Autoregressive‐asymmetric moving average models for business cycle data

Abstract

AbstractMuch business cycle research is based on an assumption of symmetric cycles, though it is frequently argued that the downturns are steeper and more short‐lived than the upturns; implying cyclical asymmetries. A new class of nonlinear autoregressive‐asymmetric moving average models is introduced. These models are able to deal with symmetric as well as asymmetric phenomena. A likelihood estimation procedure and a Wald test statistic for symmetry are presented. Evidence of asymmetry is found in US real GNP growth rates.