# Big data in central banks: 2018 survey results

As work in big data enters the mainstream for central banks, its policymaking and supervisory influence is expanding, prompting significant investment in new technologies.

This new survey – the third in a series conducted by Central Banking in association with BearingPoint1 – reports on the approach central banks take towards big data, and data management more broadly. This report has only been possible with the support and co-operation of the central bankers who agreed to take part. They did so on the condition that neither their names nor those of their central banks would be disclosed in this report.

### Key findings

• Work in big data can now be considered a mainstream activity for central banks: over half of survey respondents said they are working on a big data project.
• Big data plays an increasingly significant role in policymaking and supervisory processes. More than 60% of respondents said big data was either a “core input” or an “auxiliary input” into policymaking, an increase on last year’s survey.
• Data governance remains a work in progress for central banks. More than 60% of respondents do not believe their central bank has a clear structure for data governance.
• Spending on data in central banks is decentralised. More than 80% of respondents do not have a single allocated budget for the handling of data.
• Central banks tend to focus their big data investments on software and hardware over human resources and security.
• Ensuring sufficient quality is seen as the greatest challenge when collecting and managing data.
• Central banks increasingly look to external sources to obtain big data.
• Central banks typically make use of a combination of methods to process regulatory data collection. The most popular combination is a self-developed data platform, Excel and data-based handling.
• Central banks are increasingly bringing in new external technologies to manage data. Just under 40% of respondents said they had done this in the past 12 months.
• New technology has yet to exert a significant impact on central banks’ supervisory architecture.
• Central banks see monetary policy as standing to benefit most from big data.
• Central banks typically use a range of methodological approaches to analyse big data, with data mining being the most popular.
• Big data is established as a tool for forecasting and nowcasting.

### Profile of respondents

Survey questionnaires were sent to 130 central banks in May 2018. By June, responses had been received from 52 central banks.2 The average staff size was 2,205 – slightly below the industry average – and 31 respondents had fewer than 1,000 employees. Just over 40% of respondents were from emerging‑market countries, and the most popular source of responses was statistics departments, which accounted for nearly two-thirds.

Percentages in some tables may not total 100 due to rounding.

### Is your central bank working on a project involving big data?

It is a time of change for central banks and big data. Work involving big data can now be considered a mainstream activity: more than half of survey respondents said their central bank was working on a project in this area. For some central banks, this is a relatively new endeavour and their objectives are purely research-based, exploring prototypes and concepts. Other central banks are more advanced, examining how big data can influence policy.

The 29 central banks working on a project involving big data were drawn from across the economic spectrum, but industrial countries figured prominently, and more than half of the group were larger central banks – those with more than 1,000 members of staff. In their comments, several respondents described sophisticated frameworks in place for big data development.

A European central bank detailed the progress of its projects over the past two years: “In 2016, the statistics department launched the ‘big data forum’, which is intended to be for internal discussion. In roundtables, a team with specialists in different statistical domains tries to identify improvements to the statistics it produces using ‘unconventional’ sources of information and in that way encourages the development of big data projects. In 2017, the data integration and sharing unit was created.”

Similarly, a statistician at a large institution commented on several advances it had made: “We are working in different areas related to big data. We have conducted different projects using information obtained from Google Trends to improve forecasting. A project is also in progress analysing texts as a proxy for confidence indicators. We have used machine learning (random forests) in a project to measure the effects of fiscal anticipation in exploiting texts from television news. There is also an ongoing project in the field of media economics that uses unsupervised learning and intensive use of automatic text analysis. Moreover, some problems have been tackled using ‘big data techniques’ (such as neuronal networks) even if no big data as such has been used. Likewise, machine-learning techniques are being used to solve dynamic stochastic general equilibrium models with heterogeneous agents. More recently, a couple of initiatives have been started: one with web scraping to improve the information we have on the housing market, and another using machine learning to improve the quality of statistics.”

