The analysis and prediction of the short-run economic dynamics, or the evolution of the business cycle, often require a construction of the composite economic indicator (CEI). This indicator may be endowed with nonlinear dynamics to take care of the possible asymmetries between different phases of the business cycle. This paper suggests using the smooth transition autoregression to model the CEI. The performance of this model is compared to the already classical CEI with regime switching. Both models turn out to produce statistically equally good results in terms of forecasting the business cycle turning points.