There is a large and growing literature indicating that traditional time-series models cannot properly capture the behavior of many important economic variables. The problem is that standard time-series models are linear so that they imply a symmetric adjustment process. Consider the simple linear^(1) model: *r = ctx,.i + e, (1) where: is a stationary random variable, and e,is a white-noise disturbance such that = for every time period t.
Enders, Walter and Ludlow, Jorge, "Estimating Time-Varying ARMA Models Using Fourier Coefficients" (1998). Economic Staff Paper Series. 306.