IFC – Bank Indonesia International Workshop and Seminar on “Big Data for Central Bank Policies / Building Pathways

for Policy Making with Big

Data”

Bali, Indonesia, 23-26 July 2018

Big data for central banks 1

Bruno Tissot,
Bank for International Settlements

1 This presentation was prepared for the meeting. The views expressed are those of the author and do not necessarily reflect the views of the BIS, the IFC or the central banks
and other institutions represented at the meeting.
Big Data for Central Banks

Bruno TISSOT
Head of Statistics and Research Support, BIS
Head of Secretariat, Irving Fisher Committee on Central Bank Statistics (IFC)

International Workshop on Big Data for Central Bank Policies – Bali, 23-25 July 2018
Session 1

The views expressed are those of the author and do not necessarily reflect those of the BIS or the IFC.
Overview
 Introduction
 Financial Big Data
 Three key developments
 Challenges in handling and using big data
 Analysing CBs’ experiences
 Annexes: Selected references/ BD projects by CBs

2
 Introduction – Big Data…
• General & increasing policy interest for “Big Data” (BD)

→ the world’s most valuable resource is no longer oil, but data (The Economist)
• Term usually describes
 Extremely large data-sets
 Often a by-product of commercial or social activities
 Huge amount of granular information, typically transaction-level
 Data available in, or close to, real time
 Used to identify behavioural patterns / economic trends

→ Growing impact on information creation, storage, retrieval, methodology, analysis

3
Introduction – … for Central Banks…
• Private sector use big data to produce new & timely indicators

• New opportunities also for Central Banks (CBs) – as well as

macro-prudential authorities and financial supervisors?
 Broader and timelier range of indicators
 New statistical methodologies

• Extraction of new type of information supporting

 Economic forecasts & analysis
 Financial stability work
 Policy impact evaluation

4
Introduction – … with significant opportunities…
• Focus on sources that can effectively support micro- and macro-
economic as well as monetary and financial stability analyses
 Other big data – eg geospatial information – of lower interest

• Feedback loop inherent to policy-making authorities

 Big data sources can affect policy-making
 In turn policies implemented can generate new data-sets

• Big data provide new “business opportunities” for CBs, such as:
 Qualitative statements to decipher central banks’ communication
 Large number of big data pools generated by financial regulations
 In turn, big data can strengthen supervisors’ capacity
5
Introduction – … but also challenges…
• Specific challenges faced in handling and using big data
 Public nature of financial authorities and public trust
 Central banks concerned about ethical & reputational consequences
 Risk of misusing big data for policy actions?

• Different data quality concerns compared to private sector

 Ex: online retailers targeting potential customers based on past web
searches might find it acceptable to be “right” 20% of the time
 Such a low accuracy level looks inadequate for official statisticians

6
 Introduction – … not least due to security concerns…
• Increasing security concerns linked to internet / big data, such as:
 Risk that large private records of individual information could be accessed
and potentially misused by unauthorized third-parties
 Resilience of financial market infrastructures

• Influence on central banks’ actions

 Preserve public trust, especially when collecting data
 Supervise firms’ capability to gather and interpret security-related
information
 Set standards and best practices
 Promote cyber threat intelligence and modelling techniques

7
Introduction – … with the risk of being behind…
• CBs’ constraints compared to private firms
 Basic resources needs (IT budget, staff)
 Concerns about the lack of transparency in methodologies
 Poor quality of some data sources hampering public use

• IFC survey of central banks

 Big data work still on an exploratory mode
 Regular production of big data-based information likely to take time
 Yet increased interest esp. at senior policy level

8
Introduction – … and the need to be proactive
• Key objective for central banks is to better understand
 The new data-sets and related methodologies for their analysis
 The value added in comparison with “traditional” statistics

• Focus on pilot projects to assess how big data can help to

 Better monitor the economic and financial situation
 Enhance the effectiveness of policy
 Assess the impact of policy actions

• Possible tasks may well further expand

 Constant creation of new information/research needs
 Cf Haldane (2018): exploring behaviours in a “virtual economy”
9
I – What is Financial Big Data?
• Broad approach for BD: by-product of commercial or social
activities, providing a huge amount of very granular information

• Yet:
 Not sufficient to be large to qualify as “big data” – cf census
 Unstructured data require new tools to be processed
 Structured data-sets handled with “traditional” techniques?

