The purpose of this paper is to highlight some testing procedures, both in time/frequency framework, useful to test for significant cycles in insurance data. The US underwriting cycle is measured using the growth rates of real premiums.
In addition to the traditional AR(2) model, two new approaches are suggested: testing for a significant peak in the periodogram using Fisher g test and a nonparametric version of it, and testing for unit root cycles in insurance data.
All approaches find empirical evidence for a cyclical behaviour of the growth rates of property‐liability real premiums. Results on the length of dominant cycle still diverge, according to the approach (time/frequency domain).
Compared to the existing literature, the present study innovates in that it highlights additional testing procedures, helpful to detect significant cycles in insurance time series. The underwriting cycle is analysed through the growth rates of real premiums.
