IMF-季度GDP不可用国家的面板实况转播（英）-2023.8-36页_1mb

Summary of IMF Working Paper: Panel Nowcasting for Countries Whose Quarterly GDPs are Unavailable

Objective

This paper addresses the challenge of limited access to timely quarterly GDP data for many developing economies, including those in fragile and conflict-affected states, by proposing a panel nowcasting framework. The goal is to estimate quarterly GDP growth for data-scarce countries using statistical relationships derived from data-rich economies and supplementary nontraditional data sources, such as Google Trends and remote sensing data.

Methodology

A panel nowcasting approach is utilized, leveraging Light Gradient Boosting Regression (LGB) and Ordinary Least Squares (OLS) to estimate GDP growth based on high-frequency indicators like global commodity prices, search volumes, and economic uncertainty indices. The framework employs an ensemble method to handle missing data by averaging predictions from multiple specifications, including subsamples based on regions or economic characteristics, to reduce overfitting. Input variables are transformed into quarterly growth rates where possible, and SHAP values are used for interpretability.

Key Findings

• Country-wise out-of-sample nowcasts show potential for accuracy, with Theil U indices often below 50% for some economies, indicating significant reductions in forecast errors compared to naïve random walk models.
• Subsample analyses (e.g., Sub-Saharan Africa), improve performance when tailored to specific country groups.
• Quarterly and annual nowcasts for countries like Uganda and Sierra Leone align reasonably well with actual data, supporting policy insights, though limitations exist due to country-specific shocks and data gaps.
• Machine learning (LGB) generally outperforms OLS in accuracy, but country-specific factors (e.g., idiosyncratic events) can lead to weak nowcast performance in certain instances.

Limitations and Recommendations

• The framework is limited by large country-specific variations, which can introduce errors and bias, as seen in cases like Sierra Leone.
• Recommendations include optimizing subsample selection, improving data preparation, directly incorporating mixed-frequency annual data, and validating results with country-specific expert judgment. Future research should explore these enhancements.

Benefits

The approach provides real-time economic insights for data-poor countries, aiding policy responses faster than traditional methods, while nontraditional data helps fill gaps.