Central banks’ experience in enhancing the access to and sharing of information through data science

Abstract

Data science offers significant potential for leveraging both traditional and emerging data sources, improving statistical processes and enabling advanced data analysis in central banking. It also facilitates the secure sharing of granular data sets while protecting sensitive information. However, there are various challenges, including the need for robust IT infrastructure, organisational barriers and limited quality of secondary sources, which can constrain their use for official purposes. Fortunately, central banks are well-equipped to address these issues, not least because of their expertise as both data producers and users. Looking ahead, enhancing data management, promoting interoperability, investing in modern data platforms and fostering structured exchanges of experiences across stakeholders are essential to unlock the full potential of data in today’s digital age and effectively support central banks’ public mandates.

Introduction

The ongoing data revolution presents central banks with various opportunities to support their statistical function as well as the formulation and implementation of evidence-based policies. The exponential growth in the amount of accessible information enables more real time, detailed and multidimensional insights, particularly through the integration of novel sources such as big data into traditional macroeconomic aggregates (IFC (2024)). However, challenges often arise from the limited quality and standardisation of large or complex data sets, as well as from operational, ethical and legal barriers such as confidentiality restrictions.

Fortunately, innovative data science techniques – including big data analytics, machine learning (ML) and artificial intelligence (AI) – can be part of the solution. These tools help central banks to better tap into unstructured data, such as text. They can also improve data quality, extract meaningful insights and facilitate data access and sharing while protecting sensitive information. Against this backdrop, data science has emerged as a practical solution for central banks to fully unlock the potential and value of data.

Drawing on the IFC bulletin “Data science in central banks: enhancing the access to and sharing of data”, this policy brief underscores the benefits of data science for central banks as both producers and users of reference information. It also highlights some key obstacles, particularly with emerging and secondary data sources, and emphasises the importance of international collaboration to address these challenges.

Accessing and making data available for further analysis

Data science can enhance the use of existing as well as novel data sources, with benefits for both producers and users of economic and financial information in central banks.

For producers, data science and ML in particular play a crucial role in enhancing data quality, accuracy and timeliness. Central banks are using these technologies for tasks such as anomaly detection, error identification and uncovering patterns in financial data to support areas such as prudential supervision and anti-money laundering (Shah (2025)). ML also facilitates the production of more granular and timely statistics by leveraging real time sources and automating processes across the data life cycle, including data collection, integration and evaluation (IFC (2022)).

Turning to users, innovative techniques, such as clustering, neural networks and natural language processing (NLP), can enable the analysis of large, multidimensional and/or complex data sets. These methods help, for instance, to uncover patterns, assess market sentiment and improve forecasting accuracy for key economic indicators. In particular, NLP and large language models (LLMs) allow central banks to tap into unstructured data, such as text. This can be quite valuable for them, for instance in extracting qualitative insights from their communications to improve forecasting of macroeconomic variables such as inflation (Araujo et al (2024)).

Data science can also improve the access to and sharing of data. Advanced techniques enable information to be shared or analysed while securing an adequate level of protection, eg through privacy-enhancing technologies, such as data obfuscation and encrypted data processing. Perhaps more significantly, data science allows to share information without disclosing it. Techniques such as federated learning, for instance, enable models to be trained without revealing the raw data. Multi-party secure private computing, on the other hand, supports the sharing of computations among data holders, allowing them to retain full control over their information assets without needing access to others’ data (Ricciato (2024)).

Key obstacles in accessing, sharing and using data

While innovative data techniques can enable better access to and sharing of vast and diverse amounts of information, various challenges remain.

A key challenge is the substantial investments in IT systems required to effectively leverage data science. Long-term strategies may thus be needed to address the growing need for high-performance computing and platforms for managing diverse data types. Fortunately, central banks appear to have significantly upgraded their IT and data infrastructures over the past few decades (IFC (2020)). They are also increasingly interested in adopting cloud solutions because of their scalability and performance. Yet they are also aware of a number of issues posed by the cloud, not least increased dependency on external providers, potentially weaker information protection and concerns about data sovereignty.

The adoption of data science may also face organisational barriers. First, data silos can be an issue as they prevent users from effectively tapping into the available information due to isolated data management being split across the various functional domains of central banking. Second, the low transparency of techniques such as ML can also explain organisational risk aversion towards adopting data science, while also highlighting the need for continuous training and human oversight (IFC (2025b)). Finally, limited collaboration between data scientists, IT and subject-matter experts can prevent the full deployment of data science within the organisation. Experiences from central banks suggest that success depends on a clear strategy, sound data governance and close stakeholder collaboration, for instance by setting up organisational structures such as data science hubs (Duijm and van Lelyveld (2025)).

Finally, the deployment of data science can be hindered by limitations in data quality. Emerging data sources, such as big data, are often a byproduct of other processes and may feature a low degree of standardisation and consistency, making them unsuitable for statistical purposes without adequate preprocessing (Perrazzelli (2025)). Challenges also arise due to the lack of consistent identifiers, impeding tasks like quality assurance and integration as shown by the BIS’ experience with building a database on cryptocurrencies (Illes and Mattei (2025)).

Making data access and sharing more effective

The above considerations highlight that achieving effective and adequate data access and sharing may require further progress in four key areas.

First, data management and governance are important for ensuring data quality, usability and security through clear sets of principles and adequate resources (Križman and Tissot (2022)). Key steps might include ensuring the dissemination of high-quality information through traceable documentation and citation standards, enhancing data usability by developing common inventories and adopting a master data management approach to harmonise statistical concepts within a common data dictionary (Gonçalves et al (2025)). Data governance is also a key component of AI governance, not least to ensure adequate provenance and traceability of the data used by AI systems (Figure 1; IFC (2025b)).

Second, the development of interoperable information standards can also be instrumental in effective data use and reuse, including for training ML models (Araujo (2023)). Standards such as Statistical Data and Metadata eXchange (SDMX) provide a unified governance framework for both macro and micro data exchange and metadata management, supporting smoother integration, reduced reporting burdens and enhanced efficiency (IFC (2025c)). Efforts to ensure interoperability across standards, such as SDMX and the eXtensible Business Reporting Language (XBRL), can further support the handling of diverse data types and accommodate users’ needs. Interoperability also facilitates decentralised data architectures, allowing tailored processes to business requirements while maintaining compatibility across systems. Moreover, standards act as AI enablers by improving data discovery, ensuring consistent understanding and enhancing AI accuracy through robust metadata and documentation, thereby mitigating data quality issues (Anvar (2025)).

Third, leveraging large, diverse and complex data sets calls for modern, metadata-driven and accessible data platforms. Single access points such as the European Single Access Point and collaborative frameworks such as the Common European Data Spaces can further enhance data access and reuse. Central banks have also developed specialised solutions, such as research data centres and remote execution systems, to securely facilitate academic research on granular data (IFC (2024)).

Finally, strengthening collaboration among peers and counterparts plays an important role in advancing data science. This can involve co-developing open-source solutions, such as the BIS OpenTech and sdmx.io initiatives. It can also entail fostering a structured exchange of experiences among stakeholders, as exemplified by the series of workshops on “Data science in central banking” co-organised by the IFC and Bank of Italy (IFC (2022, 2023, 2025a, forthcoming)). Collaboration can also be further strengthened by developing common international data frameworks (Tissot (2025)). In this context, the G20 Data Gaps Initiative represents a significant and promising step towards improving access to alternative data for official statistics and promoting the effective sharing of economic and financial information for the public good (IMF et al (2023)).