The perverse effects of subsidized weather insurance.

• Author: Ben-Shahar, Omri
• Position: III. The Perverse Effects of Subsidized Weather Insurance A. Distributive Effects 2. Redistribution Under Florida's Citizens Property Insurance Corporation through Conclusion, with footnotes, p. 597-626

2. Redistribution under Florida's Citizens Property Insurance Corporation

   "[Florida] subsidized the well-to-do who live near the beach at the expense of the less-well-to-do who don't."

   --Michael Lewis, New York Times (106)

   1. Citizens' data and some initial observations

      Citizens is the Florida state-owned insurance company that sells, among other types of coverage, wind-peril insurance policies to homeowners in every part of the state. (107) While there are other state-run insurance programs, we study Florida's Citizens because Florida is the state that faces the greatest hurricane risk, and therefore its insurance program is the largest and most important. (108) As mentioned, the policies are priced according to the wind territory in which the insured property is located, of which there are roughly 150. (109) Prices are adjusted annually and have to be approved by the state Office of Insurance Regulation. (110) Statutory and regulatory caps limit the extent to which Citizens can raise its rates in any given year. (111)

      As discussed above, Citizens' actual insurance premiums charged are known--and intended--to be different than the "true risk" premiums (those representing an actuarially accurate methodology). (112) For every calendar year, Citizens publishes charts listing, for each individual policy, the actual premium and the true risk hypothetical premium, allowing a straightforward calculation of the subsidy each policy receives. (113) In 2012, there were 527,250 individual policies. (114) This is the "policy-level data." In addition, because policies are rated and priced based on the risk territory in which they are sold, and because all policies within a given territory enjoy the same proportional subsidy, some of the information can be analyzed by comparing patterns across territories. For that, we used aggregated "territory-level data." (115)

      To get a general sense of the subsidy picture, we looked initially at the territory-level data. Here, in publicly available rate filings, Citizens publishes summaries for each of the 150 risk territories, showing the total sum of premiums paid by policyholders in that territory, as well as the "indicated" rate change--that is, how much more (or less) the company would have needed to charge policyholders in that territory to break even actuarially. Here is an example:

      In Monroe territory, for example, where some of the southern Florida Keys are located, the premiums actually collected by Citizens total $38,582,378, but they fall short of Citizens' estimate of the expected risk. To be precise, an increase of 126.5% in the premium charged to each policy in that territory would be necessary to cover the shortfall. In Tampa's suburbs and in Saint Petersburg, the shortfall in premiums is more modest, 25.9% and 14.7%, respectively. Many of the highly populated Florida areas--such as Broward County, where Fort Lauderdale is located--are divided into several risk territories. As Table 1 above shows, some of these territories, like the one labeled Wind 47, receive a substantial subsidy (57.3% above the actual cost); others, like Wind 48, receive a modest subsidy (17.3%); and some, like the one labeled "Broward" (but which contains only the more inland portions of the county) are actually overcharged--they do not receive any subsidy and in fact their premium helps cover the subsidy paid to neighboring properties.

      Since there are 150 territories and they vary greatly by the amount of subsidies they receive, we wanted to see if any pattern might be discerned. To that end, we created a map of Florida by risk territories and shaded each territory according to the magnitude of the subsidy it receives. The darker the shade of gray in which a territory is represented on the map, the higher the subsidy that territory receives.

      [FIGURE 1 OMITTED]

      Figure 1 shows a remarkable but predictable pattern. Coastal territories, almost without exception, enjoy large percentage subsidies, whereas inland territories receive smaller subsidies, if they receive any subsidy at all. A similar relationship can be seen when we zoom in and look at densely populated southern Florida in Figure 2.

      [FIGURE 2 OMITTED]

      The pattern is even clearer here: the subsidies are larger in territories very close to the water. Figures 1 and 2 also help us begin to speculate about a possible relation between subsidy and wealth, since water proximity is often a feature attracting wealthy home buyers. (117) To visualize this, we plotted on the subsidy maps the location of the highest (represented on the maps by triangles) and lowest (represented on the maps by squares) wealth concentrations. The triangles mark territories in which the median home value is at least three standard deviations above the statewide median. (118) The squares mark areas more than one standard deviation below median home value. No surprise: wealthy households are more often located in the high-subsidy (more deeply shaded on the maps) territories. Poor households are more often located in the low- or no-subsidy territories.

