Real GDP Growth Nowcasting in The Caribbean

Our people are committed and passionate about improving lives in Latin America and the Caribbean, and they get to do what they love in a diverse, collaborative and stimulating work environment. At the IDB, we’re committed to improving lives. Since 1959, we’ve been a leading source of long-term financing for economic, social, and institutional development in Latin America and the Caribbean. We do more than lending though. We partner with our 48-member countries to provide Latin America and the Caribbean with cutting-edge research about relevant development issues, policy advice to inform their decisions, and technical assistance to improve on the planning and execution of projects. For this, we need people who not only have the right skills but also are passionate about improving lives.

Learning goals

1. Understand the theoretical foundations of nowcasting and the rationale for combining traditional econometric and machine-learning approaches in small open economies with limited and lagged data.
2. Build a mixed-frequency, real-time data pipeline that integrates structured economic indicators with unstructured data sources (Google Trends, satellite nighttime lights, web-scraped text).
3. Pre-select variables in high-dimensional contexts using LARS and reduce dimensionality with PCA; choose the appropriate model family for each nowcasting challenge.
4. Estimate, interpret, and ensemble a comprehensive set of econometric and machine-learning nowcasting models — from OLS and Bayesian VAR through MIDAS-ML, Dynamic Factor Models, and LSTM neural networks.
5. Run, interpret, and visualize the full IDB Caribbean Nowcasting Tool end-to-end, and translate nowcast output into policy briefings that support real-time macroeconomic surveillance, research, and policy analysis.

Learning experience and activities

Intensive in-person training over four consecutive days, combining instructor-led sessions with guided hands-on labs in Python and R. Activities included:

• Installation and environment setup for Python and R, with real-time troubleshooting.
• Live coding walkthroughs of structured and unstructured data scraping, cleaning, and pre-processing.
• Guided exercises on high-dimensional variable selection (LARS) and dimensionality reduction (PCA) using Caribbean macroeconomic datasets.
• Hands-on implementation of econometric estimators (LASSO, Ridge, Elastic Net, Bayesian VAR, MIDAS, MIDAS-ML, Dynamic Factor Models) and machine-learning estimators (Decision Trees, Random Forests, Gradient Boosting Trees, LSTM neural networks).
• End-to-end execution of the IDB Caribbean Nowcasting Tool — from data input through ensemble forecast and visualization.
• Open Q&A and office hours for participants to apply the tool to their country of focus.

Content

• The mixed-frequency nowcasting problem: ragged-edge data, vintages, and information sets.
• Programming foundations for economic analysis: Python (pandas, numpy, scikit-learn) and R (tidyverse, dplyr).
• Data sourcing and pre-processing: API scraping, deflation, seasonal adjustment, and growth-rate transformations.
• Unstructured data for nowcasting: Google Trends and satellite nighttime lights (VIIRS).
• Variable pre-selection with Least Angle Regression (LARS) and dimensionality reduction with Principal Component Analysis (PCA).
• Econometric nowcasting models: OLS, LASSO, Ridge, Elastic Net, Bayesian VAR, MIDAS, MIDAS-ML, Dynamic Factor Models.
• Machine-learning nowcasting models: Decision Trees, Random Forests, Gradient Boosting Trees, LSTM neural networks.
• Nowcasting pipeline architecture: single-model walkthrough and scaled-up ensemble execution.
• Nowcast evaluation, visualization, and integration into policy briefings.

Assessment

This badge has been issued after the recipient successfully attended all sessions and submitted the final exercise.

Estimated effort

It took approximately 22 hours to complete this course.

Badge type

Completion badge