In this paper, we propose a generalised specification of a time varying transition probability Markov switching model for U.S. industrial production index. The model is specifically designed to investigate the presence of asymmetries in the shape of the cycle, given its relevance in the debate about long-run effects of recessions on output level. We can think about asymmetries in the shape of the cycle along two main dimensions. First, we can think about patterns of variation in growth rates over the course of expansions and recessions. Second, we can consider to which extent recessions are simply negative expansions. The model, estimated using Bayesian methods, generates posterior probabilities of being in recessions which correspond to the NBER dated recessions, provides support to the presence of a recovery early in expansions, consistent with what found in the literature, and estimates the shape of recessions to be linear, contrary to some previous parametric studies. When we investigate the ability of our specification to produce plausible business cycle features, where those features are a set of statistics proposed by Harding and Pagan (2002), we find that the model is able to capture all of them. Finally, the effects of recessions on long-run output level implied by our specification lie between those predicted by two important benchmark models of this literature.