Determining appropriate levels of data aggregation in a linear programming model

LP models as aids to annual budgeting are often too large to be economically viable unless they employ some form of aggregation. The degree of aggregation is usually determined informally and/or intuitively when the formulation is developed. This note describes some experiments and results with a model where the level of aggregation in the generated LP may be controlled by user-supplied parameter values. The effect of various aggregation levels may then be investigated in a straightforward, explicit manner while the model is still being developed. The information thus gained may serve as a guide to the most appropriate aggregation level for any particular subsequent planning use of the model. The approach adopted here for annual budgeting may prove practical and useful in other application areas where the main disadvantage expected of an otherwise satisfactory LP formulation is the size of the LP model and consequent excessive computing time required.