Multi‐national underwriting cycles in property‐liability insurance : Part II – model extensions and further empirical results

Purpose – The purpose of this article is to examine the existence of underwriting cycles in property‐liability insurance for Switzerland, the USA, and Japan over a period of 40 years (1957‐1997), it aims to overcome some of the limitations of a pure autoregressive process of second order (AR(2)). Design/methodology/approach – Loss‐ratio data for the three countries are used for the recent period 1957‐1997. To test for the existence of cycles and calculate their length, we start with autoregressive processes of second order and include more variables that should be taken into account. We also estimate a cointegration model in order to separate long‐term and short‐term dynamics. Findings – The hypothesis of cycles of six years in length no longer holds globally and it is necessary to go further than estimating simple AR(2) processes. Different models or specifications seem adequate for different countries because of the varying structure of the insurance markets. Cycles are still found for the USA and most estimations for Switzerland, whereas most specifications for Japan have not revealed cycles. Originality/value – Most empirical work on underwriting cycles has so far been done on US data. This study investigates insurance cycles for three countries on three continents. Based on the autoregressive process of second order and the estimations presented in the companion to this article (i.e. Part I), the model is extended by inclusion of more variables and some of the time‐series characteristics that were ignored in many previous studies are dealt with.