. Therefore, the results of the PSM cannot be directly interpreted as estimates of information asymmetries that cause moral hazard or adverse selection. Instead, the PSM results provide evidence of effects on damage of risk selection and risk mitigation behavior 8

11 by insured households, which can lead to adverse selection and moral hazard problems when these are not observed by the insurance company. Our application of PSM follows the methodology described in Hudson et al. (2014), which consists of a two-step procedure. The first step refines the control (the non-insured group) and treatment groups (the insured group) to a sub-sample that is strongly comparable with one another with respect to risk defined as vulnerability, exposure and hazard. The second step compares the damage outcomes of the control and the treatment groups in order to estimate the effects of the treatment, which is in this case having an insurance policy. Using actual flood damage suffered as the dependent variable, the presence of a behavioral change due to insurance would be detected though an average treatment effect on the treated (ATT) estimate that is significantly different from zero. A confounding bias due to risk selection into insurance has been filtered out from this ATT, which provides an indication of the importance of risk traits in determining both damage outcomes and insurance purchasing. As is common, we report results of five different matching methods in order to provide an informal robustness check. The reason is that the unconfoundedness assumption that underlies PSM is more likely to hold if the estimated ATTs are all reasonably close to one another. See Appendix A for more details on PSM. In the second analysis of the U.S. data we are primarily focused on investigating the role of known deductible levels on the likelihood of undertaking any short and long-term preparation activities, while simultaneously controlling for subjective and objective measures of risk and previous hurricane damage. Specifically, for only those respondents indicating having homeowners insurance we undertake three separate statistical estimations: 1) Heckman sample selection model where the selection stage is a probit model of the likelihood of undertaking any short-term preparation and the outcome component of the model estimates effects of the explanatory variables on the actual number of preparation activities undertaken; 2) a probit model of the likelihood of having window protection in-place; and 3) a probit model of the likelihood of having done any other home mitigation. 4. Results Results for Germany Table 1 provides results of a probit model of how a household s flood insurance purchase relates with mitigation activities, and risk information. These results do not provide evidence for moral hazard since lack of undertaking two of the three mitigation measures are not significantly related with the likelihood of having a flood insurance policy, while those households who did not employ water barriers are 6.4% less likely to have flood insurance. The significant marginal effects of the 3 information variables further complement this finding. These variables show that individuals who were less proactive in understanding and coping with the flood risk that they face are also less likely to have flood insurance. As we expect risk perception to be related to the purchase of insurance, mixed results are found for the subjective risk variables. This is because the variable of the future likelihood of being flooded again is statistically insignificant, while the catchment area risk culture proxy is the single most powerful influence. The two objective risk variables included in this probit model of being located in a 100 year floodplain and being flooded before are insignificant. These variables are traditionally viewed as being important determinants of flood insurance purchases. There insignificance in this application can be explained through the importance of the risk reducing 9

12 behavioral and informational variables in capturing differences in risk preferences, which outweigh the importance of subjective and objective risk factors. Overall these probit results provide evidence in favor of advantageous selection as the proactive actions of a policyholder in mitigating risks are positively related with flood insurance purchases. This is further confirmed by estimating a bivariate probit model, in line with the common approach developed by Chiappori and Salanie (2000), between having an insurance policy and employing a mitigation measure. In this approach two separate probit models are simultaneously estimated. This allows for the cross correlation between the error terms of the two probit models to be estimated. This correlation (rho) is an estimate of the unobserved relationship between having insurance and carrying out mitigation measures. A positive rho indicates advantageous selection as based on a unobserved relationship the insured are more likely to carry out mitigation; a negative rho results in moral hazard or the same reasoning. The estimated bivariate probit models confirms advantageous selection with a positive statistically significant rho. Thereby confirming the results of the univariate probit model viii. The PSM analysis of flood damage outcomes provides further support for the absence of a moral hazard effect. Flood risk is a function of exposure (the value of what can be damaged), hazard (the probability and intensity of a flood), and vulnerability (susceptibility of the building or contents to damage) (Kron, 2005), while flood damage is a single realization of the risk faced. The presence of risk selection or moral hazard would cause systematic differences in these variables between the insured and the non-insured samples. A mean comparison of experienced flood damage between groups of individuals with, and without, flood insurance reveals that insured individuals in Germany suffered significantly higher flood damage to contents and buildings (Table 2). It can be argued that because a mean comparison contains the ATT and a selection bias (Angrist and Pischke, 2008) it estimates the combined effect of the risk selection and the behavioral element of moral hazard. Risk selection would be expected to increase damages, while the behavioral effect of having insurance coverage is more ambiguous as insurance could cause individuals to become more lax or insured individuals may take more mitigation measures because they are generally very careful (risk averse). It is thus an empirical issue to estimate whether individuals with flood insurance experience higher, or lower, flood damage than uninsured individuals. Panel A of Table 3 presents summary statistics of the hazard experience. It shows that an element of adverse risk selection may be present because the treatment group scores higher on various hazard indicators than the control group. In other words, respondents with flood insurance suffered from a worse flood event, as the difference in water levels shows. This suggests that households with flood insurance face a higher flood risk as the overall shape of the water-level distribution can be argued to be the same across different flood magnitudes, although it is centered around different locations. Furthermore, water-level can be considered to be the most important variable of influence on flood damage (DEFRA, 2006; Merz et al. 2010). In conclusion, adverse risk selection may be present in the German flood insurance market since the insured households faced higher risk than non-insured households, which indicates that higher risk households have a greater incentive to purchase flood insurance coverage. Problems with adverse selection can, in practice, be limited by reflecting in insurance premiums the higher flood risks of individuals in floodplains who demand flood insurance than individuals in floodplains who do not buy flood insurance. The PSM estimates can be regarded as providing an indication of the presence of a moral hazard effect, because the confounders remove risk selection effects on flood damage. Table 2 10

