Central banks, as both producers and users of statistical information, have been increasingly interested in making the most of the wealth of data available in modern societies. But they are also fully aware of the need to address the important challenges associated with dealing with granular data sets. These data sets typically comprise micro-level records (such as at the individual account, instrument or transaction level) as well as disaggregated data (that are below the level of aggregated statistics and have a higher likelihood of identifying reporting units). The challenges they pose include not only the large size of these data sets and their often limited quality, but also their complexity due to the high level of details they entail and the sheer variety of their formats.
Despite these challenges, granular data are appealing for a variety of reasons. The most obvious is that they can offer a level of precision and comprehension into economic phenomena that aggregation may typically mask or hide. Further, granular data provide sufficient flexibility and adaptability to tackle a wide scope of issues. The insights gained can enable central banks to have a better understanding of emerging behaviours and lead to more targeted and effective policy prescriptions. Additionally, since many micro data sources are a by-product of the increasingly digitalised world, they provide information that can be timelier and more resilient to disruption, such as during financial crises. Moreover, granular data can support the production of novel economic indicators (for example on economic confidence or uncertainty) as well as experimenting with alternative ones. Finally, they have proved to usefully complement traditional macroeconomic statistics, for instance by helping to assess real-time accuracy and address gaps or missing data.
While granular data offer a level of detail and insight not previously possible with aggregate sources, central banks must confront two important issues. First, these data inherently lack the thorough production processes and quality assessments of typical macroeconomic statistics. Second, to provide value they must be situated within a coherent framework and contextualised so that they can ultimately inform policy decisions.
Fortunately, central banks’ experience shows that data science and its large set of new tools and techniques can be very effective at tackling these issues. Four aspects deserve careful consideration:
- Granular data in their raw form are often unstructured, unverified and dispersed. Consequently, lengthy pipelines are typically required to receive, transform and validate the information collected before it can be useful. Due to the vastness of the data, these operations can hardly be labour-driven and require a high degree of automation. Fortunately, recent advances in artificial intelligence (AI) and machine learning (ML) offer great opportunities to harness the insights from granular data, by performing a wide variety of tasks such as anomaly detection, real-time monitoring, pattern recognition and predictive analytics.
- The defining feature of granular data is the possibility of identifying individual reporting units. While there are serious considerations related to privacy, central banks in isolation might not be able to process the vast quantity of granular data and benefit fully from the insights they offer. Finding the means to collaborate with other authorities, the industry and/or academia without jeopardising privacy or confidentiality requires new and innovative ways to work with and share granular data.
- A third important feature relates to the governance and standardisation of granular data. Unlike aggregate statistics, for which common frameworks are usually in place to ensure consistency and comparability, harmonising granular data across countries is more challenging for a variety of reasons. These include varying levels of technological infrastructure, idiosyncratic regulatory and legal frameworks, as well as different degrees of data availability and access. Given the interconnected nature of today’s globalised world, developing universally adopted frameworks and norms related to granular data appears essential.
- A majority of key policy models and analyses at central banks have been built over the years based on macroeconomic statistics. Important challenges need to be addressed in order to update or provide new models that leverage granular data, and to reconcile granular data with their aggregate counterparts. Looking forward, making the most of the opportunities provided by granular data calls for developing appropriate mitigation measures to safeguard their security, address their quality problems and ensure the usefulness of the information provided in a transparent way, especially for supporting policy decisions.