Affordability of National Flood Insurance Program Premiums: Report 1 (2015)

4

The Insurance Purchase Decision

Achieving high rates of flood insurance purchase has been a challenge for the National Flood Insurance Program (NFIP). Congress therefore requested FEMA to consider the effect of premium increases on purchase of flood insurance when proposing an affordability framework (HFIAA 2014). That possibility was also to be part of the analysis called for in BW 2012, Section 100236. This chapter discusses the decision to purchase insurance, focusing mainly on the effect of premiums on purchase decisions. The chapter contrasts a standard model of choice found in the economics literature with behavioral models of choice. These two choice models provide the necessary context for reviewing empirical data on factors that affect insurance purchase decisions. Insights that can be useful for FEMA’s efforts to make flood insurance purchase more attractive to households are presented on the basis of this literature review.¹

RATIONAL ACTOR MODEL OF CHOICE

The rational actor choice model (well known in the economics literature) posits that insurance buyers estimate the probability of events, such as flooding, and their adverse consequences. With an assessment of possible adverse consequences in mind, the individual considers whether to pay a particular premium each year to avoid the adverse consequences if the event occurs. As part of this thought process, the individual will consider

__________________

¹Household reluctance to purchase flood insurance is not surprising, given the reluctance to purchase other lines of insurance (Kunreuther et al., 2013).

different deductibles and coverage limits. There are important exceptions to this simple model; one is risk aversion. People who are averse to risk might be willing to purchase insurance at a premium that exceeds the annual expected loss. To illustrate, suppose a risk-averse consumer is willing to pay an annual premium of $12 to insure against a loss of $100 that has a 1 in 10 chance of occurring in any year. The expected loss in this scenario is $10. The additional $2—the risk premium—reflects the amount above the expected loss that the individual is willing to pay for insurance.

Risk aversion still requires the rational actor to understand insurance. It is necessary for individuals to take time and effort to evaluate options and related financial considerations. This “deliberative thinking” is assumed by the rational actor model (see Kahneman, 2011; Kunreuther and Pauly, in press). Deliberative thinking involves systematic and effortful behavior that often requires complex computations and the use of formal logic.

The rational actor model is most often used to formulate and test hypotheses about the role of prices in decision making. In the flood insurance purchase decision, the price would be the premium paid. The hypotheses are that higher premiums will affect the amount of coverage purchased or the decision to purchase at all. Another hypothesized influence on purchase would be the price of a substitute; in this case, one substitute for having an insurance claim paid is receiving disaster aid. Although the price of insurance is the premium paid for a selected level of coverage, the price of disaster aid is zero. The expected amount of disaster aid, however, depends on an individual’s perception of the generosity and timeliness of aid. Even if the price of aid is zero, low expectations of aid may make it an imperfect substitute for insurance. If an individual expects aid to be generous, however, it may discourage purchase of insurance.

Effects of Premiums on Purchase of Flood Insurance

Despite conceptual difficulties,² many investigators have attempted to estimate how a change in the price of insurance coverage (the premium)

__________________

²Specifying a price variable can be a particularly difficult process. The most easily obtained data are for the average premium per household. Because of the rating structure, however, the average premium is much larger than the marginal premium for most households, and not in a predictable way. Furthermore, a particular household is assigned a premium on the basis of various structural features, the rating zone, first-floor elevation relative to base flood elevation (BFE), and the chosen deductible amount. That means that any price variable is probably colinear with other variables in the demand function, and this biases any measure of a price effect. Although the marginal, rather than average, premium may seem less afflicted with statistical biases, it is more difficult to obtain at a household level and relevant only to questions about the amount of insurance coverage purchased, rather than about whether insurance is purchased at all.

might affect the decision to purchase insurance or the amount of insurance to purchase. The metric for this effect is the price elasticity of flood insurance purchase decisions. In 1983, the U.S. General Accounting Office (GAO; now the Government Accountability Office) developed an econometric demand function for number of policies issued as a function of several variables, including average premium paid after adjustments (GAO, 1983a). Data was obtained for the years 1978—1982. The result was an estimate of elasticity of policies in force with respect to average premium of –0.38. That defines a relatively inelastic relationship, in which an increase in premium of 1.0% will result in a decrease in the number of policies equal to 0.38%.

