In this paper the exponential smoothing methods of forecasting are rationalized in terms of a statistical state space model with only one primary source of randomness. Their link, in general terms, with the ARMA class of models ( both stationary and nonstationary cases) is also explored.