• Choice of the relevant metric

 Volume of data?
 Specific characteristics of big data-sets?
 Timing issue? “Big data” 10 years ago versus today
10
I –Financial Big Data: 3 main BD groups…
• Definition by United Nations Department of Economic and Social
Affairs

• Big data type of information classified in three groups, as a

product of:
1. Social networks (human-sourced information, eg blogs, videos,
searches)
2. Traditional business systems (process-mediated data, such as data
produced by commercial transactions, e-commerce, credit cards)
3. The internet of things (machine-generated data, such as data produced
by pollution/traffic sensors, mobile phone information, computer logs)
11
I – Financial Big Data: … with a key distinction…

• 1. Unstructured data-set (often quite large):

 By-product of a non-statistical activity – “produced organically”
 Different from the datasets produced for traditional statistics, which are
structured by design

• 2. Data-set with large records, relatively well-structured

 Difficult to handle because of size, granularity or complexity
 Even “simple” structured datasets can benefit from big data techniques

12
I – Financial Big Data: … some judgment…
• Room for judgment, depends on features such as the “Vs”
 Volume (number of records and attributes)
 Velocity (speed of data production, eg tick data)
 Variety (for instance structure and format)
 Veracity (accuracy / uncertainty of large individual records)
 Valence (interconnectedness of the data)
 Value (often a by-product of an activity, can trigger a monetary reward)

• Features characterising big data can be very diverse

• Information content also quite heterogeneous

13
I – Financial Big Data: … 2 main sources for CBs…
• CBs see Big Data as comprising the variety of large-scale
information requiring/benefiting from “non-traditional” tools to
be processed & analysed
• Two data sources relevant for central banks:
 Restricted view: the “internet of things”-type of unstructured data,
heavily used by the private sector
→ Public interest eg Google Trends, but not really the core
 Large registers, by-products of 3 types of activities: financial,
commercial & administrative
→ Key issues include confidentiality and quality
14
I – Financial Big Data: … and 4 main types of BD-sets…

15
I – Financial Big Data: …with overlaps…
• Increasing part of the information collected on the web can be
the result of financial, commercial or administrative activities
• Cf recent expansion of “Fintech”
 “Technology-enabled innovation in financial services that could result in
new business models, applications, processes or products with an associated
material effect on the provision of financial services” (FSB, 2017)
 Parallel innovations: big data, mobile phone, internet, artificial intelligence
• Multiple applications that blur traditional boundaries
 Digital currencies (Bitcoin)
 Various applications in payments, crowdfunding, smart contracts, robot
advice, credit risk assessments & contract pricing
16
I – Financial Big Data: … practical issues…
• In practice CBs deal with various & heterogeneous “big data”
 Usually not directly produced for a specific statistical purpose, as in the
cases of traditional census or survey exercises
 Indirectly, data sources can be exploited for addressing statistical
information needs that may independently exist
 “Smart data”: treatment of the raw, “organic” data is key

• Public authorities just be at the beginning of making sense of all

the increasing volume and variety of data
 Use of specific data sources depend on policy questions
 Eg payment systems data: of interest to supervisors and tourism analysis

17
I – Financial Big Data: … complexity…
• Micro-level BD universe is complex and evolves over time
 Interaction between data available and specific policy needs

• Transforming data into information requires

 Merging different sources, with common identifiers
 Dealing with inconsistent observations
 Choosing a particular source
 Aggregating, by using parent relationship and rules
 Choices may depend on circumstances (“time dependency”)

18
I – Financial Big Data: … example
• Example: BIS International Debt Security issuance statistics
 Micro aggregation derived from large security-by-security data-sets
 Data collection based on a “traditional” residency concept…
 … and a "nationality basis“ (include debt issued by foreign affiliates)

• Constructing nationality-based statistics requires to

 Identify the perimeter of global firms
 Reclassify individual units
 Consolidate granular information at the group level
 Tasks both time-dependent and source-dependent

→ Handling large & complex data can benefit from BD techniques

19
II – Three key developments

• Big Data as a result of the combination of three key

developments in the financial area
The internet of things
Digitalisation
Expansion of micro financial data-sets in the aftermath of the
Great Financial Crisis (GFC) of 2007-09

