imf experience with big data and machine learning · big data . build the global data commons. be...
TRANSCRIPT
![Page 1: IMF Experience with Big Data and Machine learning · Big Data . Build the Global Data Commons. Be agile. in identifying. data needs for . effective surveillance. Aim. at greater cross-country](https://reader031.vdocuments.us/reader031/viewer/2022040212/5e7cfd0ff3820661ac7d62f8/html5/thumbnails/1.jpg)
IMF Experience with Big Data
![Page 2: IMF Experience with Big Data and Machine learning · Big Data . Build the Global Data Commons. Be agile. in identifying. data needs for . effective surveillance. Aim. at greater cross-country](https://reader031.vdocuments.us/reader031/viewer/2022040212/5e7cfd0ff3820661ac7d62f8/html5/thumbnails/2.jpg)
Overview of the presentation
Background – why are we now focusing on big data?
Change – how are we responding to the challenge?
Examples – what big data projects?
![Page 3: IMF Experience with Big Data and Machine learning · Big Data . Build the Global Data Commons. Be agile. in identifying. data needs for . effective surveillance. Aim. at greater cross-country](https://reader031.vdocuments.us/reader031/viewer/2022040212/5e7cfd0ff3820661ac7d62f8/html5/thumbnails/3.jpg)
The Background – How we did things in the past…… Traditionally the IMF has depended almost entirely on structured data
Two kinds of data: Operational data
Official data
Recurring crises – Common feature – data gaps
IMF handling huge amounts of structured data
This structure has posed a constraint on our efforts to use big data
![Page 4: IMF Experience with Big Data and Machine learning · Big Data . Build the Global Data Commons. Be agile. in identifying. data needs for . effective surveillance. Aim. at greater cross-country](https://reader031.vdocuments.us/reader031/viewer/2022040212/5e7cfd0ff3820661ac7d62f8/html5/thumbnails/4.jpg)
How are we addressing the issue?Recently, the IMF has introduced a number of initiatives to take advantage of the potential with big data.
2016 Staff Discussion Note: “Big Data: Potential, Challenges and Statistical Implications”
2018 New data strategy
2019 Establishment of a new Data Services division in the Statistics Department with
responsibility for big data Creation of a Data Scientist position within the IMF “Job family” Establishment of a “Community of interest” within the institution Innovation lab New Infrastructure
![Page 5: IMF Experience with Big Data and Machine learning · Big Data . Build the Global Data Commons. Be agile. in identifying. data needs for . effective surveillance. Aim. at greater cross-country](https://reader031.vdocuments.us/reader031/viewer/2022040212/5e7cfd0ff3820661ac7d62f8/html5/thumbnails/5.jpg)
Big Data in the Fund
New questions and new indicators
Bridging time lags in the availability of official statistics
Supporting the timelier forecasting of existing indicators
As an innovative data source in the production of official statistics
![Page 6: IMF Experience with Big Data and Machine learning · Big Data . Build the Global Data Commons. Be agile. in identifying. data needs for . effective surveillance. Aim. at greater cross-country](https://reader031.vdocuments.us/reader031/viewer/2022040212/5e7cfd0ff3820661ac7d62f8/html5/thumbnails/6.jpg)
Overarching Strategy
Support the use of Big Data
Build the Global Data Commons
Be agilein identifying
data needs for effective
surveillance
Aimat greater
cross-country data
comparability
Address data weaknesses
Ensure secure and
seamless access and sharing of
high quality data
6
New Data Strategy
Approved by the IMF Executive Board on March 9, 2018
Supporting big data is a strategy priority in the Fund
![Page 7: IMF Experience with Big Data and Machine learning · Big Data . Build the Global Data Commons. Be agile. in identifying. data needs for . effective surveillance. Aim. at greater cross-country](https://reader031.vdocuments.us/reader031/viewer/2022040212/5e7cfd0ff3820661ac7d62f8/html5/thumbnails/7.jpg)
The Strategy Sets Three Recommendations on Big Data
Address Skill Gaps
Support Big Data in Fund Operations
Peer-Learning
![Page 8: IMF Experience with Big Data and Machine learning · Big Data . Build the Global Data Commons. Be agile. in identifying. data needs for . effective surveillance. Aim. at greater cross-country](https://reader031.vdocuments.us/reader031/viewer/2022040212/5e7cfd0ff3820661ac7d62f8/html5/thumbnails/8.jpg)
New organizational structure
Review of our data function
Established a Data Services Division in Statistics with responsibility for big data.