Later, Price Waterhouse Coopers LLP (PWC), after searching the existing literature on price effects, reviewed the 1983 GAO study (PWC, 1999). It was noted that the GAO data covered premiums of $41.50-$88.00 during the 1978—1982 period (roughly $114-$241 in 2014 dollars). PWC adopted the –0.38 elasticity reported by GAO for premiums in that range but then assumed (without evidence) that the price effect would increase for larger premiums, eventually reaching an elasticity of –0.76.

A 1999 study estimated elasticity of policies in force with respect to average price at –0.32, similar to that in the 1983 GAO study (Browne and Hoyt, 2000). It also modeled total insured amount and found it to be considerably more price elastic than policies in force (–1.22). A 2000 report examined data on a sample of 11,000 properties drawn from 18 coastal counties (selected by FEMA) and estimated price elasticity by using an expected utility maximization framework (Landry and Kriesel, 2000). The model had the fraction insured properties as a dependent variable and derived estimates in two ways: using weighted least squares and using maximum likelihood. Neither approach produced a statistically significant relationship between the average premium and the probability of purchasing insurance.

A RAND Corporation study used a national sample of 5,472 single-family homes in a logistic model to estimate flood insurance purchase (Dixon, et al., 2006). On the basis of results of the model, the elasticity of the probability of purchasing insurance with respect to price was estimated to be –0.06; in this case, price is defined as the premium cost per $100, averaged over the total coverage for a property.

A 2008 study collected data on 1,692 properties in two coastal counties of North Carolina and used the data in a Tobit model to estimate flood insurance coverage elected (Landry and Jahan-Parvar, 2008). The formulation is able to recognize the fact that flood insurance coverage is a bounded variable: it cannot be less than zero or greater than $250,000. Two sets of marginal insurance premiums were estimated by using different assumptions for deductible amounts and level of coverage with respect to

replacement cost. The authors also considered the effect of premium subsidies, where they existed, and responses to a household survey designed to elicit more possible explanatory variables. Price elasticities were computed for various combinations of the data sets. On the basis of the high-premium models and the largest data set, price elasticity of insurance coverage was estimated at –0.26 for nonsubsidized properties and –2.09 for subsidized properties. Introducing the household level data from the survey reduced the nonsubsidized elasticity to –0.12. For the low-premium alternative, three model specifications were used for each set of marginal premiums. For the high-premium alternative, price elasticities ranged from –0.12 to –0.49. For the low-premium alternative, all results are somewhat more elastic, although the authors cautioned that these are upper bounds on the true effect.

A 2014 study use a Tobit model of insurance coverage purchased, including two alternative measures of marginal premium (high and low) among the explanatory variables (Howard, 2014). This analysis of insurance demand included estimates of consumer surplus. No price elasticities were reported.

Another 2014 study collected 32 years of data for 153 counties in Georgia (Atreya et al., 2014). The study is notable for several things, such as the inclusion of county-level data on race, education, and age. The dependent variable is policies in force per 1,000 of population; the price variable is average premium cost per $1,000 of coverage (in 2010 dollars). All data are county-level aggregates or averages. Two model specifications and three estimation methods were used, for a total of five sets of results. Price elasticity values ranged from –0.14 to –0.31. The most inelastic value (-0.14) reflects explicit correction for serial autocorrelation. It also was noted that flood insurance purchases increase with educational attainment, with increased proportion of black households, and with age.

Only a few conclusions can be drawn from the literature:

• Overall, the probability of insurance purchase is quite inelastic with respect to either the average cost of coverage (Dixon, et al., 2006) or the marginal cost of coverage (Landry and Jahan-Parvar, 2008). Results of various studies yielded elasticity values in the range of –0.38 (GAO, 1983b) to –0.06 (Dixon et al., 2006); results in the vicinity of -0.10 were more common.
• Subsidized policy-holders may be much more responsive to changes in marginal price than those with full risk premiums (Landry and Jahan-Parvar, 2008).
• Total coverage purchased may be considerably more elastic with respect to average premium cost than the probability of purchasing insurance (Browne and Hoyt, 2000).

• The probability of insurance purchase by households subject to the mandatory insurance provision is slightly less responsive to the average premium than the probability of purchase by households not subject to the requirement (Dixon, et al., 2006). Mandated-insurance households also purchased slightly more coverage than other policy holders (Landry and Jahan-Parvar, 2008).

