Many macro-economic variables display cyclical behavior where amplitude and cycle lengths do not seem constant over time. There are various time series models that can describe such data features, but a common property of these models is that longer-term out-of-sample forecasts lose these features. In this paper we propose a stochastic-mean duration-dependent Markov Switching model, extending the work of Durland and McCurdy (1994), which meets this drawback. We show that our model can produce forecasts with cyclical behavior while also incorporating the uncertainty in the level of the future states and the timing of the switches. We discuss representation, parameter estimation and inference of this new model. We illustrate the model using the monthly percentage growth in the number of building permits in the Netherlands from 1996 to 2012. An additional benefit of our model is that it provides insights into the way any information on the level of building permits growth becomes available over time. Hence, recursive inference of our model can lead to useful monitoring of the data.