Academic journal article Journal of Risk and Insurance

Underwriting Cycles in Property and Liability Insurance: An Empirical Analysis of Industry and Byline Data

Academic journal article Journal of Risk and Insurance

Underwriting Cycles in Property and Liability Insurance: An Empirical Analysis of Industry and Byline Data

Article excerpt

INTRODUCTION

Recent studies have observed the presence of cyclical underwriting returns within the property-liability industry and have proposed competing theories to explain the insurance underwriting cycle. Important proposals include: (1) the extrapolation hypothesis by Venezian (1985); (2) the rational-expectations/institutional-intervention hypothesis by Cummins and Outreville (1987); (3) the fluctuation-in-interest-rates hypothesis suggested by Doherty and Kang (1988) and Doherty and Garven (1992); (4) the capacity-constraint hypothesis by Winter (1988, 1989), Cummins and Danzon (1991) and Gron (1994); and (5) the change-in-expectations hypothesis by Lai and Witt (1990, 1992). These theoretical works have attempted to explain the insurance underwriting cycle, yet empirical studies have not fully tested these hypotheses and have yet to determine the exact causes of the underwriting cycle.(1)

Specifically, much empirical support for these competing theoretical works has not incorporated a comprehensive study that simultaneously tests the hypotheses. Most of the earlier empirical studies examine one specific hypothesis at a time, thus ignoring competing theories that emphasize other variables. This omission of relevant variables may cause variable specification errors. Though a recent study by Niehaus and Terry (1993) has resolved some specification problems, many important issues remain unresolved regarding the time-series behavior of insurance premiums. Our study attempts to examine the causes of the underwriting cycle by (1) rigorously testing the causal relationship and the dynamic interactions between premiums and losses by using a vector autoregressive model (VAR); (2) simultaneously testing the five hypotheses listed above; and (3) examining how premiums respond to shocks in selected variables including surplus, interest rates, the conditional variance of losses, and the variances of interest rates. More importantly, our study focuses on the causes of the underwriting cycle for individual lines because the determinants of a cycle may be different for different lines.

We extend the literature that examines the underwriting cycles in several ways. First, this paper uses more powerful tests that capture the interactions among premiums, losses and other variables. Thus, our tests provide more information than a simple statistical hypothesis test for rejection or non-rejection. Earlier studies analyzing relationships between premiums and other variables rely heavily on regression analysis that may not provide complete analysis of the data and which may lead to questionable results.(2) For example, Niehaus and Terry (1993) find the regression coefficients of lagged surplus variables on premiums have opposite signs for one and two periods (see Panel B, Table 4, Niehaus and Terry 1993), and such results clearly give rise to ambiguous interpretation. Grace and Hotchkiss (1995) also utilize VAR approach but they focus their analyses on model parameters while we use variance decomposition analysis to test various hypotheses. In addition, Grace and Hotchkiss use only industry data, we use both industry and by-line data.

Second, this study is the first to examine the impulse response functions of premiums to shocks in variables such as losses, interest rates, surplus and uncertainty variables using by-line data. This analysis provides evidence of how premiums respond to unexpected changes in the other variables, thus, providing insight into possible causes for cyclical trends in premiums. Both Haley (1995) and Grace and Hotchkiss (1995) report some results of impulse response functions. Haley uses a by-line cointegration analysis of underwriting margins and interest rates, thus, his analysis is only limited to one variable, namely, interest rate. Grace and Hotchkiss also report the results of impulse response functions using industry data while this study uses by-line data.

Third, gaps in hypothesis testing remain--recent studies have not tested the change-in-expectations hypothesis. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.