Effect of Expectations for Future Disaster Aid on Demand for Flood Insurance

Some major flood events lead to presidential declarations and trigger the availability of federal disaster aid. Publicity surrounding such aid often sends a message that large amounts of money are being distributed and may create an impression that a substantial fraction of households flood losses will be compensated. A widely shared perception of generous postdisaster aid might depress the demand for flood insurance (Kousky and Shabman, 2015). The implication of generous post-disaster grants might lead many to view post-disaster aid as a substitute for flood insurance. In reality, federal disaster aid is limited to specific events, is uncertain, is modest in scale, is mostly for repair of public infrastructure or for protection against future damages, and offers little to households that are not insured (ibid.); thus, it may not fully substitute for insurance. It is still possible, however, that a widely shared perception of generous post-disaster aid depresses the demand for flood insurance (Kousky and Shabman, 2012).

An empirical demonstration of this effect may be difficult to obtain. A 2000 paper included previous disaster assistance as an explanatory variable and hypothesized that past experience with high levels of disaster aid would reduce the demand for insurance (Browne and Hoyt, 2000). The result was a small, but statistically significant, positive relationship between disaster aid and insurance purchases. The authors attributed that unexpected result to collinearity: both disaster aid and insurance purchases are thought to positively correlate to the level of risk. Another 2000 study found no evidence of demand suppression by disaster aid (Landry and Kriesel, 2000). Although examined in other contexts (e.g. Herring, 2005; Brown and Finkelstein, 2008), there are few empirical findings on disaster assistance in the United States. One examination of insurance purchases after receipt of federal disaster aid for flood events in Florida found that receipt of individual assistance had a crowding-out effect on flood insurance purchases (Kousky and Michel-Kerjan, 2014). A 2006 study found only a small relationship between insurance takeup rates and disaster aid and only for compensation with respect to damaged property (Dixon et al., 2006). The authors noted, however, that this finding could be because those who

receive assistance do not have the means to purchase insurance coverage or because much disaster assistance is for losses not covered by insurance.

None of the reviewed studies directly investigated property owners’ perceptions regarding the future availability of disaster aid, so there is no basis for ruling it in or out as a factor in the demand for insurance. Several laboratory experiments and surveys have asked individuals if they consider disaster aid when making insurance decisions; these have the benefit of assessing how perceptions before a disaster can influence the purchase decision, but there is also concern that answers to surveys may not reflect real-world purchase decisions. Results vary in the literature. Usually, if individuals are told about assistance, it will lower their willingness-to-pay for insurance; but without such a prompt, they may not consider disaster aid when making insurance decisions (Kunreuther et al., 1978; van Asseldonk, Meuwissen, and Huirne, 2002; Botzen and van den Bergh 2012; Petrolia et al., 2013; Raschky et al., 2013).

The rational actor model would suggest that the possibility of disaster aid will discourage the purchase of insurance, but there is no consistent or persuasive empirical evidence of this effect. At best, the effect of perceptions of aid on the flood insurance purchase decision remains an open question.

BEHAVIORAL MODEL OF CHOICE

Behavioral models of choice argue that nonfinancial considerations and intuitive thinking can be used to understand choices. Intuitive thinking relies on mental shortcuts when foregoing, purchasing, or canceling insurance on the basis of such reactions as anxiety or regret; it uses simple decision rules (heuristics) that are influenced by personal experience with events, such as a flood and its consequences. The heuristics require less effort in making a decision than the detailed analyses implied by the deliberative rational actor decision process.

Nonfinancial Considerations

Whether to purchase insurance is a risk management decision, but it may not be based solely on financial considerations (Krantz and Kunreuther, 2007). For example, a homeowner may buy insurance to reduce anxiety about suffering a large uninsured loss (and thus to provide peace of mind) or to avoid regret when a flood occurs about not having purchased a policy. There is an extensive literature on how nonfinancial considerations influence individuals’ risk management decisions (e.g., Finucane et al., 2000; Loewenstein et al., 2001). For example, some people claim that they refuse to fly not because they fear a crash, but because they anticipate and dislike feeling anxious about a crash while they are on a plane; however, people

who cannot avoid anxiety about a loss may still find opportunities to reduce this emotion by taking protective measures. That may partially explain the demand by the few who purchase flight insurance. Similarly, individuals might pay more for insurance if they fear a specific event (for example, home damage from a flood) than if they are not concerned about the event even if the actual expected losses are the same. Regret and disappointment are different from anxiety, as they are experienced mainly after a loss rather than before, but anticipation of these emotions also can influence decisions.

