I want to forecast world market prices of several commodities. Different tests for linearity show that the prices of most commodities are nonlinear. A textbook (Teräsvirta et al., 2011, p.5) says the following: “In the linear univariate case, e.g. differencing the series can transform the series to take advantage of the theory for stationary models. In the nonlinear nonstationary case all this becomes more difficult….” In the rest of the book stationarity is hardly mentioned as a topic at all. My questions are then; How important is testing for stationarity and differencing in nonlinear time series? Should I consider stationarity at all, and in case how?
Thanks in advance!