      These maps reflect the territory-based data, comparing the treatment of the 150 different insurance risk territories. Next we wanted to see if the distribution of the subsidies was indeed correlated with the distribution of wealth. To do so, we needed more information about policyholders' wealth. For this we used two metrics:

      (i) Home Value: Home value is a good proxy for affluence because it is a significant component in people's net wealth, accounting for roughly twenty-five percent of the aggregate households' net worth in America. (119) Citizens' policy-level data do not include home values, but they do list the zip codes of the insured properties. Thus, we were able to use publicly available information about median household value within the zip code in which the insured property is located. (120) It is of course likely that less affluent households are blended into the pools that we count as wealthy (those in high-median-home-value zip codes), but this measure allows us to look at overall trends in average subsidy allocations.

      (ii) Coverage Limit: Citizens' policy-level data include an entry for the amount of insurance purchased under each policy. Since insurance law does not allow the purchase of coverage exceeding the value of the property, we can use the coverage amount as an estimate of the lower bound of the property's value. (121) This will help us test whether people who own lower-valued homes receive a greater or smaller insurance subsidy. (122)

      To further visualize the relation between subsidy and wealth, we used the zip-code-level household-value data. For each zip code, we know the median household value, and we computed the average dollar value subsidy for all Citizens' policies issued in that zip code, taken from Citizens' policy-level data. When we did this for all 904 Florida zip codes, we got the following scatter plot:

      [FIGURE 3 OMITTED]

      * The relation between absolute insurance premium subsidy and median household value. Each observation on the graph represents a Florida zip code region (N=904).

      The trend line is positive, suggesting that zip codes with higher-valued homes receive higher per-policy subsidies.

      A similar picture emerges if we look at policy-level data and ask whether high-value insurance policies (those presumably attached to high-value homes) receive a higher or lower subsidy. Before we turn to the connection between the value of the homeowners' insurance policy and the size of the subsidy, let us first provide a rough picture of the range of coverage limits found in the data. As it turned out, the lowest coverage limit in the dataset (presumably for the policy covering the lowest-valued home in the dataset) was for $29,200. (123)

      The highest coverage limit was for $1,000,000. The mean was $228,888. And the median was $196,700. The policies broke down further as follows:

      Table 2 Policy Value Percentile Coverage Limits 1% $92,000 5% $113,600 10% $126,900 25% $152,400 50% $196,700 75% $262,000 90% $361,800 95% $463,900 99% $745,000 Thus, the smallest 1% of the policies had coverage limits of $92,000 or less, the smallest 5% had coverage limits of $113,600 or less, and so on. At the other end of the spectrum, the top 10% had coverage limits of $361,800 or greater, and the top 1% had coverage limits of $745,000 or greater.

      To link these coverage limits, and thus (indirectly) home values, with the size of Citizens' subsidies, we divided Citizens' policies into five quintiles according to the policy coverage amount. For each quintile, we calculated the average subsidy. Again, we see a clear picture: higher quintiles of wealth get a higher absolute subsidy. (124)

      [FIGURE 4 OMITTED]

   2. Empirical analysis

      In order to measure the disproportionate benefit of the insurance subsidy on the affluent, we used Citizens' policy-level data. For each policy, we looked at two measures of subsidy. First, we looked at the straightforward "absolute subsidy," which is the difference between the premium charged and the hypothetical premium reflecting full risk. Since Citizens reports the "indicated rate change" necessary to bring the actual premium to the full-risk level, this absolute subsidy for each policy is simply the premium charged for that policy times the indicated rate change for that policy.

      But the absolute subsidy may tell an incomplete story. A $300 subsidy for a low-coverage policy of, say, $50,000, may be a relatively more significant factor than a $500 subsidy for a high-coverage policy of $500,000. We therefore wanted to measure the relative subsidy each policy is getting. To do this, we created a synthetic benchmark in which the subsidy pool (the total amount of subsidy for all policies within the dataset) is divided pro rata across the policies, under the (counterfactual) assumption that all policies receive the same indicated rate change--the same percent discount. We denoted this...