For example, consider this common behavior: homeowners purchase flood insurance after suffering damage in a flood and then cancel their policies when several consecutive years pass without experiencing any flood damage. One explanation of this behavior is that reducing anxiety in anticipation of a flood and reducing regret if a flood occurs are both important goals immediately after suffering water damage; the cause of the loss is deeply etched in the purchaser’s recent memory. Buying insurance is easy to justify to oneself and others because a flood has just occurred. Several years later, many people may find that the prospect of a flood no longer intrudes on their peace of mind, so they are less anxious about its consequences.

Mental Shortcuts

A second departure from the rational actor model is the process by which individuals consider risk information. This process has been termed intuitive thinking. The literature describing intuitive thinking is vast and at times uses different terms to describe the same phenomenon. Three select findings from the literature on insurance purchase decisions are presented in this section: prospect theory, status quo bias (a reluctance to consider alternatives to the current condition), and availability heuristic (considering the most recent event that occurred most recently to be the most likely).

Prospect Theory

Kahneman and Tversky (1979) proposed prospect theory to explain how individuals make choices when outcomes are characterized by a probability distribution. Prospect theory argues that individuals misperceive probabilities, having a tendency to underweight small probabilities and overweight larger ones. If the probability of an event is perceived to be extremely low, the likelihood is considered to be zero. Empirical studies reveal that individuals tend to experience the pain of a loss twice as strongly as the enjoyment of the gains of the same magnitude. Stated simply, individuals tend to be loss-averse relative to their reference point (Tversky and Kahneman, 1991). For example, a controlled laboratory experiment found that many individuals bid zero for insurance coverage against low-

probability events, apparently viewing the probability of a loss as so small that they are not interested in protecting themselves against it (McClelland et al., 1993).

Status Quo Bias

A flood insurance purchase decision is made when a homeowner buys a house in the floodplain and is considering whether to purchase flood insurance for the first time or when a policy expires and a homeowner has to decide whether to renew it. There is evidence that many individuals are reluctant to depart from the status quo (not having insurance, or holding a policy that is expiring) even though there may be substantial benefits to them from doing so (Samuelson and Zeckhauser, 1988). With respect to consumer insurance decisions, changes in laws in Pennsylvania in 1990 and in New Jersey in 1988 provided an opportunity to examine the impact of the status quo as a reference point on the choice of automobile policies. Insurance laws in the two states differed with respect to the status quo (that is, the default option). In New Jersey, motorists had to change their existing insurance policy to acquire the full right to sue that would result in a higher premium. In Pennsylvania, the status quo was the full right to sue and motorists had the opportunity to reduce their insurance premium by giving up some of their rights to sue. When offered the choice between these two policies, only about 20% of New Jersey drivers chose to acquire the full right to sue. In Pennsylvania, 75% of the insured population retained their current policy, which gave them the full right to sue (Insurance Information Institute, 1992). Similar results were obtained in a hypothetical study with 136 university employees. Interestingly, the effect was even larger in the real world than in the controlled experiment (Johnson et al., 1993).

Availability Heuristic

In some situations, individuals assess the probability of an event on the basis of the ease with which they can imagine its occurrence (Tversky and Kahneman, 1973). The availability mental shortcut implies that individuals are more interested in buying insurance coverage after a disaster because it is highly salient. Indeed, it has been found that takeup rates of flood insurance policies in the United States increase right after a disaster event and then slowly decline (Gallagher, 2014). The flood insurance market offers more striking empirical evidence on that point. A 2012 study examined the number of new policies issued by the program and their durations through 2009 for those residing in both Special Flood Hazard Areas (SFHAs) and non-SFHAs by using the entire portfolio of the NFIP over the period 2000–2009 (Michel-Kerjan et al., 2012). Of the 841,000 new policies in

2001, only 73% were still in force 1 year later; after 2 years, only 49% of the original 2001 policies were still in place; and in 2009, only 20% were still in place.