20
II – 3 Developments: Internet of things (1: new data)

• Significant experience in recent years in collecting information

generated by the wide range of web and electronic devices
 Search queries, clicks on specific pages, display of information and text
online, social media messages…

• Can be used to complement “standard” statistical processes

 In general, the (near) real time availability of web data can allow for
getting rapid information and improving timeliness
 Approach to estimate current patterns and forecast them in advance of
actual publication dates (nowcasting exercises)

21
II – 3 Developments: Internet of things (2: inflation)
• Example: “scraping” prices posted online by retailers
 Exercises typically limited to specific inflation components (eg volatile
fresh vegetables’ prices)
 Process appears robust, scalable and can be automatised
 Important challenges: capturing unit-level prices, product characteristics,
quantities, adequate weights

• Billion Prices Project (MIT):

 Enhanced international comparisons of price indexes
 Dealing with measurement biases
 Addressing distortions in international relative prices

22
II – 3 Developments: Internet of things (3: house prices)
• Example: collection of housing prices on the web
 Scraping prices displayed by real estate agencies
 Capturing the various housing characteristics posted in advertisements
facilitates the calculation of quality effects (hedonic prices)
• Challenges
 Collecting the information in a comprehensive & structured way
 Weighting schemes

→ Particularly relevant for economies with less developed statistics

→ In more advanced countries also: property prices often derived from low
frequency surveys / for a limited number of cities
23
II – 3 Developments: Internet of things (4: real activity)
• Real-side economic indicators
 Job web announcements: indicators of business activity & unemployment
 Monitor consumption of durable goods (eg cars)
 Overall level of the economy / specific sectors (eg tourism) / areas

• But this use has been relatively incremental and limited, even
for national statistical agencies in advanced economies, and often
targeted at:
 Methodological improvements (eg quality adjustment)
 Reducing reporting lags and data revisions
 Alternative to the organisation of large surveys (eg India)
24
II – 3 Developments: Internet of things (5: new insights)
• Possibility of capturing unsuspected data patterns
 “Traditional” statistical modelling to infer economic relationships
 BD algorithms to incorporate various effects without ex ante assumptions
 Techniques can be implemented easily and in an automated way

• Opportunity to incorporate qualitative information

 Clicks from web searches, twitter messages or posts on social medias
 Incorporating sentiment and agents’ expectations for measuring risks,
changing preferences, causality patterns…
 Factors that play an important role during crises and are quite difficult to
model (non-linearities, network effects)

25
II – 3 Developments: Internet of things (6: drawbacks)
• Data quality issues
 Errors, typos and self-fulfilling expectations
 Need to collect consistent information but goods are not kept identical
 Announcement prices can differ from actual transaction prices
 Advertisements remain posted after economic transactions are settled
 Accuracy of the information that individuals (or robots!) input to the web

• Key limitation is that the data are not well structured
 Details on the location of a transaction / job offer difficult to get
 Underlying information can be collected several times

26
II – 3 Developments: Internet of things (7: challenges)
• Technical challenges
 Use of new techniques (eg web-scraping) and methodologies

• More fundamental challenges?

 Limited interpretability of “black box” calculations
 Mining data and the need to derive meaningful conclusions from an
economic perspective
 CBs need to present a consistent “story” when communicating policy

27
II – 3 Developments: Digitalisation (1: new information)
• Expanded access to digitalised information
 Rise in textual information moving to the web (while not produced by
internet activities strictly speaking)
 Reference documents can be digitalised, accessed and analysed like
“web-based” indicators
• Can be more easily and automatically exploited through ad hoc
BD techniques: eg text semantic analysis
 Extraction of textual information of interest
 Characterising text attributes and similarities
 Classifying information content (eg tone of central banks’ messages)
 Assessing the impact of external factors (eg circumstances, policy actions)
28
II – 3 Developments: Digitalisation (2: new opportunities)
• Techniques can also be used to measure impact on economic
agents’ expectations
• Structured way to assess policy communication
 Perceived stance of public authorities’ communication
 Impact of this communication / action in view of the messages expressed
in reaction by stakeholders
 Formation of public expectations

→ Complement traditional “event studies” (eg central bank actions)

→ Provides opportunities for fine-tuning policy communication
29
II – 3 Developments: New financial statistics (1: a revolution?)
• Revolution in financial statistics observed since the GFC
 Limitations of aggregated data: consider those institutions that are
systemic on an individual basis
 Need to measure the distribution of macro indicators, look at “fat tails”
and go “beyond the aggregates”
 Revolution comparable to the 1930s for the real accounts?

