Scholarlink

<p>Accurate inflation forecasting is critical for economic decision-making by individuals, institutions, and policymakers. Recently, machine learning (ML) techniques have gained attention for their potential to enhance forecasting performance. However, no existing Systematic Literature Review (SLR) has synthesized the collective effectiveness of these models. This article addresses that gap by conducting a comprehensive SLR of ML-based approaches to inflation forecasting, structured around four dimensions: (i) forecasting trends (indicators predicted, univariate vs. multivariate, etc.), (ii) datasets and evaluation methods and metrics, (iii) ML techniques and their comparative performance against each other and traditional models, and (iv) capabilities of ML models in terms of strengths and weaknesses. Following the guidelines of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA), we analyze 96 studies published between 2008 and June 2024, covering four forecast horizons (1, 3, 6, and 12&#xa0;months). The findings confirm that model performance is horizon-dependent. Lasso Regression excels at short- and mid-term horizons, while XG-Boost and Random Forest dominate long-term forecasts. LSTM performs well at 3&#xa0;months but declines beyond, whereas deep MLPs show potential at mid-range horizons when adequately tuned. Ridge Regression improves with horizon length, while Elastic Net, SVR, and shallow MLPs generally underperform. Overall, ML models outperform traditional methods and exhibit variable performances depending on temporal dynamics and data structure. The review underscores the importance of aligning model choice with forecasting horizon, data characteristics, and optimal model tuning. It also offers practical guidance by identifying top-performing models per horizon, aiding researchers and practitioners in related fields.</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

Modeling inflation with machine learning: a cross-horizon systematic review

  • 期刊论文

  • 发表日期：

    2025年12月9日

查看全文
  • Hassenet Slimani,
  • Chemseddine Chourabi

摘要

Accurate inflation forecasting is critical for economic decision-making by individuals, institutions, and policymakers. Recently, machine learning (ML) techniques have gained attention for their potential to enhance forecasting performance. However, no existing Systematic Literature Review (SLR) has synthesized the collective effectiveness of these models. This article addresses that gap by conducting a comprehensive SLR of ML-based approaches to inflation forecasting, structured around four dimensions: (i) forecasting trends (indicators predicted, univariate vs. multivariate, etc.), (ii) datasets and evaluation methods and metrics, (iii) ML techniques and their comparative performance against each other and traditional models, and (iv) capabilities of ML models in terms of strengths and weaknesses. Following the guidelines of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA), we analyze 96 studies published between 2008 and June 2024, covering four forecast horizons (1, 3, 6, and 12 months). The findings confirm that model performance is horizon-dependent. Lasso Regression excels at short- and mid-term horizons, while XG-Boost and Random Forest dominate long-term forecasts. LSTM performs well at 3 months but declines beyond, whereas deep MLPs show potential at mid-range horizons when adequately tuned. Ridge Regression improves with horizon length, while Elastic Net, SVR, and shallow MLPs generally underperform. Overall, ML models outperform traditional methods and exhibit variable performances depending on temporal dynamics and data structure. The review underscores the importance of aligning model choice with forecasting horizon, data characteristics, and optimal model tuning. It also offers practical guidance by identifying top-performing models per horizon, aiding researchers and practitioners in related fields.