Although some of these individuals may have sold their homes and cancelled their policies because they moved, the large percentage decrease in the policies in force can be only partially explained by migration patterns. Data from the annual American Community Survey over the period covered by the flood insurance dataset revealed that the median length of residence was 5-6 years—somewhat higher than the median tenure of flood insurance of 2-4 years.

That finding of higher insurance purchase after catastrophe is often true even when premiums increase (unlike NFIP policies). A prime example is the purchase of earthquake insurance after a major seismic event. Surveys of owner-occupied homes in counties in California that were affected by the 1989 Loma Prieta earthquake showed a significant increase in coverage (Palm, 1995): just before the disaster, 22.4% of the homes had earthquake coverage; 4 years later, 36.6% had coverage—a 63% increase. The possibility of a future earthquake was now more salient, so many individuals decided to purchase insurance to gain peace of mind.

The availability of mental shortcuts also implies that before a disaster, the perceived likelihood of another disaster is perceived to be much lower than estimated by experts (Tversky and Kahneman, 1973). For example, consider floods in August 1998 that damaged property in northern Vermont, an area that had not experienced a recent major natural disaster. Of the 1,549 victims of this disaster, FEMA found that only 16% of homeowners who were in flood-prone areas had insurance, even though 45% were required to purchase it (Tobin and Calfee, 2005; Michel-Kerjan et al., 2012). These findings imply that lenders were not enforcing the regulation or that property owners were finding ways to avoid lender enforcement. In the case of Hurricane Sandy in 2012, only about 20% of New York City households that were inundated had flood insurance at the time of the disaster (NYC, 2013).³

Framing

Framing refers to the way in which outcomes are described as gains or losses relative to a reference point, which can either be the status quo or another value (Kunreuther and Weber, 2012). One way to encourage individuals to invest in protection is to reframe the probability of risk so that people perceive potential future disasters as above their threshold level of

__________________

³Many areas inundated by Hurricane Sandy and the associated surge were outside of then-designated SFHAs.

concern. Research shows that simply adjusting the time frame can have a significant effect on the perception of risk. For example, people were more willing to buckle their seatbelts when they were told that they had a one-in-three chance of an accident over a 50-year lifetime of driving, rather than a 0.00001 chance in each trip (Slovic et al., 1978; Kunreuther et al., 2013). Similarly, describing flood probabilities in terms of the number of “1 in 500 year” possible floods during the 30-year life of a mortgage may have greater meaning than telling someone that there a 0.2% chance of a flood in any year.

Suppose that a person is provided with a concrete scenario highlighting the damage to property from a future flood and that this question is posed “How would you fare financially if you did not have insurance and suffered a future loss from a storm similar to Hurricane Sandy?” Individuals at risk may decide that they should purchase coverage rather than regret not being financially protected if they suffer a severe loss. More generally, calling attention to the benefits of insurance by focusing on a specific event such as Hurricane Sandy, is likely to be more effective in increasing takeup rates than framing a general message in terms of reducing damage from future floods or hurricanes. Even before 9/11, controlled experiments revealed that consumers are willing to pay more for insurance against a plane crash caused by terrorists than for flight insurance against any event, a counter-intuitive finding in that by definition “any event” includes a terrorist attack (Johnson et al., 1993; Kunreuther et al., 2013).

IMPLICATIONS FOR ENCOURAGING PURCHASE

Deliberative Thinking

The concept of deliberative thinking is a process that aims to provide concrete and meaningful comparisons in helping people understand a given risk, and removing misperceptions that people may have about that risk. People generally have difficulty in evaluating low-probability risks, but often form more accurate perceptions when numbers are presented in the context of familiar situations. A raw probability number, such as 1 in 1 million, may be an abstract concept, but people can more readily interpret such a number if it is compared with the risk of an automobile accident (1 in 20), or the risk that lightning will strike one’s home on one’s birthday (less than 1 in 1 billion).

Misperceptions of the likelihood and generosity of aid may influence an NFIP purchase decision. This suggests that decisions on whether to purchase insurance could be affected by providing accurate information on the limits of federal aid. Another example of a misperception is the view of insurance as an investment. Some insured individuals do not feel justified

in continuing to pay premiums when they do not collect on their policies. They view insurance as a poor investment rather than recognizing the fact that they have not suffered any losses for the last few years (Kunreuther et al., 1978). Insurance is a risk-management mechanism, however, and a person should value not having a loss rather than thinking that money was wasted in the premiums. People who view insurance as an investment misunderstand its purpose.

