## Big Data in Macroeconomic Statistics - From Innovation to Capacity Development

## Source details

**Canonical URL:** [Big Data in Macroeconomic Statistics - From Innovation to Capacity Development](https://www.imf.org/-/media/files/capacity-developement/d4d/9th-meeting/big-data-for-macroeconomic-statistics-from-innovation-to-capacity-development.pdf)

## Other formats

- [Markdown version](/-/media/files/capacity-developement/d4d/9th-meeting/big-data-for-macroeconomic-statistics-from-innovation-to-capacity-development.pdf.md)
- [Structured JSON version](/-/media/files/capacity-developement/d4d/9th-meeting/big-data-for-macroeconomic-statistics-from-innovation-to-capacity-development.pdf.json)

---

### Event and authorship
- Presentation dated JUNE 22, 2026 delivered at D4D FUND STEERING COMMITTEE MEETING by Iyke Maduako, Data Scientist, Big Data Center, Statistics Department, IMF.
- Working paper authors listed: Andinet Woldemichael & Iyke Maduako.
- Contact: BigDataCenter@imf.org.

### Every economic activity leaves a digital trace — Big data sources
- Satellite imagery indicators: NIGHTLIGHTS, NO2 EMISSIONS, VEGETATION INDEX (NDVI, EVI referenced elsewhere).
- Digital transactions: mobile money and other digital payment traces.
- Scanner data: prices.
- Administrative data: tax records and other administrative registers.
- Text data and Unstructured data (Natural Language Processing / Text Analysis).
- Ship and trade-related sensors: AIS/PortWatch (sensors on maritime vessels), Passenger tracking (Passengers on Board / FlightRadar24).
- Labor market data: hiring and skills.
- Internet connectivity and mobility data.

### IMF experience and impact on capacity development
- Big Data Center has trained more 180 officials from member country institutions across all regions, including central banks, national statistical offices, ministries of finance, and ministries of planning.
- Key training topics:
  - Overview of big data sources and analytical methods for satellite data, ship tracking data, unstructured/Text data, and online searches.
  - Nowcasting GDP.
  - PortWatch and AIS data to monitor trade disruptions and estimate trade.
  - Natural Language Processing / Text Analysis.
  - Strategic Planning for Big Data.
  - Partnerships and Collaborations on private sector data.
- Emphasis on prioritizing durable capacity in fragile, data-constrained (FCS) settings through hands-on sessions with Ministries of Finance and Customs.

### AI-enabled workflows and tools
- AI-enabled end-to-end workflows for trade statistics: from raw customs records to validation and visualization.
  - Use of Gemini in Google Colab to lower the technical barrier: staff process customs data with AI help; automated inconsistency checks; reproducible, transferable workflow.
  - Analysis & quality assurance via Power BI dashboards: trade indicators dashboard, validation against PortWatch estimates, production-ready, SNBS-led tools.
- Somalia National Bureau of Statistics example: AI for improving production workflows.
- Bank of Uganda example: An end-to-end automated and scheduled Nowcasting Framework.

### Big data applications for national accounts
- Predicting informal activity at granular level:
  - Multi-layer ML pipeline using Satellite, Infrastructure, Places, Population, GHSL data at District Level.
  - Satellite-derived indicators: Land Cover, NDVI, NDBI, Night-Time Light, GHSL BUILT ENV. (Built –S, Res and Non-res; Built_Characteristics; Built_Volume; Degree of Urbanization).
  - OpenStreetMap / Overture Places: Vendors, Kiosks, Markets, Food Stalls, Repair, Retail, Taxi, etc.
  - Infrastructure: Roads (Count/Type).
  - Training target: Country Informal Enterprise Survey Data.
  - Feature extraction and ML methods: RF / XGBoost etc.
  - Output: District-Level Predicted Informal Enterprise Density.
- Leveraging open-source buildings data to address scarcity of timely and granular residential building data where housing censuses are not always conducted, are outdated, or not geotagged.
- Working paper explores an application of Satellite-Based Census of Residential Buildings.

### Big data for high-frequency indicators and nowcasting
- Nowcasting scalable framework:
  - For countries where quarterly GDP data is unavailable or available with significant delays, satellite data and machine learning-driven nowcasting provide timely GDP estimates.
  - Inputs: Traditional and non-traditional high-frequency indicators (e.g., nighttime lights, NO2, vegetation indices).
  - Methodology: Horse racing over15 Machine learning models and selecting the best model based on the out-of-sample RMSE.
  - Output: Timely YoY GDP growth nowcasts.
  - Capacity development: IMF staff and country authorities in AFR, APD, MCD, EUR, and WHD have been trained on methodology.
  - NowHub is a collaboration between STADS / Big Data Center, AFR, RES, and other depts.
- Country examples: Lesotho (manufacturing, industry and other fossil fuel combustion proxies), Nigeria (agro-potential and yield proxies).

### AIS and PortWatch for trade nowcasting
- Goal: Build monthly global trade estimates (value & volume) aligned with how official statistics are compiled.
- Approach:
  1. Physical Volume: Use AIS data to estimate cargo volume (in metric tons) by vessel type: tankers, dry bulk, and containers.
  2. Trade Value: Multiply physical volume by vessel-type-specific unit values (US$/ton), derived from trade data.
  3. Trade Volume: Deflate trade value using export/import price indices to obtain trade volume in constant prices.
- Methodology attribution: Arslanalp et al (2025) — Nowcasting Global Trade from Space.
- PortWatch provides strong correlation with official trade estimates and enables early detection of trade patterns well before changes are reflected in official customs data.
- Data ecosystem components: Maritime Data Aggregator, AIS Receiver Network, AIS Satellite Network, UN Global Platform.

### Big data applications for climate risk and gridded statistics
- Ongoing projects include Gridded GDP (US and Canada examples) and Gridded Capital Stock (Bangladesh example).
- Outputs and tools:
  - Gridded GDP by industry.
  - Percentage buildings exposure to flooding by Area vs Value.
  - GeoPulse Event-Tracking for measuring exposure in near real-time.

### Key takeaways and conclusion
- Big Data Center supports country teams and authorities—regardless of local statistical capacity—with timely and trusted signals of economic activity to guide surveillance and policy advice.
- Emphasis on reproducible, transferable workflows, AI-assisted processing, capacity development, and use of multiple big data sources to augment and nowcast official macroeconomic statistics.

---


_Source: https://www.imf.org/-/media/files/capacity-developement/d4d/9th-meeting/big-data-for-macroeconomic-statistics-from-innovation-to-capacity-development.pdf_