Parameter Proliferation in Nowcasting: Issues and Approaches—An Application to Nowcasting China’s Real GDP
IMF Working Papers, October 24, 2025
Source details
- Canonical URL
- Parameter Proliferation in Nowcasting: Issues and Approaches—An Application to Nowcasting China’s Real GDP
Other formats
Bibliographic details
- Authors: Paul Cashin, Fei Han, Ivy Sabuga, Jing Xie, Fan Zhang
- Published: October 24, 2025
- Series: IMF Working Papers
- DOI: https://doi.org/10.5089/9798229027212.001
Research objective and scope
- Evaluate three approaches to address parameter proliferation in nowcasting:
- Variable selection using adjusted stepwise autoregressive integrated moving average with exogenous variables (AS-ARIMAX).
- Regularization in machine learning (ML).
- Dimensionality reduction via principal component analysis (PCA).
- Empirical application: nowcasting China's annualized real GDP growth rate.
- Data and estimation window:
- Utilizes 166 variables.
- Models estimated from 2007Q2 to 2019Q4 using rolling-window regression.
- Pseudo out-of-sample comparison for 2020Q1 to 2023Q1.
Methodology and models compared
- Variable-selection approach: AS-ARIMAX implemented within Bridge-type models.
- Regularization approaches: Ridge Regression, LASSO, and Elastic Net.
- Dimensionality reduction approach: principal component analysis (PCA) feeding into dynamic factor models (DFM).
- Nowcasting model families evaluated:
- Bridge
- MIDAS
- U-MIDAS
- Dynamic factor model (DFM)
- Machine learning techniques (Ridge Regression, LASSO, Elastic Net)
- Evaluation framework:
- Rolling-window regression estimation over 2007Q2–2019Q4.
- Pseudo out-of-sample performance comparison over 2020Q1–2023Q1.
Key findings
- LASSO results:
- The LASSO method outperforms all other models, but only when guided by economic judgment and sign restrictions in variable selection.
- Simpler models:
- Bridge models combined with AS-ARIMAX variable selection yield reliable estimates nearly comparable to those from LASSO.
- Effective variable selection is crucial for capturing strong signals, enabling simpler methods to perform well.
- Implication on approaches:
- Regularization (LASSO) can dominate when combined with domain knowledge and sign constraints.
- Dimensionality reduction via PCA and other techniques may be less effective than targeted variable selection in this application.
Practical takeaways and recommended practices
- Prioritize effective variable selection to capture strong predictive signals; AS-ARIMAX is a viable approach for Bridge-type models.
- When using ML regularization methods (e.g., LASSO), incorporate economic judgment and sign restrictions to improve performance.
- Simpler models with disciplined variable selection can achieve near-state-of-the-art nowcasting accuracy without full-scale ML pipelines.
Source: IMF Working Paper "Parameter Proliferation in Nowcasting: Issues and Approaches—An Application to Nowcasting China’s Real GDP" by Paul Cashin, Fei Han, Ivy Sabuga, Jing Xie, and Fan Zhang (October 24, 2025).
Content in this bundle
- Working Paper