FEMA has prepared a two-page brochure that illustrates this comparison (FEMA, 2012a). Support for deliberative thinking might include expanding the material into a broader educational effort. Expanded educational materials might explain not only the limits of aid but also the uncertainty of the aid being secured. In a broader context, educational materials might increase understanding of the purpose of insurance. In fact, FEMA’s FloodSmart program represents a significant effort to inform the public of the benefits of purchasing flood insurance (FEMA, 2014a). In addition, a user-accessible financial decision support tool to help persons to compare the financial consequences of purchasing with those of not having insurance and relying on aid. Dissemination and use of these materials might also be considered. One possibility might be to make such materials available to the community floodplain managers and to write your own (WYO) agents who can work with property owners to assess the merits of aid versus insurance for their particular situation.

Choice Architecture

Households have many financial decisions to make and limited time in which to make them. Some households will use some of the time for a deliberative process when making an insurance purchase decision. Others may use mental shortcuts. In recognizing that some (maybe many) households will use mental shortcuts, the goal of increasing purchase may be served by paying attention to “choice architecture” (Thaler and Sunstein, 2008). The authors of this 2008 article argued that people’s choices often depend on how options are presented. For example, consider the situation regarding the NFIP mandatory purchase requirement. There has been continuing attention to enforcement of the mandatory purchase of flood insurance for properties in SFHAs that have federally insured mortgages. As in the past, the focus has been on the lending sector, but there is still less than full compliance. Efforts to increase enforcement of mandatory purchase by focusing on the borrowing sector have been only partly successful. An analysis of the sequence of decisions that includes the property owner who may be reluctant to purchase any form of insurance (for the reasons discussed), the WYO agent, whether the commission for each policy purchased is enough to encourage following up when a policy is dropped, the lender, and who-

ever owns the mortgage over time can identify possible reasons for failures to maintain insurance and suggest policy actions to increase compliance. The focus appropriately would be on the household decision because if households do not choose to drop insurance in the first place, enforcement will be less of a challenge.

An illustration suggests how choice architecture might result in people’s buying and maintaining flood insurance policies. An analysis of the entire portfolio of the NFIP revealed that more than half of all NFIP policies (mandatory and voluntary) were canceled between 2-4 years after purchase (Michel-Kerjan et al., 2012). This illustration applies equally to voluntary and mandatory purchase. Currently, the choice context is for households to purchase insurance on an annual basis; that is, they need to evaluate each year whether to renew. If there has not been a flood in the preceding year, the availability heuristics (discussed above) may work against a decision

to renew. If the renewal choice were for multiple years, however, the effect of availability heuristics might be minimized and presenting the choice as a multiyear purchase might take advantage of the status quo heuristic (see Box 4-1).

SUMMARY

A long-standing objective of the NFIP has been to increase purchases of flood insurance. Household decisions on whether to purchase insurance can be understood through different models of choice that in turn have been the foundation for empirical studies of the insurance purchase decision. A conclusion from a review of this literature is that no single strategy will increase purchase of NFIP policies. As FEMA improves its efforts to increase takeup rates through risk communication and other efforts, the literature reviewed here may offer insights that can improve the effectiveness of these programs (Kousky and Shabman, 2015).

• The original NFIP legislation expected NFIP premiums to be priced at reasonable levels to promote voluntary purchase of NFIP policies. Empirical studies have found that premium prices may affect takeup rates although the size of that effect is small. The effect of the availability of disaster aid on insurance purchase decisions is uncertain.
• Studies have found that people may use intuitive thinking, as opposed to systematic consideration of the cost of premiums in relation to expected claim payments, when choosing to forego insurance or to cancel an existing policy.
• The combination of acknowledgement of intuitive thinking and the effects of premiums on insurance purchase decisions suggests that lower premiums alone will not increase takeup rates substantially.
• Keeping NFIP premiums at reasonable levels can be part of any strategy to maintain compliance with mandatory purchase requirements and increase voluntary takeup rates. A multipart strategy to motivating purchase of NFIP policies can be designed using insights from the behavioral sciences literature.

This page intentionally left blank.