Review of job competencies Data Scientist
Data Architect
Data Analyst/Officer
![Page 9: IMF Experience with Big Data and Machine learning · Big Data . Build the Global Data Commons. Be agile. in identifying. data needs for . effective surveillance. Aim. at greater cross-country](https://reader031.vdocuments.us/reader031/viewer/2022040212/5e7cfd0ff3820661ac7d62f8/html5/thumbnails/9.jpg)
Big data “Community of Interest”
Inter-departmental forum to share learning and results
Focus on developing skills as much as highlighting research
A community of “equals” Works outside the IMF departmental structure
No direct management review
![Page 10: IMF Experience with Big Data and Machine learning · Big Data . Build the Global Data Commons. Be agile. in identifying. data needs for . effective surveillance. Aim. at greater cross-country](https://reader031.vdocuments.us/reader031/viewer/2022040212/5e7cfd0ff3820661ac7d62f8/html5/thumbnails/10.jpg)
New Innovation Lab
Previous approach Ideas for change reviewed through departmental structures High costs of failure
New approach Innovation “competitions” Share ideas (TED-Talks, forums) Funds available to “try things out” It is “OK to fail”
Practical impact – lots of big data ideas came through the innovation lab 109 ideas submitted 10 ideas selected 6 ideas tested with a proof of concept
![Page 11: IMF Experience with Big Data and Machine learning · Big Data . Build the Global Data Commons. Be agile. in identifying. data needs for . effective surveillance. Aim. at greater cross-country](https://reader031.vdocuments.us/reader031/viewer/2022040212/5e7cfd0ff3820661ac7d62f8/html5/thumbnails/11.jpg)
Some big data projects
Lots of decentralized projects by individual researchers
Limited resources available
Innovation lab Providing resources
Know-how
Forum for results
SWIFT
• Global flow of funds
Sentiment indicators (Google Trends)
• Nowcasting; High Frequency Indicators
Week at the beach
• Tourism index
Scanner data (Indonesia)
• CPI, retail sales
![Page 12: IMF Experience with Big Data and Machine learning · Big Data . Build the Global Data Commons. Be agile. in identifying. data needs for . effective surveillance. Aim. at greater cross-country](https://reader031.vdocuments.us/reader031/viewer/2022040212/5e7cfd0ff3820661ac7d62f8/html5/thumbnails/12.jpg)
SWIFT
Built SWIFT messages database
Provides real transaction data on payments and trade between countries around the world.
Downside – access to this data is very expensive
![Page 13: IMF Experience with Big Data and Machine learning · Big Data . Build the Global Data Commons. Be agile. in identifying. data needs for . effective surveillance. Aim. at greater cross-country](https://reader031.vdocuments.us/reader031/viewer/2022040212/5e7cfd0ff3820661ac7d62f8/html5/thumbnails/13.jpg)
Sentiment indicators
Google Private API provided by Google for Social Science Research.
Used to nowcast various macro-indicators.
Central bank announcements
Financial Times – “Fear” indicators used in the vulnerability exercise
![Page 14: IMF Experience with Big Data and Machine learning · Big Data . Build the Global Data Commons. Be agile. in identifying. data needs for . effective surveillance. Aim. at greater cross-country](https://reader031.vdocuments.us/reader031/viewer/2022040212/5e7cfd0ff3820661ac7d62f8/html5/thumbnails/14.jpg)
Tourism data
Composite price index (basket of items typically consumed on a beach holiday) in collaboration with TripAdvisor
Recent examples of applications to Fund policy work include WHD REO 2016,
2018 Grenada article IV report
2017 Maldives article IV.
![Page 15: IMF Experience with Big Data and Machine learning · Big Data . Build the Global Data Commons. Be agile. in identifying. data needs for . effective surveillance. Aim. at greater cross-country](https://reader031.vdocuments.us/reader031/viewer/2022040212/5e7cfd0ff3820661ac7d62f8/html5/thumbnails/15.jpg)
Developed an interactive database available to research staff
![Page 16: IMF Experience with Big Data and Machine learning · Big Data . Build the Global Data Commons. Be agile. in identifying. data needs for . effective surveillance. Aim. at greater cross-country](https://reader031.vdocuments.us/reader031/viewer/2022040212/5e7cfd0ff3820661ac7d62f8/html5/thumbnails/16.jpg)
Indonesia Scanner data project
Joint project with Statistics Netherlands and Bank Indonesia
Part of our Technical Assistance
Objectives: Use scanner data to obtain high frequency estimates of the CPI
Obtain more robust high frequency retail sales estimates
Project ongoing – initial results are promising, especially for the CPI
![Page 17: IMF Experience with Big Data and Machine learning · Big Data . Build the Global Data Commons. Be agile. in identifying. data needs for . effective surveillance. Aim. at greater cross-country](https://reader031.vdocuments.us/reader031/viewer/2022040212/5e7cfd0ff3820661ac7d62f8/html5/thumbnails/17.jpg)
Concluding remarks
The institution has recognized the need to develop capacity in the area of big data
This required some organization changes New data strategy
Dedicated division
New career stream
Lots of projects were underway, but decentralized.
Need to create space for innovation
Last issue – infrastructure – we are starting to think about it now……