Evidence on the Time Series Properties of Insurance Premiums and Causes of the Underwriting Cycle: New Support for the Capital Market Imperfection Hypothesis

Abstract

Introduction Numerous studies document that underwriting profit rates and insurance prices in the property-liability insurance industry have been cyclical in the postwar period (for example, see Cummins and Outreville, 1987; Doherty and Kang, 1988; and Venezian, 1985). A hard where prices and profits are high and coverage is restricted tends to be followed by a soft where prices and profits are low and coverage is expanded. The cycle in property-liability underwriting profits has a period of about six years.(1) The of this underwriting cycle has been the subject of much debate. In a perfect market, a price cycle would occur only if insurers' expected costs--the present value of expected losses--were cyclical. Most explanations, however, attribute the cycle to imperfections on the supply side of the property-liability insurance market.(2) Cummins and Outreville (1987) present a model in which insurers set prices rationally, but in which almost all relevant information in forming expectations of future losses is contained in past loss reports. These assumptions, along with regulatory and accounting lags, induce a cycle in reported profits and prices, Venezian (1985) attributes the cycle to the extrapolative methods used in the rate setting process. Several recent articles show that capital market imperfections can generate a cycle. For example, Winter (1989) assumes that the quantity of insurance is constrained by an insurer's equity and that internal equity is a cheaper source of capital than external equity (Myers, 1984). Consequently, a reduction in capital (e.g., due to unexpected losses) causes industry supply to contract and insurance prices to increase.(3) These hypotheses have different implications for the determinants of premiums, which are tested using time series regressions similar to the causality tests proposed by Granger (1969) and Sims (1972). In a perfect market with rational expectations, insurers set premiums equal to the present value of expected future losses, where expectations are formed using all relevant information. Premiums are therefore the best predictors of future losses in the sense that premiums aggregate all information about future losses.(4) In the terminology of Sim's causality tests, premiums cause losses; that is, current premiums can be explained by future loss payments. After controlling for expected future loss payments, other information such as past losses and past surplus will not help explain premiums. In contrast, the capital market imperfection hypothesis (Winter, 1989; also see Gron, 1992, and Cummins and Danzon, 1992) implies that premiums are not the best predictors of actual losses because past surplus affects premiums. Venezian's (1985) model implies that premiums are not the best predictors of actual losses, because premiums are mechanically set based on the realizations of past losses. Accordingly, past loss experience affects premiums. Although data problems limit the strength of the conclusions, our results suggest that market imperfections play an important role in insurance pricing. Consistent with the capital market imperfection hypothesis, the evidence suggests that past values of surplus affect premiums. Weaker evidence is presented that past values of losses affect premiums, which is consistent with Venezian's hypothesis. The article proceeds as follows. In the next section, the hypotheses are described in more detail. In the third section, the methodology and data are presented along with potential econometric problems. The results are then summarized. The article concludes with a short summary and some caveats. Hypotheses Perfect Markets/Rational Expectations Hypothesis In a perfect market, the price of insurance would be determined by discounting expected cash flows by the appropriate discount rates. To be more specific, consider a competitive insurance market with no transaction costs and no information costs. …