Beyond Linear Models: Evaluating Tree-Based, Instance-Based, and Deep Learning Methods for Carbon Market Forecasting
線形モデルを超えて：カーボン市場予測のための木ベース、インスタンスベース、深層学習手法の評価 (AI 翻訳)
Nadirgil O
🤖 gxceed AI 要約
日本語
本稿は、28の多様な炭素価格ドライバーを対象に、深層学習、インスタンスベース、木ベースなど多様なモデルを同一データセット上で比較する初の網羅的ベンチマーク研究である。結果、石炭価格が最大のドライバーであり、英国10年債利回り、S&P500、米国イールドカーブスプレッド、日経平均が続く。予測性能ではETとKNNが最上位層、DNN・RF・SHAP-XGB・XGBが第二層、LSTMとSVMが最下位となった。
English
This paper presents a comprehensive multi-model benchmarking study for carbon price forecasting, using 28 diverse drivers and comparing deep learning, instance-based, and tree-based models on the same dataset. Coal price is identified as the dominant driver, followed by UK 10-year bond yield, S&P 500, US yield curve spread, and NIKKEI 225. ET and KNN form a statistically superior top tier, with DNN, RF, SHAP-XGB, and XGB in the second tier, while LSTM and SVM rank last.
Unofficial AI-generated summary based on the public title and abstract. Not an official translation.
📝 gxceed 編集解説 — Why this matters
日本のGX文脈において
日本では炭素価格の変動要因として海外市場の影響が示唆され、GXリスク管理や投資判断に示唆を与える。また、日経平均がドライバーとして挙げられたことは、日本市場の連動性を示す。
In the global GX context
This study advances carbon market forecasting by benchmarking multiple ML models on a global set of drivers, highlighting the importance of non-EU financial markets. It provides robust methodology for practitioners and regulators to assess price drivers and model performance.
👥 読者別の含意
🔬研究者:Provides a rigorous benchmark framework for carbon price forecasting, including feature selection and model comparison methodologies.
🏢実務担当者:Offers actionable insights on key carbon price drivers (e.g., coal, equity indices) and identifies best-performing models (ET, KNN) for trading or hedging strategies.
🏛政策担当者:Highlights the influence of global capital flows and non-EU markets on carbon pricing, relevant for designing efficient carbon markets.
📄 Abstract（原文）
<title>Abstract</title> <p>This study introduces a comprehensive and systematic multi-model benchmarking perspective on carbon pricing by jointly employing extensive carbon price feature selection and forecasting research through exploring 28 categorically and geographically diversified carbon price drivers across a large portfolio of Deep Learning (DL), instance and tree based models on the same dataset and time period. Methodology is equipped with rigorous statistical analysis, including Jarque-Bera (JB) test, Augmented Dickey-Fuller (ADF) unit root test, Pearson correlation and Variance Inflation Factor (VIF) multicollinearity screening, and ARCH-LM heteroskedasticity testing with GARCH(1,1) filtering. All models are optimised via exhaustive grid search with time-series cross-validation, and predictive accuracy is assessed using eight complementary metrics, in addition cross pair performance differentials are verified by the HLN-corrected Diebold-Mariano and the Nemenyi Post-Hoc rank tests. Results identify coal price as the dominant carbon price driver with the UK 10-year bond yield, S&P 500, US yield curve spread (US SPR), and NIKKEI 225 emerging as secondary determinants. Notably, non-EU equity and bond markets consistently outrank their European counterparts, reflecting the influence of global capital flows on carbon pricing. In terms of forecasting performance, ET and KNN constitute a statistically superior first tier, followed by a second tier comprising DNN, RF, SHAP-XGB, and XGB, while LSTM and SVM rank last.</p>
🔗 Provenance — このレコードを発見したソース
- Research Square https://doi.org/10.21203/rs.3.rs-9172688/v2first seen 2026-06-05 04:40:09
🔔 こうした論文の新着を逃したくない方は キーワードアラート に登録（無料・3キーワードまで）。
gxceed は公開メタデータに基づく研究支援データセットです。要約・翻訳・解説は AI 支援で生成されています。 最終的な解釈・検証は利用者が原典資料に基づいて行うことを前提とします。