Scholarlink

<p>Factor-based approach using Principal Component Analysis (PCA) is an effective machine learning tool for forecasting with dimension reduction that has many applications in statistics, economics, and finance. This paper considers a Supervised Screening and Regularized Factor-based (SSRF) forecasting framework that targets high-dimensional predictor vectors with complex signal–noise structures. The framework adopts a structured four-step procedure that integrates both static and dynamic forecasting mechanisms to extract informative signals from noisy predictors. The static approach selects predictors via marginal correlation screening and scales them using univariate predictive slopes, while the dynamic approach screens and scales predictors based on time series regression with lagged predictors. PCA is subsequently applied to the scaled predictors to extract latent factors, followed by LASSO regularization to further refine predictive accuracy. Through simulation studies, we evaluate the effectiveness of the SSRF framework and identify practical strategies for parameter adjustment in high-dimensional settings. Using Chinese macroeconomic data from January 1996 to December 2019, we show that the SSRF method achieves competitive performance to several commonly used forecasting techniques.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Supervised Screening and Regularized Factor-Based Approach to Forecasting China’s Macroeconomic and Financial Indices

  • 期刊论文

  • 发表日期：

    2026年7月30日

查看全文
  • Sihan Tu,
  • Zhaoxing Gao

摘要

Factor-based approach using Principal Component Analysis (PCA) is an effective machine learning tool for forecasting with dimension reduction that has many applications in statistics, economics, and finance. This paper considers a Supervised Screening and Regularized Factor-based (SSRF) forecasting framework that targets high-dimensional predictor vectors with complex signal–noise structures. The framework adopts a structured four-step procedure that integrates both static and dynamic forecasting mechanisms to extract informative signals from noisy predictors. The static approach selects predictors via marginal correlation screening and scales them using univariate predictive slopes, while the dynamic approach screens and scales predictors based on time series regression with lagged predictors. PCA is subsequently applied to the scaled predictors to extract latent factors, followed by LASSO regularization to further refine predictive accuracy. Through simulation studies, we evaluate the effectiveness of the SSRF framework and identify practical strategies for parameter adjustment in high-dimensional settings. Using Chinese macroeconomic data from January 1996 to December 2019, we show that the SSRF method achieves competitive performance to several commonly used forecasting techniques.