American Journal of Economics 2015, 5(3): 394-403 DOI: 10.5923/j.economics.20150503.11 A Markov Switching Three Regime Model of Tunisian Business Cycle Imed Medhioub Department of Finance and Investment, College of Economics and Adimistration Sciences, Imam Muhammad Ibn Saud Islamic University (IMSIU), Riyadh, Saudi Arabia Abstract A great interest is accorded to the non-linearities in modelling economic time series. In this context, Medhioub (2007, 2010, 2011) has proved that Markov switching models can capture the business cycle asymmetries of Tunisian economic activity. In this paper, we propose a three regime Markov switching model to analyse the Tunisian business cycle. The results have demonstrated that the Tunisian business cycle is well characterized by three distinct growth rate phases: a recession regime, a moderate growth regime as well as a high growth regime. Based on the filtered probabilities obtained through the Markov switching models, we conclude that the three state model displays a better out-of sample forecasting performance than the two state one. Furthermore, the prior-recognition of the economic transition relating to a new phase of the economic cycle, can be considered as an adequate dating evaluation of the economic cycles in order to present forecasts concerning the real economic activity fluctuations of the industrial production in Tunisia. Keywords Asymmetric business cycles, Markov switching model, Three state model, Industrial production, Turning points, Forecasts 1. Introduction We have recently accorded a great importance to the significant differences existing between the different cycle phases in order to analyse the economic activity. It is obvious that these differences are mainly due to the business cycle asymmetries. The basic idea of this view is that downturns are more violent but less frequent than upturns. However, by using the techniques based on non linear models we try to analyse and identify the economic cycles relating to such economy. In the economic literature, parametric and non parametric tests have been developed to verify if there exist any asymmetries in the data (such as the deepness and steepness tests of Sichel and the triple test of Randles for example). Medhioub (2007) has proved that the monthly data of the Tunisian industrial production for the period 1994–2004 exhibits some cyclical asymmetries. For this reason, we have used the non linear two-state Markov switching model to analyse the economic activity in Tunisia. The use of parametric tests supports the three-state model better than the two-state model. By choosing with this model type, a three-regime Markovchain is then adopted by us. Hence, the economic interpretation of these three regimes is as follows: ● A low growth regime: this regime is characterized by * Corresponding author: ahmathiob@imamu.edu.sa (Imed Medhioub) Published online at http://journal.sapub.org/economics Copyright © 2015 Scientific & Academic Publishing. All Rights Reserved a negative growth rate, and is therefore associated to the classic recession phases. ● An intermediate growth regime or a regime of moderate expansion: for this phase, we suppose that the economic growth rate is below the trend associated to the growth rate (a growth cycle weak phase) without recession. ● A high growth or high expansion regime: for this regime, we suppose that the economic growth rate is above the trend associated to the growth rate (a strong phase of the growth cycles). There is, however, a theoretical reason for which we assume the possibility of using such interpretations concerning the regimes. Yet, in the empirical applications, the relation existing between the different phases of the three regimes and the economic cycles is strong. However, it must be noted that this third regime interpretation considers implicitly a constant long term growth rate. In this paper, we propose to use the three state Markov switching model in our analysis of the Tunisian business cycle. The advantage of these switching processes, frequently indicated in the literature, lies in their capacities to take into account the asymmetry that cannot be caught by linear models. In the Markov switching models, the asymmetries can be explained by introducing the notion of transition probabilities. We have now, at our disposal, an abundant literature which uses the regime switching models for the monthly and quarterly dating of the different economic cycle turning points. In fact, due to the existence of cyclical asymmetries