Nonlinearities in business cycle: SETAR models and G7 industrial production data

In this paper we enquire if nonlinear or linear models produce iid residuals using the BDS test for the G7 industrial production series. We found that nonlinear models produce iid residual for the USA, and evidence of NL are also detected for Canada and Germany, while Japanese data does not reject the alternatives, and the L model is rejected for Italy, France and the UK. On this evidence, 'white chaos' data generation processes cannot be rejected. We evaluate the linear and nonlinear impulse response functions finding that: long-run multipliers are smaller for European countries; long-run asymmetries are evident for the USA, Canada and West Germany. Differently from previous research, our results show that the non-linear hypothesis has to be taken seriously, but with some caution because of not entirely satisfactory forecast performance and asymmetries (this is especially true for the US series). On the other hand, the hypothesis of linearity can be rejected with greater confidence, since it almost always fails to encompass the alternative.