The Financing of Catastrophe Risk
The price of catastrophe reinsurance in the United States has fluctuated markedly in recent years. These fluctuations are commonly associated with the pattern of catastrophe occurrences. For example, catastrophe losses during the period 1992-94 totaled $38.6 billion in 1994 dollars, exceeding the cumulative total of losses during 1949-91 of $34.6 billion. During this three-year period, prices on catastrophe-reinsurance cover more than doubled and then began to decline thereafter. What drives such changes in price? Does the demand for reinsurance shift, does the supply of reinsurance capital change, or do both occur?‘
If catastrophe losses lead to a decrease (leftward shift) in supply, then we would expect to see increases in price coupled with declines in quantity after an event. Of course, a decline in supply is possible only in the presence of some form of capital market imperfection. If capital markets were perfect, the supply curve for reinsurance would be perfectly elastic. In this case, regardless of losses, the price of reinsurance would be fixed, where the “price” of a contract is best thought of as the ratio of premiums to actuarially expected losses covered under that contract. Capital market imperfections would imply that the marginal cost of producing reinsurance is increasing in the quantity supplied. Thus, these imperfections lead to an upward-sloping supply curve, which (all else equal) can shift back as a result of reinsurer losses. Such a supply shift increases price and reduces quantity. As with price, it is best to think of this “quantity” as the actuarially expected loss covered by reinsurance.
On the other hand, catastrophe losses may lead to increases in demand. Rightward demand shifts can be thought of as the result of an actual or perceived increase in actuarial losses covered by a given contract. We call this probability updating. Naturally, it would seem possible to identify such demand shifts from the fact that they lead to an increase in price and quantity. Thus, conditional on a loss, an absolute decline in the quantity of reinsurance purchased would be evidence of important leftward shifts in supply even if there were also positive increases in demand.
We look for such absolute declines in quantity, but, in addition, we pursue the probability-updating hypothesis further. While it is impossible to distinguish between probability updating and capital market imperfections on the basis of the behavior of aggregate price indices over time, it is possible to distinguish between them on the basis of the behavior of cross-sectional changes in reinsurance prices. Specifically, probability updating ought to vary across contracts, with larger price increases associated with contracts for which more probability updating occurs. We therefore examine cross-sectional price increases in response to an event and determine the extent to which they are explained by relative contract exposures.
To see how this works, consider a catastrophe loss caused by a winter freeze in New England. We might expect such a loss to affect strongly (and positively) the distribution of prospective losses due to freeze andor the distribution of prospective losses due to other perils in New England. After all, the event may cause people to recognize how much damage a freeze can do or to learn about the replacement costs of certain physical assets in New England. However, such updates in knowledge would have little or no import for the distribution of catastrophe losses outside New England, where freezes do not occur. Specifically, little would be learned about loss exposures in California (which faces primarily earthquake risk), the Southeast (which faces primarily hurricane risk), or Texas (which faces primarily windstorm risk). Under probability updating, it follows that contracts with relatively little exposure to freeze and/or to the Northeast region ought to have relatively small price increases. In this way, we are able to further distinguish between capital market imperfections and probability updating.
Our identification strategy is made possible through the use of a unique and detailed data set from Guy Carpenter and Company, by far the largest catastrophe-reinsurance broker for U.S. catastrophe exposures. These data include all 1970 and 1994. They allow us to measure prices and contract losses and to go about the complex process of estimating each contract’s exposure to different event types and regions.
To preview our results, we find that supply, rather than demand, shocks are more important for understanding the effect of losses on reinsurance prices and quantities. Capital market imperfections therefore appear to be the dominant explanation. There is limited evidence for probability updating, and what evidence there is suggests that the effect is of a small magnitude. The magnitudes of the supply effects are large: after controlling for relative contract exposure, a $10 billion catastrophe loss raises average contract prices by between 19 and 40 percent and reduces quantity of reinsurance purchased by between 5 and 16 percent.
The rest of the paper is organized as follows. Section 5.1 sets out our identification strategy and the structure of our empirical tests. Section 5.2 describes our data sources. In section 5.3, the calculation of contract exposure and price is discussed in detail. We devote considerable attention to the calculation of exposure, which requires a number of involved steps. Section 5.4 provides a brief graphic analysis. The empirical testing is carried out in section 5.5. Section 5.6 summarizes and offers our conclusions.