A Neural Network Model for Activity-Cost Relationship Estimation in Performance-based Budgeting

One of the concerns of budgeting researchers and managers is how to relate performance data to budget as one of the key concepts of performance-based budgeting. One of the most intricate parts of performance-based budgeting is how to attribute activities to resources and determine the shares of resource drivers. In most customary costing and budgeting methods, it is usually assumed that there is a linear relationship between activities and costs. However, a cost function is not always linear in practice and presuming it as linear would lead to faulty calculations in the budgets of activities, outputs and plans. This article uses artificial neural networks to solve the problem of estimating the relationship between activities and resources (costs). The data of Tejarat Bank of Iran were used to train and test the neural network model. The distinguishing characteristic of this model in comparison to other models is that it considers cost-cost center relationship as non-linear. Special architecture of the suggested network, i.e. multilayer feed-forward network architecture with jump connections, makes it possible not only to forecast costs but also to extract the share value of the resource drivers out of the model. When the results of the suggested model on the value of drivers were compared with the results of experts’ opinion poll on the resource drivers, an acceptable difference was shown.