13 shows that the expected difference in flood damages between the groups with, and without, flood insurance is lower once adverse risk selection has been controlled for using PSM. The latter removed effects on flood damage resulting from risk-related factors that determined whether people purchased flood insurance. The ATT estimates suggest that a behavioral change by insured people does not increase flood risk, because the difference in damage between the insured and non-insured is statistically insignificant. For this to be the case the behavior of the two groups must be rather similar with respect to vulnerability to flood hazard which is influenced by the undertaking of damage mitigation measures. For moral hazard to be present the insured group would have to undertake fewer protective measures. The summary statistics displayed in panel B of Table 3, and the results of Table 1 do not indicate this to be true. Moreover, the insured group also seems to be more informed about the risk they face as well as being more likely to be a part of a flood support network. It is also arguable that the insured group is more risk averse than the non-insured group, as every member of the insured group employed at least one of the flood coping measures indicated in Table 3, while only 56% of the non-insured group did so ix. Therefore, it is possible that the higher level of risk aversion has reduced any negative moral hazard effect. It may be the case that insurance has directed individuals to undertake actions that improve their ability to cope with floods, such as joining a flood support group, although this is unlikely based on the results of Thieken et al. (2006). The previous results are based on data from both the Elbe and Danube River catchment areas; however, there may be differences between these two catchments. Because of historical reasons the flood insurance cultures in the two catchment areas have developed differently. The Elbe catchment is mainly located in the former German Democratic Republic, where flood insurance was a part of the compulsory insurance policies household must have. Even after the reunification of Germany, a large number of households in that area still have an equivalent set of contracts, while insurance penetration in former West Germany (including the German part of the Danube catchment) is much lower (Thieken et al., 2006). In order to investigate if the PSM results are being driven by regional effects, the model was estimated using only the sample of households located in the Elbe catchment area, and again, but restricted to, the Danube catchment. The results of these spilt sample models are presented in Table B1 in Appendix B. These results are broadly similar to the pooled model results, and do not provide evidence of a moral hazard effect after controlling for risk selection into insurance. Moreover, a split sample analysis of the relation between flood preparedness activities and flood insurance coverage (Table B2) reveals that the positive relation between insurance and mitigation is stronger in the area where the decision to buy flood insurance is more consciously made (the Danube catchment). In conclusion, our results for Germany provide evidence for adverse risk selection in this flood insurance market, while we find the opposite of a moral hazard effect. We observe that individuals with flood insurance take more flood risk mitigation measures and are more proactive informing themselves about the risk, which suggests that they are more risk averse. Next, a similar set of probit models will be applied to the U.S. sample in order to examine if the German findings also hold for the U.S. context. In addition, the U.S. data allow for examining the influence of the deductible on flood preparation activities of insured households which may counteract possible moral hazard effects. Results for the United States 11

