Scholarlink

<p>This study investigates the improvement of inflation forecasting precision by the incorporation of sophisticated time-series filtering methods (Hodrick–Prescott, Baxter–King, and Kalman) with several machine learning and deep learning frameworks (ANN, RNN, LSTM, GRU, CNN, Random Forest, and XGBoost). Monthly U.S. Federal Reserve economic statistics from 1959 to 2024 are utilized to compare models through mean squared error (MSE) and the coefficient of determination (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(R^2\)</EquationSource> </InlineEquation>). Our results indicate that Hodrick–Prescott filtering most efficiently eliminates noise and emphasizes the cyclical dynamics of inflation, allowing ensemble approaches to attain MSE <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(&lt; 0.001\)</EquationSource> </InlineEquation> and <InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(R^2&gt; 0.99\)</EquationSource> </InlineEquation>, while rendering neural networks viable competitors (<InlineEquation ID="IEq4"> <EquationSource Format="TEX">\(R^2&gt; 0.93\)</EquationSource> </InlineEquation>). Baxter-King filtering and multivariate Kalman methods exhibit instability and unnecessary complexity, resulting in inferior performance compared to unfiltered baselines. We recognize the Producer Price Index (PPI) as a principal leading indicator, capable of elucidating up to 84% of inflation variance with only 2 to 3 essential factors. These findings highlight the paramount significance of preprocessing quality relative to model complexity and provide an enhanced strategy for monetary policy development and economic planning.</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

Analysis of Improvements in Inflation Prediction Performance through the Use of Hybrid Filter Models and Many-to-one Neural Networks

  • 期刊论文

  • 发表日期：

    2026年1月27日

查看全文
  • A. J. Martínez Casares

摘要

This study investigates the improvement of inflation forecasting precision by the incorporation of sophisticated time-series filtering methods (Hodrick–Prescott, Baxter–King, and Kalman) with several machine learning and deep learning frameworks (ANN, RNN, LSTM, GRU, CNN, Random Forest, and XGBoost). Monthly U.S. Federal Reserve economic statistics from 1959 to 2024 are utilized to compare models through mean squared error (MSE) and the coefficient of determination ( \(R^2\) ). Our results indicate that Hodrick–Prescott filtering most efficiently eliminates noise and emphasizes the cyclical dynamics of inflation, allowing ensemble approaches to attain MSE \(< 0.001\) and \(R^2> 0.99\) , while rendering neural networks viable competitors ( \(R^2> 0.93\) ). Baxter-King filtering and multivariate Kalman methods exhibit instability and unnecessary complexity, resulting in inferior performance compared to unfiltered baselines. We recognize the Producer Price Index (PPI) as a principal leading indicator, capable of elucidating up to 84% of inflation variance with only 2 to 3 essential factors. These findings highlight the paramount significance of preprocessing quality relative to model complexity and provide an enhanced strategy for monetary policy development and economic planning.