• Unprecedented efforts to collect more information on the

financial sector – the Data Gaps Initiative (DGI) endorsed by G20
 High demand for large, granular and complex data-sets
 Collected at the level of institutions, transactions & instruments

30
II – 3 Developments: New financial statistics (2: CBs’ interest)

• Fundamental factors explaining why CBs’ have been leading

the way for collecting such financial big datasets
 Go beyond aggregated indicators
 Make a better use for policy of available/expanding micro-level datasets
 Realisation that a huge amount of information is already available and
could be better exploited (eg administrative data)

• Focus on very granular information, derived from various

sources, and more complex compared to “typical” web-based
data

31
II – 3 Developments: New financial statistics (3: CCRs)
• Example: rising demand for detailed loan-by-loan / security-
by-security information
 Central credit registries (CCRS) have become the largest data-sets
maintained by some central banks
 Europe’s AnaCredit: “analytical credit dataset”
 US FRBNY Consumer Credit Panel: detailed information on consumer
debt and credit derived from individuals’ reports
• Data are well structured, but reporting is highly granular
 Multiple attributes: 200 attributes per data point on a monthly basis (and
on a daily basis for a subset) for AnaCredit
 Often complex to aggregate / analyse
32
II – 3 Developments: New financial statistics (4: specificities)
• Information often derived from confidential operations (tax
registers, banks’ books)
 Richness across the population of interest (eg capturing very small
enterprises)
 Usually collected regularly over a long period of time
 But need for anonymization / confidentiality protection
• CBs learning from private sector
 Increased experience in dealing with large data-sets (eg production of
“stress tests”)
 Supervisors of financial firms to develop their expertise in these areas too
33
III – Challenges
• Handling big datasets requires significant resources and proper
arrangements for managing the information
• Using big data in policy-making creates opportunities but is not
without risks
• Key implications
 Explains why public authorities’ actual use of big data is still limited, at
least in comparison to the private industry
 Significant time and effort needed before any regular production of big
data-based information for supporting CBs’ statistical and analytical work
on a large scale

34
III – Challenges in handling big data (1)
• Resources and proper arrangements for managing BD
 Sheer size of the data-sets
 Lack of structure
 Often limited quality of raw data

• The statistical production process itself has to be adapted

 Work to appropriately collect, clean, reconcile and store BD
 Usually, BD produced without standard quality controls of “traditional”
statistics (while public authorities put a lot of attention on those issues)
 Significant number of false/inconsistent/missing records

35
III – Challenges in handling big data (2)
• Need to set up a clear and comprehensive information
management process
 Data acquisition
 Data preparation
 Data processing
 Data validation

• A major area is IT
 Large processing costs, difficult & expensive technology choices
 Sophisticated statistical techniques: “BD algorithms”, “ML techniques”, “AI”
 Public authorities with less budget compare to private sector
36
III – Challenges in handling big data (3)
• New issues in terms of confidentiality protection and security
 Large amount of data provided by users through their web-based activities
 Large financial datasets require the handling of transaction-level,
potentially highly confidential, information
 Data privacy issues may increase with the development of big data and
Fintech firms

• Potentially wider implications

 Operational incidents can lead to significant privacy and legal issues, with
financial consequences
 Cf European General Data Protection Regulation (GDPR, 2018)
37
III – Challenges in handling big data (4)
• A key risk: reputation risk
 Peculiar position of central banks if private information is reported to
them but not protected adequately
 Especially for regulatory-type data collections
 But internet-based information, often a by-product of commercial
activities, can also pose significant legal, financial, reputational & ethical
issues
• Operational implications
 Public statisticians tend to be “cloud computing-adverse”
 Preference to operate in a “secluded” data environment
 But could reduce opportunities to use BD techniques in the marketplace
38
III – Challenges in handling big data (5)
• Ongoing substantial internal organisational changes to deal
with big data
 Cf creation of internal centres for big data statistics, “data lakes”, “internal
clouds”