The dynamic nature of the field is highlighted by the novelty of the work: more than half of the 29 central banks reported that their projects were less than a year old. A respondent from a large central bank was testing the waters in this area with a prototype: “As part of our granular data model, we started a proof of concept to validate granular data.” The largest central bank reported that several projects are in progress.

Comments also indicated that administrative data – data collected via the work of governments and other official sector institutions – is a beneficial source for central banks starting out in the field. A respondent from the Americas noted their central bank’s data sources: “Tax information, which is used to compute a monthly activity index and yearly GDP.” A large central bank in Asia was interested in real estate developments: “Currently, we are using web-based and administrative data to understand developments in the property market.”

A central bank from a developing economy was keen to ensure data quality: “The central bank is working to optimise financial market statistics and monetary statistics in order to reduce duplication and contradictions in the data collected.” The largest institution in this survey commented that it was working on “loan loss risk analysis and big data collection for the financial institution’s supervision.”

### Has this project been initiated in the past 12 months?

For 13 central banks, these projects have been in process for a considerable period of time. These central banks tended to be smaller – 10 had fewer than 1,000 head of staff. The duration of these projects ranged from just over one year to as much as 10 years. An industrial-economy central bank commented: “This line of action has been active for some years, involving new and already existing data sources and business requirements.”

Twenty‑three central banks – 45% of respondents – are not involved in big data projects. This group largely comprised emerging-market central banks. In their comments it was clear that, for some, big data is an area of interest for the future, with several indicating work would soon start.

### Which best represents your central bank’s view of big data?

Big data is set to play an increasingly influential role in policymaking and supervisory processes. More than 60% of respondents said big data was either a “core input” or an “auxiliary input” into policymaking – an increase on the results of last year’s survey.3 Interestingly, the number of central banks that see it as an auxiliary input into policymaking is almost double that of last year, while the proportion that sees it as an active “area of research” has decreased three percentage points from the 2017 survey figure of 42%.

A central bank based in Oceania said: “Data captured should ultimately be a core input for policymaking and supervisory processes.” Administrative data is widely seen as an auxiliary input: “The central bank has a well-organised statistical business process model. So our database includes large amounts of information necessary for monetary and financial statistics production in line with the international standards. However, in some analytical materials, we use administrative data and other sources as an auxiliary input.”

Of those that see big data as a core input, half are from developing‑economy countries, with all of these institutions smaller than the survey average. Staff numbers on average were higher for the 19 central banks treating big data as an auxiliary input. In Asia, a central bank noted that its projects were focused on nowcasting: “We have used big data for nowcasting GDP, but we still need other traditional models or methods to forecast its long‑term trend.”

Twenty central banks said big data is an active area of research, with respondents from larger institutions typically holding this view. Many reasoned the projects were still in preliminary stages, and therefore big data was too underdeveloped to be used in policymaking. Several comments received, however, suggest this will change over time. A central banker from Europe said: “The central bank’s big data projects are mostly still at the research stage. If successful, they may become an auxiliary input into policymaking.”

Central banks’ views of big data and its role in policymaking has not generally shifted in the past 12 months; 90% of respondents said their view was unchanged.

### Does your central bank have a clear data governance structure for the management and collection of data?

Data governance remains a work in progress for central banks. More than 60% of respondents said their central bank – in their view – does not have a clear structure for data governance. This finding indicates a slight increase on last year’s survey. It is interesting to compare those with the ambition to use big data for policymaking purposes (61% of respondents) and their governance structure. Of the 61%, half do not have a clear data governance structure.

Of the 32 central banks that do not have a clear data governance structure in place, two-thirds are from emerging-market countries and are typically smaller institutions. Comments from these respondents can be broadly divided into three categories: employing a decentralised strategy, placing the responsibility solely with the statistics department and outsourcing the process. A central bank in Europe fell into the first category: “The central bank pursues, at the moment, a bottom‑up strategy. We have clear roles and responsibilities for each dimension of data governance, such as data quality and IT security in the responsible divisions.” A central banker from the Americas similarly commented: “Data governance is decentralised. Expertise is spread across departments with focal points in different departments.” A central banker in Africa added: “Each department has its own personnel and methodology of keeping/storing data.”

