Scholarlink

<p>The association between electricity consumption and economic growth has evolved substantially in high-income countries, where energy-efficient transitions and sectoral shifts have contributed to a decoupling of the two. This study explores the predictive capacity of electricity consumption in forecasting GDP across six major European economies, Germany, France, Italy, Spain, the Netherlands, and Switzerland, using both traditional econometric techniques and deep learning models. While Granger causality tests, vector autoregression (VAR), impulse response functions (IRF), and forecast error variance decomposition (FEVD) reveal weak or negligible structural links between electricity and GDP, these findings support growing evidence of decoupling in advanced economies. In response, we implement Long Short-Term Memory (LSTM) neural networks to discover non-linear, temporal patterns in electricity data that can improve short-term GDP forecasting. To ensure robustness, we applied a walk-forward validation strategy using a 20-quarter rolling window. While in-sample performance appeared strong, out-of-sample forecasting revealed poor generalizability in most countries, highlighting the limitations of electricity data alone. Our findings underscore the limitations of linear models in decoupled economic systems and emphasize the role of electricity consumption as a potentially useful, but insufficient, predictive input when processed through non-linear machine learning frameworks. These results carry important implications for policymakers, energy analysts, and economists concerned with real-time economic monitoring and sustainability transitions in Europe.</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

Watts and Wealth: Forecasting the Economic Pulse of Europe Through Electricity Consumption

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年9月11日

查看全文
  • Robin Kunju Mol Raj,
  • Marek Vochozka,
  • Yelyzaveta Apanovych

摘要

The association between electricity consumption and economic growth has evolved substantially in high-income countries, where energy-efficient transitions and sectoral shifts have contributed to a decoupling of the two. This study explores the predictive capacity of electricity consumption in forecasting GDP across six major European economies, Germany, France, Italy, Spain, the Netherlands, and Switzerland, using both traditional econometric techniques and deep learning models. While Granger causality tests, vector autoregression (VAR), impulse response functions (IRF), and forecast error variance decomposition (FEVD) reveal weak or negligible structural links between electricity and GDP, these findings support growing evidence of decoupling in advanced economies. In response, we implement Long Short-Term Memory (LSTM) neural networks to discover non-linear, temporal patterns in electricity data that can improve short-term GDP forecasting. To ensure robustness, we applied a walk-forward validation strategy using a 20-quarter rolling window. While in-sample performance appeared strong, out-of-sample forecasting revealed poor generalizability in most countries, highlighting the limitations of electricity data alone. Our findings underscore the limitations of linear models in decoupled economic systems and emphasize the role of electricity consumption as a potentially useful, but insufficient, predictive input when processed through non-linear machine learning frameworks. These results carry important implications for policymakers, energy analysts, and economists concerned with real-time economic monitoring and sustainability transitions in Europe.