• Another key area is staff

 Various skills needed: IT, data science and methodology, legal expertise…
 A “war for talent”? A competition with the private sector that may be
difficult for CBs…
 Additional issues: compensation, career path, management

39
III – Challenges in handling big data (6)
• How to enhance existing information management processes?
 Goal: flexible production of relevant information out of data points
 “Traditional”, template-driven data collections to be replaced by accessing
granular data from various sources

• Requirements
 Greater harmonisation of data-sets, statistical standards, identifiers and
dictionaries
 International efforts eg to develop global Legal Entity Identifiers &
automated data exchanges standards (XBRL, SMDX, ISO 20022)

40
III – Challenges in handling big data (7)
• Better integration of various IT systems among both authorities
and reporting entities

• Recent “Fintech innovations” to facilitate secure data transfer

mechanisms:
 Distributed ledger technology (DLT) to enable network participants to
securely propose, validate and record information to a synchronised
ledger distributed across the network
 Each transaction can be recorded in a batch (a “block”) and added to the
full transactions’ history (the “blockchain”)

• Involvement of private service providers (“regtech” industry)

41
III – Challenges in using big data (1)

• Big data opportunities for policy use but is not without risks
 Immediate benefits: lower production costs, new insights, production
speed
 To be balanced against potential large economic and social costs of
misguided policy decisions

• Key question: does “big data” provide a more accurate picture of

economic reality?

42
III – Challenges in using big data (2)

• Risk of conveying a false sense of accuracy and precision

 Problem exacerbated by the organic nature of BD:
 Data often self-reported or by-product of social activities
 Coverage bias unknown, can be significant (eg social media users)

→ Extremely large big data samples may thus compare

unfavourably with (smaller) traditional probabilistic samples –
precisely designed to be representative of the population of interest

→ Key misperception of the intrinsic value of big data

43
III – Challenges in using big data (3)
• Risk of undermining public policy?
 Effectiveness (if data are providing wrong signals)
 Reputation/legitimacy

• Might systematically alter decision-making?

 Greater ability to monitor the economy in real time: bias towards
responding quickly and more frequently to news, encouraging shorter
horizons?
 Greater reliance on “big data”-based analyses of sentiment: risk of
excessively fine-tuning policy communication based on perceived
expectations rather than actual economic developments?

44
IV – Analysing CBs’ experiences (1)
• Proper information management frameworks needed to make
the most of big data so as to:
 Address challenges faced when handling and using big data
 Avoid the risk of focussing on cumbersome management tasks – cleaning,
documenting, organising data – instead of using the information

• Key is to make sense of the data collected, with a coherent

information management framework

• CBs are following step-by-step approaches, with specific use

cases instead of “big bang” solutions
45
IV – Analysing CBs’ experiences (2)
• Pressing challenges
 Combination of internet / digitalisation / new post-crisis initiatives
 Authorities just at the beginning of making sense of the increasing
volume, granularity and variety of information
 “Connecting the dots is as important as collecting the dots, meaning the
right data” (Caruana, 2017)

• Fundamental distinction between “data” and “information”

 “Traditional” official statistics were “designed data” collected for a specific
purpose
 By definition were organised in order to extract meaningful information
 Key difference with “organic”-type big datasets
46
IV – Analysing CBs’ experiences (3)
• Avoiding the risk of confusing “data” and “information”
 Need to complement, not replace, designed data with (organic) BD sets
 Calls for ensuring a continuum: from the collection of BD to statistical
processing and the extraction of valuable information for policy use

• Various ingredients:
 Proper IT infrastructure
 Adequate statistical applications (including big data analytics)
 Legal and HR support in terms of skill-sets
 Good co-ordination to have a consistent and holistic information
production chain

47
IV – Analysing CBs’ experiences (4)
• Central banks have already started to rethink their information
management processes to:
 Be able to access internet data-sets and big data techniques
 Handle the new data collections initiated after the GFC

• No one-size-fits-all approach, as it depends on

 Characteristics of each data collection
 Country circumstances
 Actual policy needs