The second cohort noted that their statistics departments alone were responsible for big data projects, as this is their remit. Three central banks typified this sentiment in their comments. A small central bank said: “The central bank co-ordinates its statistical activities and data collections through its statistics department.” Equally, a Europe‑based central bank commented that its “statistics and reporting department is responsible for data management, but we do not have a clear structure.”

Outsourcing data management – the preference of the third group – was described by a statistician from a large central bank in Europe: “The initial data processing is procured and outsourced.”

Four central banks noted they are in the process of developing a governance structure. An African respondent said that their central bank is “currently working on [an] expanded structure to integrate statistics and other department stakeholders.”

At central banks with a governance structure, some respondents highlighted the novelty of this, and described how they look to prioritise and support big data projects by restructuring departments and maximising input to central bank policy. Implementation was a relatively new endeavour for two central banks in Europe; one introduced a governance structure last year: “In June 2017, the bank established the data governance committee chaired by the chief data officer and the first deputy of the bank.” The second example was put into place earlier this year: “As of April 2018, a new section has been formed to facilitate more effective data sharing and data analytics within the central bank.”

A central bank with just over 1,500 members of staff made considerable changes to accommodate big data developments: “The central bank has a governance structure for analytic data that supports its monetary policy and financial system functions. This includes a central data and statistics office and an analytic environment governance committee that oversees the use of financial resources to acquire data, software, hardware and to fund big data projects, among other things.”

The establishment of clear data governance structure has led one European central bank to reassess the roles and responsibilities of its employees: “The central bank has a clear role division between employees throughout the data phase: collection, processing, analysis and storage. Functions according to the role distribution are approved in job descriptions.”

### Is the data governance gap widening?

Three years of surveying central banks on big data has highlighted a gap between those that do not see their data governance as clear and those that do. This gap appears to be widening. In 2016, those who felt governance was not clear held the slimmest of majorities (see chart below). In 2017 this majority increased to seven and, in 2018, is more than 10. Fewer than 40% see their data governance as clear. So the immediate questions are: why? And: does it matter?

Taking the latter first, it is difficult to answer “no”, certainly given the interest and activity in big data and the importance central banks place in governance. So, while one can commend respondents for their honesty, from a policy perspective the trend is heading in the wrong direction.

Second, to the “why?”. This is harder, not least as the composition of survey respondents has changed over the three years. But, from comments to this question – and others – one can imagine cases where initiatives, policy interest and activity are running ahead of structure and management. Big data cuts across departments, divisions and policy. Quite simply, it is disrupting central banking, and changing governance structures to meet that disruption takes time and effort.

Where governance is clear, as an extended comment this year reveals, the structure reflects these rich seams of interest, but assigns clear responsibility at departmental level, with a single person responsible for overall co-ordination: “The bank has an integrated model of information management based on an information governance structure, which attributes the responsibility of operational management, in co-ordination with the IT department, to the statistics department. Each department of the bank has a data steward responsible for the content and management of the data it produces. There is also a master data steward, who co‑ordinates the activity and oversees general guidelines.”

Does your central bank have a clear data governance structure for the management and collection of data?

### Does your central bank have a single allocated budget for the handling of data – including big data?

Spending on data in central banks is, for the most part, decentralised. More than 80% of respondents do not have an allocated budget for handling data. In their comments, central bankers explained that the budget for data was either allocated per division/department or per project.

Two central banks have no assigned budget as the tasks are carried out by employees as part of their day-to-day work. This was the case for a central bank in Africa: “The bank employs statisticians and assistant statisticians who are mainly responsible for the collection – including management of database – and dissemination of data.”