Abstract

Authors :Anggraini Widjanarti, Akhmad Zacky Nugraha, Mohammad Khoyrul Hidayat, She Asa Handarzeni,, Insan Istafada

The rapidly evolving data landscape is compelling central banks to modernize statistical and analytical capabilities to remain relevant, trusted, and impactful for public policy. As digital transformation accelerates, the demand for evidence-based policymaking increasingly relies on granular data. Yet, wider utilization of such data is constrained by confidentiality requirements, personal data protection obligations, and practival barriers to safe data sharing for cross-departmental analytics or collaborative research initiatives. In this context, Bank Indonesia is exploring synthetic data as a strategic Privacy-Enhancing Technology (PET) to enable responsible collaboration while safeguarding sensitive information.
This paper presents a case study on generating and assessing synthetic data credit card transaction data derived from Integrated Commercial Bank Reports. We implement two state-of the-art synthetic data generators, Conditional Tabular Generative Adversarial Networks (CTGAN) and CopulaGAN, and evaluate the resulting datasets across three core dimensions: fidelity, utility, and privacy, to ensure ethical and reliable methodological implementation. Empirical results from fidelity testing indicate that synthetic data effectively replicates core statistical distributions, achieving overall quality scores of 84.47% via CTGAN and 88.86% via CopulaGAN. From a utility perspective, the synthetic outputs support descriptive analysis and the exploration assessment of transactional dynamics, enabling analysis of consumption patterns and market developments without direct exposure to confidential microdata. From a privacy perspective, the generated datasets are fully synthetic and contain no direct duplicates of original records, thereby reducing exposure to record-level identification risks.
Overall, the findings suggest that synthetic data can serve as a practical PET for central bank statistics use cases, expanding analytical accessibility under controlled condition, supporting cross-unit collaboration, and reinforcing trust through responsible data governance.