Improving the Budgeting of Framework Agreement Volumes with Time Series Models in Governmental Procurement

Abstract

Companies still use existing data and predictive analytics too rarely to support their decision making. The use of analytical models has also been neglected in public procurement research. This thesis addresses the utilisation of time series analytics in supporting decision making, improving budgeting, and forecasting public procurement volumes through a case company. The aim is to examine how well time series models can forecast the volumes of governmental procurement, and whether these models can improve the budgeting accuracy of the case company.

Hansel Ltd is a non-profit central purchasing body of the Finnish central government whose purpose is to save public funds by increasing the productivity of governmental procurement. Hansel’s operations focus on tendering and maintaining framework agreements in accordance with the national and the EU legislation. Since around 87 percent of Hansel’s revenue is comprised of framework agreement specific service fees, their allocation is a strategically important decision for achieving zero profit. The level of service fees is based on the budgeted framework agreement volumes, which unfortunately differ from the actual volumes. In recent years, the overall budget has differed from the actual total on average 5 percent, but the differences between the budgeted and the actual volumes vary from few to even 100 percent at the framework agreement level.

The ability of time series models to forecast future framework agreement volumes and the effects of the models on Hansel’s budgeting process was studied by modelling 26 framework agreement subtotals. ARIMA models, which predict future values through previous values and forecast errors, were used in the modelling. ARIMA models capture efficiently the temporal dependence of values with a finite number of parameters, making modelling and creating accurate forecasts straightforward. The forecast accuracy of the models was compared to the current budget numerically with the root mean squared error and the mean absolute percentage error as well as graphically.

The forecasts of the models were about 6 percentage points more accurate than the current budget. Also, the differences to the actual volumes were decreased on average 37 percent at the framework agreement subtotal level. In addition, implementing the models as a part of the budgeting process would reduce the number of steps and overlapping work in the process and increase the transparency of budgeting when the budget was based on theory rather than on the intuition of the top management. By increasing the budgeting accuracy and enhancing the budgeting process the models would also improve the allocation of adequate service fees and achieving the zero profit objective. As the ARIMA models were found to be competent to forecast framework agreement volumes accurately, this thesis has practical implications also outside Hansel. Framework agreements have established a key role in European public procurement, and the benefits of utilizing time series models for public procurement units other than Hansel should be further studied.