A mathematical model first proposed by Box and Jenkins in 1967 for forecasting and prediction in time series analysis based on the variable’s past behaviour. It produces very accurate short-term forecasts but requires a large amount of past data. The method involves determining what type of model is appropriate by analysing the autocorrelations and partial autocorrelations of the stationary data and comparing the patterns with the standard behaviour of the various types. The parameters of the model can then be estimated to provide the best fit to the data.