Forecasting Recoverable Spares Box-Jenkins Time Series Techniques

In recent years, Air Staff directed a comprehensive review of recoverable spares forecasting due to significant underestimates of spares requirements. The purpose of this study was to determine if time series forecasting models could accurately forecast demand for aircraft recoverable spares. Box-Jenkins time series analysis was used to analyze and develop forecasting models for ten C-135 recoverable spares. Two different Box-Jenkins models were developed to forecast demand for each spare. These forecasts were compared to actual demand and to forecasts done using simple exponential smoothing. The first type of Box-Jenkins model built was the multivariate (transfer function) model. In these models, flying hours are the independent/ input variable and demand is the dependent/output variable for forecasting. The second type of model is the univariate model in which past demand relationships are used to forecast demand. The three types of models forecast one quarter of demand. The results were compared to the actual demand for the quarter. Results showed low correlation between flying hours and demand in the transfer function models. Though each type of model forecasts well, simple exponential smoothing had better results for the short term (three months) forecast. In the majority of forecasts, the three models overestimated demand. Keywords: Forecasting, Time series analysis, Box-Jenkins, Spare parts, Recoverables, Theses. (SDW)