Abstract

Central banks require high-quality, granular and timely statistics to support monetary policy formulation, payment system oversight, and financial system stability. Rapid digitalisation and rising analytical complexity increase the need for flexible data environments that enable methodological experimentation, robust model testing, and end-to-end digital transformation of statistical processes. However, these ambitions are constrained by strict confidentiality, personal data protection obligations, and the imperative to preserve public trust. In this context, synthetic data offers promising capability, provided its role is clearly bounded and governed.
This paper proposes a practical framework for implementing synthetic data within central bank statistical systems. The framework is designed around three objectives: (i) creating a safe experimentation space via data augmentation to test and compare statistical methods before implementation, (ii) enabling privacy-preserving transparency and controlled sharing, and (iii) accelerating digitalised processing through realistic augmentation that strengthens model performance while maintaining governance discipline.
The framework is constructed through a structured methodology combining internal governance and regulatory analysis with external literature review and benchmarking, followed by iterative evaluation with academic and industry stakeholders. It is aligned with the Generic Statistical Business Process Model (GSBPM) and integrates synthetic data across four key stages: methodological development, statistical processing, dissemination, and evaluation. To protect statistical integrity, the framework establishes explicit usage rules by classifying use cases into encourage, conditionally allowed, and discouraged practices.
Governance is anchored on two complementary pillars: synthetic data as a Privacy-Enhancing Technology (PET) to enable safe experimentation and controlled data sharing through systematic management of re-identification risks, and as a methodological augmentation instrument to strengthen analytical capability. These pillars are guided by core principles of trustworthiness, relevance, utility and quality, security and privacy, robustness, and transparency, supported by a lifecycle approach spanning data understanding, preprocessing, generation, evaluation, and risk assessment.
Although grounded in Bank Indonesia’s context, the framework provides transferable guidance for central banks seeking to institutionalise synthetic data in a way that supports innovation and modernising their statistical systems while safeguarding the credibility of official statistics or public trust.