Time series forecasting of quarterly railroad grain carloadings

The participants in the grain logistics system need forecasts of railroad grain carloads. Although forecasting studies have been conducted for virtually every mode, no forecasting studies of quarterly railroad grain transportation have been published. The objectives of the paper are (1) specify a US quarterly railroad grain transportation forecasting model, and (2) empirically estimate the model. The selection of explanatory variables requires that they have a theoretical relationship to railroad grain transportation supply and/or demand, and that the data for the explanatory variables are published in quarterly frequency. However, there are relatively few potential explanatory variables that are published quarterly and those that are available appear to have weak correlation with quarterly railroad grain carloadings. The economic process generating quarterly railroad grain carloadings is quite complex and very difficult to model with regression techniques. Given this problem and the focus on short run forecasting, a time series model was employed to forecast quarterly railroad grain carloadings. An AR(4) model was estimated using the Maximum Likelihood estimation procedure for the 1987:4–1997:4 period. The actual railroad grain carloadings for this period were compared to the forecast carloadings generated by the time series model. For 92% of the 37 quarters the percentage difference between the actual and forecast values was 10% or less. Of the 9 annual observations, the per cent difference between the actual and forecast value was less than 2.6% for 8 of the 9 years.