Using industry and by-line data, we examine the causes of insurance cycles in a vector autoregressive model. Some of the important findings are summarized below. First, the uncertainty variable explains significant portions of forecast errors of premiums. Second, the significant factors that determine premiums are different for different lines. Third, investment incomes in general are more important for long-tail lines than short-tail lines. Evidence on the response of premiums to shocks suggests that all one-time shocks to variables tend to be relatively permanent. The overall results seem to imply that no single hypothesis is able to explain the insurance cycle.