48
 Annex (1): Selected references
Bank for International Settlements (BIS) (2018). Cryptocurrencies: looking beyond the hype, BIS Annual Economic Report, chapter V.
Bean, C. (2016). Independent review of UK economic statistics, March.
Bholat, D. (2015). Big data and central banks, Bank of England, Quarterly Bulletin, March 2015.
Borio, C. (2013). The Great Financial Crisis: setting priorities for new statistics, Journal of Banking Regulation.
Caruana, J. (2017). International financial crises: new understandings, new data, Speech at the National Bank of Belgium, Brussels, February.
Cavallo, A., & Rigobon, R. (2016). The Billion Prices Project: Using Online Prices for Measurement and Research, Journal of Economic Perspectives, Spring 2016, Vol 30(2):
151-78.
Cœuré, B. (2017). Policy analysis with big data, speech at the conference on “Economic and Financial Regulation in the Era of Big Data”, Banque de France, Paris, November.
Financial Stability Board (FSB) (2017). Financial Stability Implications from FinTech.
Glass, E. (2016): Survey analysis – Big data in central banks, Central Banking Focus Report, 2016.
Haldane, A. G. (2018). Will Big Data Keep Its Promise? Speech at the Bank of England Data Analytics for Finance and Macro Research Centre, King’s Business School, 19
April.
Hammer, C., Kostroch, D., Quiros, G., & Staff of the IMF Statistics Department (STA) Internal Group (2017). Big data: potential, challenges, and statistical implications, IMF
Staff Discussion Note, Staff Discussion Notes (SDN)/17/06, September.
Hill, S. (2018). The Big Data Revolution in Economic Statistics: Waiting for Godot... and Government Funding, Goldman Sachs US Economics Analyst, 6 May.
Irving Fisher Committee on Central Bank Statistics (IFC) (2015). Central banks’ use of and interest in ‘big data’, October.
Irving Fisher Committee on Central Bank Statistics (IFC) (2017). Proceedings of the IFC Satellite Seminar on “Big Data” at the ISI Regional Statistics Conference 2017, IFC
Bulletin, no 44, September.
Meng, X. (2014). A trio of inference problems that could win you a Nobel Prize in statistics (if you help fund it), in Lin, X., Genest, C., Banks, D., Molenberghs, G., Scott D., &
Wang, J.-L. (eds), Past, present, and future of statistical science, Chapman and Hall, 2014, pp 537–62.
Nymand-Andersen, P. (2015). Big data – the hunt for timely insights and decision certainty: Central banking reflections on the use of big data for policy purposes, IFC Working
Paper, no 14.
The Economist (2017). The world’s most valuable resource is no longer oil, but data, 6th May edition.

49
Annex Big data areas Types of data-sets
Foreign trade operations /
investment transactions
Examples of projects
Balance of payments statistics eg tourism, exports
(2): Taxation / payroll / unemployment Employment, wages, business formation (SMEs)
insurance
Selected Administrative Central balance sheet offices Performance vulnerabilities assessment
records
BD Loans registers Measurement of credit risk, FX exposures

projects Financial market supervisors

Public financial statements
Network analysis, exposures
Corporate balance sheet, group-level supervision
by central Financial market activity indicators
Internet clicks
Payments systems, Trade repositories
Google searches
banks social networks
Digitalised content / text
confidence & economic sentiment
policy communication , analysis of expectations
Web-based Websites’ scraping Various uses
indicators Job portals Employment / activity
Prices posted directly on websites Measure specific components of the CPI, PPIs, Inflation nowcasting /
forecasting,
Pricing strategy analysis
Real estate agencies House price indices
Credit card operations Payments patterns, Tourism
Commercial Mobile operators
data-sets Mobile positioning data (eg travelers’), Financial inclusion
Geo spatial information National statistical system Tasks
Credit institutions Balance sheet exposures, Investor behaviour/expectations
Settlement operations Operational risks, Market functioning
Securities issuance Security-by-security databases
Financial data-
sets Market liquidity Bid/ask spreads
Custodians records Securities holding statistics
Tick-by-tick data Real-time analysis of financial patterns

50
 Thank you!!

Questions?
[email protected]
[email protected]

51