We generalize the Gaussian mixture transition distribution (GMTD) model int
roduced by Le and co-workers to the mixture autoregressive (MAR) model for
the modelling of non-linear time series. The models consist of a mixture of
K stationary or non-stationary AR components. The advantages of the MAR mo
del over the GMTD model include a more full range of shape changing predict
ive distributions and the ability to handle cycles and conditional heterosc
edasticity in the time series. The stationarity conditions and autocorrelat
ion function are derived. The estimation is easily done via a simple EM alg
orithm and the model selection problem is addressed. The shape changing fea
ture of the conditional distributions makes these models capable of modelli
ng time series with multimodal conditional distributions and with heterosce
dasticity. The models are applied to two real data sets and compared with o
ther competing models. The MAR models appear to capture features of the dat
a better than other competing models do.