University of Tasmania
Iskandar_whole_thesis.pdf (1.97 MB)

The development of government cash forecasting model : a case study for the Indonesian government

Download (1.97 MB)
posted on 2023-05-28, 09:48 authored by Iskandar,
The ability to predict the future cash required to fulfil government responsibilities and public services deliveries is crucial not only for the domestic economy but also for a potential spread to other communities. Discussions on the interconnection between government spending and economic development has been a prominent research area in the field of economic studies. However, the 2010 Greek crisis taught world a lesson that, regardless of existing causality, sustainable economic growth relies on the ways in which government manages expenditure. Moreover, public expenditure management (PEM) sees the national budget as an instrument to influence the economy through several features. One of them is cash management which focuses on ensuring the availability of government money to deliver public services in the most effective way. An effective government cash management (GCM) facilitates the requirements for the government to fulfil its responsibilities and public services deliveries while maintaining economic stability. Furthermore, a reliable government cash forecasting model is essential for an effective GCM. In this thesis, the researcher has developed a government cash forecasting model that meets an acceptable level of accuracy and materiality for use by government cash managers. In doing so, the most appropriate variables were identified for use in the model and a number of statistical methods were evaluated and tested to be used to construct the model. The methodology undertaken by this study was as follow. The government cash forecasting model developed utilised historical daily data of Indonesian government expenditure following three steps: (1) attribute selection, (2) modelling, and (3) performance evaluation processes. Several techniques based on statistical, machine learning, and hybrid methods were tested independently and then each was compared with the other to assist in developing the most accurate forecasting model based on performance evaluation measurements. In the modelling phase, the following methods were used. The Autoregressive Integration Moving Average with Exogenous Variables (ARIMAX) technique was chosen to represent the statistical modelling method. The machine learning methods tested utilised multiple artificial neural network techniques including Feed-forward Neural Networks (FFNN), Cascade-forward Neural Networks (CFNN), Radial Basis Function Neural Network (RBFN), Generalised Regression Neural Network (GRNN), Extreme Learning Machine (ELM), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU). For the hybrid method, a combination of ARIMAX and Nonlinear Autoregressive Neural Network (NARNN) techniques was used. The results show that an accurate government cash forecasting model that meets an acceptable level of materiality for the cash manager can be achieved by identifying and including the most significant variables which influence government expenditure through an attribute selection process and the application of an artificial neural network machine learning-based method. In this study, it was found that the most appropriate variables to build a government cash forecasting model are the total daily available fund for intermittent expenditure, the week of the month, the month of the year, and policy implementation, while the GRU was the most accurate technique. This study contributes to the existing literature and practice in its development of a statistically robust and accurate method to forecast government expenditure. Notwithstanding that Indonesian data only was used in this research, the procedures used in this study and the forecasting model developed are applicable to other governments and public sectors.


Publication status

  • Unpublished

Rights statement

Copyright 2019 the author

Repository Status

  • Open

Usage metrics

    Thesis collection


    No categories selected


    Ref. manager