ARIMA vs ARIMA using Error Correction and Ensemble Model
This paper explores the possibility of improvement in pure ARIMA model through introduction of error correction adjustments in the context of forecasting business metrics like number of churns. It also provides a detailed comparison of performance metrics of ARIMA model and an ensemble model that uses error corrected forecast of ARIMA, exponential smoothing (ES) and moving average. In the real life experimentation, mean absolute percentage error (MAPE) obtained from individual models are used as a weightage of the ensemble model.
Forecasting technique like the Delphi method is a type of judgemental forecast whereas methods like ARIMA and Exponential Smoothing (ES) are statistical methods. ARIMA and ES are the two most widely used methods of forecast. ARIMA and ES method can handle different type of time series data which include trend, seasonality and level. Naïve based forecast where last observation is considered as a forecast and forecast based on simple average of a given number of past observations is also famous in some industry domains like finance etc.
To read the entire whitepaper, please fill the form.