The More the Merrier? A Machine Learning Algorithm for Optimal Pooling of Panel Data

IMF Working Papers, February 28, 2020

Source details

Canonical URL: The More the Merrier? A Machine Learning Algorithm for Optimal Pooling of Panel Data

Title: The More the Merrier? A Machine Learning Algorithm for Optimal Pooling of Panel Data
Authors: Marijn A. Bolhuis, Brett Rayner
Date: February 28, 2020
Core objective: Leverage insights from machine learning to optimize the tradeoff between bias and variance when estimating economic models using pooled datasets.
Key methodological claim: Develop a simple algorithm that estimates the similarity of economic structures across countries and selects the optimal pool of countries to maximize out-of-sample prediction accuracy of a model.
Empirical application: Nowcasting output growth with a panel of 102 countries.
Main empirical summary: The authors are able to significantly improve forecast accuracy relative to alternative pools.
Additional observation from the authors: The algortihm improves nowcast performance for advanced economies, as well as emerging market and developing economies, suggesting that machine learning techniques using pooled data could be an important macro tool for many countries.

Algorithm purpose:
- Estimate similarity of economic structures across countries.
- Select an optimal pool of countries to balance bias and variance for pooled estimators.
Intended outcome:
- Maximize out-of-sample prediction accuracy of the specified model.
Language in source (verbatim emphasis on algorithm naming variations as presented):
- "We develop a simple algorithm that estimates the similarity of economic structures across countries and selects the optimal pool of countries to maximize out-of-sample prediction accuracy of a model."
- "We apply the new alogrithm by nowcasting output growth with a panel of 102 countries..."
- "The algortihm improves nowcast performance for advanced economies, as well as emerging market and developing economies..."

Data and scope:
- Panel of 102 countries used for nowcasting output growth.
Performance findings:
- Significant improvement in forecast accuracy relative to alternative pools (as stated by the authors).
- Improvements observed across:
- Advanced economies
- Emerging market and developing economies

Practical implication: Machine learning techniques that optimally pool cross-country data can enhance macroeconomic nowcasting and forecasting for a wide range of countries.
Suggested value: Such methods could be an important macro tool for many countries, improving out-of-sample prediction accuracy by selecting more appropriate pooled samples.