Predicting Firms’ Taxpaying Behaviour Using Artificial Neural Networks: The Case of Indonesia

Due to the complexity of tax and the time and resources needed to monitor and examine tax returns,
tax noncompliance is challenging to detect. Big data and sophisticated analytics might help tax
authorities extract actionable data insights. Using income tax record data, this paper employs an
Artificial Neural Networks (ANN) model to predict and discover the determinants of firms’ taxpaying
behaviour. To the best of the author’s knowledge, this study is the first to apply ANN to exploit the
taxpaying behaviour of Indonesian firms. This work examined 538,254 firm-level administrative data
across fiscal years 2014 and 2019 to predict the magnitude of tax payment based on seven variables
of interest: types of tax returns, gross profit margin, operating profit margin, other business income
ratio, other business expense ratio, positive fiscal adjustment ratio, and negative fiscal adjustment
ratio. Multi-Layer Perceptron Neural Network-based models were trained to predict three categories
of taxpaying measurement—i.e, Corporate Tax Turnover Ratio (CTTOR)—across varying magnitudes
of annual turnover. The models predicted the firms' taxpaying behaviour with an average accuracy
rate above 92%. The implementation of artificial intelligence also allows this study to identify
heterogeneous channels responsible for firms’ taxpaying behaviour across groups. This study finds
other business income and positive fiscal adjustment to be significant predictors of taxpaying
behaviour for small and medium firms. In contrast, operating profit margin, other business expenses,
and negative fiscal adjustment are prominent predictors for large corporations. The findings will assist
decision-makers in tax administrations about potential areas of misreporting, enabling them to
develop evidence-based and effective policy actions.