14 Table 4 presents the relationship between the lack of undertaking any hurricane risk reducing behavior and having homeowners (model 1) and flood insurance coverage (model 2) in the U.S. We have pooled the data across the three hurricanes and control for any unobserved hurricanespecific fixed effects through hurricane dummy variables (Irene, Isaac, Sandy), with Sandy being the omitted category. x Negative coefficients signs across both pooled models (1) and (2) indicate that those survey respondents that engage in no short or long-term ex-ante property risk reducing behavior preparation, window protection, mitigation - are less likely to have homeowner s or flood insurance compared to those that do engage in these activities. And these effects are statistically significant in both models at the 1 percent and 5 percent levels. That is, those without homeowners or flood insurance are more vulnerable due to a lack of risk mitigation measures, and thus those that have insurance do not exhibit evidence of moral hazard. For example, for two otherwise average U.S. respondents, the probability of having homeowners (flood) insurance for those that engage in no preparation activities is 23 (12) percentage points less than those that engage in any preparation activities. We see similar statistically significant percentages for those that engage in no window protection (4 and 11 percent less likely to have homeowners and flood insurance respectively), and those that engage in long-term mitigation (12 percent less likely to have flood insurance) as compared to those that engage in either risk-reducing activity. For both models (1) and (2), those that have experienced previous hurricane damage are more likely to have homeowners and flood insurance than those respondents that have not experienced hurricane damage, which is statistically significant at the one percent level. While those that are further from the coast are more likely to have homeowners insurance, respondents located in the 100 year floodplain are more likely to have flood insurance, with a decreasing effect in distance from the coast as indicated by the negative coefficient sign. Hurricane Isaac respondents are also more likely to have flood insurance as compared to those from Sandy. This finding is likely a remnant from Hurricane Katrina striking the same Isaac geographic area in 2005 causing massive flooding damage. Lastly, those respondents that engage in no ex-ante personal risk reducing behavior (have no plans to evacuate) are more likely to have homeowner s insurance. These results suggest a trade-off in risk aversion to property losses vs. risk aversion to personal harm among our respondents. The results in Table 4 do not indicate the presence of moral hazard in regard to U.S. natural disaster coverage. However, if insurers are concerned that moral hazard may be occurring, then one way to offset this is through the use of a deductible. The deductible forces the insured to have skin in the game by making them at least partially responsible for any losses incurred. In the U.S. separate wind and hurricane deductibles ranging from 1 to 5 percent of the insured value of the home provide a potentially substantial incentive to homeowners. Unfortunately for insurers relying on a deductible to offset moral hazard behavior, our survey data suggest that homeowners are not aware of their deductible amount, or if they are aware believe it to be relatively low. For example, from our 1,442 respondents who indicate that they have homeowners insurance, 62 percent do not know what their deductible is xi. Furthermore, only 12 percent believe it to be greater than $1000. For only these respondents indicating having homeowners insurance, Table 5 presents the pooled hurricane results from the Heckman sample selection (model 1) and probit (models 2 and 3) estimations. These models aim at primarily statistically investigating the deductible s role in the likelihood of undertaking any short and long-term preparation activities, while controlling for having flood insurance in-place, previous hurricane damage experience, the perceived level of safety, and objectives measure of risk. xii Separate bivariate probit estimations were also run for 12

15 each model (1) to (3) with having a known deductible as the dependent variable of the other equation. Chi-squared results on rho for pre-event preparation and window protection indicated these as independent equations (see Appendix B). Similar to the Table 4 results, we see little evidence of moral hazard for the insured as the likelihood of undertaking any short or long-term preparation activities, as well as the number of preparation activities undertaken, has a positive and statistically significant relationship with having flood insurance in place. As would be expected, coefficient signs indicate that having experienced hurricane damage in the past, feeling less safe, and living in the one hundred year flood plain are generally positively related to undertaking more preparation activities. Although having experienced damage is the only statistically significant variable. In terms of the deductible coverage, the model (1) selection stage coefficient value of having a known deductible does not suggest this increases the likelihood of undertaking any preparation activities. From the model (1) second stage results in comparison with those having an unknown deductible amount (the omitted category), coefficient signs generally indicate that knowing one s deductible increases the number of preparation activities undertaken, but only knowing that you have a deductible greater than $2500 is statistically significant. Similarly, from models (2) and (3) while most coefficient signs on the various deductible levels are positive, only knowing the deductible is $2500 or greater has a statistically significant impact on the likelihood of undertaking window protection or long-term mitigation. The lack of statistical significance on any of the deductible variables in the three models, other than for the highest deductible levels (> $2500) is notable, especially since only 12 percent of respondents believe their deductible to be greater than $1000, if they know it at all. These results indicate the deductible s relative lack of importance in incentivizing short-term preparation or longer-term mitigation ahead of the hurricane for our insured respondent sample. 5. Conclusion It is often suggested that adverse selection and moral hazard create problems in establishing wellfunctioning markets for natural disaster insurance. Here we estimate the behavioral required component of adverse selection and moral hazard for natural disaster insurance markets in Germany and the U.S. utilizing field survey data. Flood insurance purchases in Germany appear to be either insignificantly or positively related with flood preparation activities of households. This evidence suggests the opposite of a moral hazard effect. Moreover, Propensity Score Matching has been applied to estimate the influence of adverse risk selection and behavioral changes as a result of having flood insurance on experienced flood damage by households living along the Elbe or Danube Rivers in Germany. The results show that adverse risk selection can occur since households with flood insurance experienced a worse hazard during past flood events in the Elbe or Danube catchments. However, flood damages do not significantly differ after controlling for this adverse risk selection effect, meaning that behavioral changes related to moral hazard have not heightened the vulnerability of insured households to floods. Actually the opposite of a moral hazard effect may be present since individuals with flood insurance in Germany are more likely to have undertaken one of the suggested flood coping measures than uninsured households. These findings are overall robust to a split sample analysis of respondents located in the former German Democratic Republic and former West-Germany. This split sample approach is of interest since flood insurance purchases in the latter may be more of a conscious choice and less driven by habits than in the former German Democratic Republic, where flood coverage was 13