Making the Most of Big Data for Financial Stability Purposes

Introduction

There is general policy interest for “big data”, which is fundamentally changing the character of the information available to public authorities. To put it simply, this term usually describes extremely large data-sets that are often a by-product of commercial or social activities and provide a huge amount of granular information, typically at the level of individual transactions. This form of data is available in, or close to, real time and can be used to identify behavioral patterns or economic trends. Its impact in terms of information creation, storage, retrieval, methodology and analysis has gained increasing importance, and the private sector is already using to a significant extent data patterns from such data-sets to produce new and timely indicators.

The flexibility and immediacy of big data also provide new opportunities for public authorities involved in financial stability policies. In particular, central banks, macro-prudential authorities and financial supervisors are showing an increasing interest, especially at senior policy level.¹ They can have access to a broader and timelier (often on a real-time basis) range of indicators, opening up the possibility of extracting more timely economic signals, applying new statistical methodologies, supporting economic forecasts and financial stability analyses, and obtaining rapid feedback on policy impacts.

Yet there are important peculiarities as regards the role of big data for financial stability issues. Financial stability needs are quite specific, and not all big data fields are equally promising from this perspective. For instance, central banks’ focus is more on those big data sources that can effectively support micro- and macro-economic as well as monetary and financial stability analyses. In contrast, other types of big data sources – such as geospatial information, an increasingly important source for national statistical offices – is often of lower interest to central banks.

Not only is the focus specific, the challenges faced in handling and using big data for financial stability purposes are also particular. For instance, a key input in financial stability assessments relates to the identification of pockets of micro fragilities that can have a system-wide impact – but basing such an identification on large granular data-sets can be difficult because of, for instance, confidentiality protection rules or commercial copyrights. Another example is the feedback loop that is inherent to policy-making authorities: big data sources can affect central bank policy-making, and in turn the policies implemented can generate new data-sets. This relates to the more general Lucas critique, and is a clear distinctive mark compared to other public statistical bodies in the field of big data. One obvious example is the growing number of qualitative statements that can be used to decipher central banks’ communication, by applying big data-related text mining techniques, and which in turn can modify other agents’ actions. Another example relates to the large number of big data pools generated by various financial regulations, since public authorities in charge of supervising financial institutions and monitoring specific market segments can request a lot of (complex and highly rich) information. In turn, big data can strengthen supervisors’ capacity by providing insights into large amounts of unstructured data (Basel Committee on Banking Supervision (BCBS), 2018).

Another key specificity relates to the public nature of financial authorities and the degree of trust they benefit in the pursuing of their mission of promoting monetary and financial stability. They put a great emphasis on the degree of confidence they enjoy in the society and are usually the first to be concerned about possible ethical or reputational consequences of using big data. Moreover, the risk of misusing big data has to be carefully considered, since policy decisions based on wrong data could have incommensurable consequences. Indeed, the public and private sectors may have different areas of concerns from this data quality perspective: for instance, online retailers targeting potential customers based on past web searches might find it acceptable to be “right” once out of five times, but official statisticians would usually consider such an accuracy level of 20% as completely inadequate…