Feature Selection for Rare Earth Element Price Forecasting - A Hybrid Econometric and Machine Learning Approach
(2025) DABN01 20251Department of Economics
Department of Statistics
- Abstract
- This thesis addresses the need for robust price forecasting in the volatile Rare
Earth Elements (REEs) markets. With a specific focus on neodymium, given it is
the most utilized REEs, we investigate how feature selection can be effectively performed
to guide REE price forecasting by proposing and evaluating three distinct
methodologies: Adaptive Lasso, Post-Double Selection (PDS), and an integrated
approach combining PDS with Granger Causality testing. After applying these
feature selection methods, a Random Forest is then employed in a rolling window
forecasting strategy with dynamic selection. Our empirical analysis reveals that no
single feature selection method consistently optimizes forecasting accuracy across all
horizons.... (More) - This thesis addresses the need for robust price forecasting in the volatile Rare
Earth Elements (REEs) markets. With a specific focus on neodymium, given it is
the most utilized REEs, we investigate how feature selection can be effectively performed
to guide REE price forecasting by proposing and evaluating three distinct
methodologies: Adaptive Lasso, Post-Double Selection (PDS), and an integrated
approach combining PDS with Granger Causality testing. After applying these
feature selection methods, a Random Forest is then employed in a rolling window
forecasting strategy with dynamic selection. Our empirical analysis reveals that no
single feature selection method consistently optimizes forecasting accuracy across all
horizons. While no feature selection method demonstrates superiority in every forecasting
horizon, the PDS with Granger Causality approach proves most effective for
short-term forecasting, suggesting that statistical inference can enhance short-term
accuracy. Conversely, the PDS method generally underperforms due to its tendency
to retain an overly dense set of predictors. The study highlights the inherent
challenges in forecasting REE prices, particularly due to frequent periods of high
volatility. This research contributes valuable insights into developing more accurate
and reliable REE price prediction tools, emphasizing the importance of tailoring
feature selection strategies to specific forecasting horizons. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9202593
- author
- Evaldsson, Jonathan LU and Sigfússon, Jóhann Ari LU
- supervisor
- organization
- course
- DABN01 20251
- year
- 2025
- type
- H1 - Master's Degree (One Year)
- subject
- keywords
- Rare Earth Elements, Price Forecasting, Feature Selection, Adaptive Lasso, Granger Causality
- language
- English
- id
- 9202593
- date added to LUP
- 2025-09-12 09:04:03
- date last changed
- 2025-09-12 09:04:03
@misc{9202593, abstract = {{This thesis addresses the need for robust price forecasting in the volatile Rare Earth Elements (REEs) markets. With a specific focus on neodymium, given it is the most utilized REEs, we investigate how feature selection can be effectively performed to guide REE price forecasting by proposing and evaluating three distinct methodologies: Adaptive Lasso, Post-Double Selection (PDS), and an integrated approach combining PDS with Granger Causality testing. After applying these feature selection methods, a Random Forest is then employed in a rolling window forecasting strategy with dynamic selection. Our empirical analysis reveals that no single feature selection method consistently optimizes forecasting accuracy across all horizons. While no feature selection method demonstrates superiority in every forecasting horizon, the PDS with Granger Causality approach proves most effective for short-term forecasting, suggesting that statistical inference can enhance short-term accuracy. Conversely, the PDS method generally underperforms due to its tendency to retain an overly dense set of predictors. The study highlights the inherent challenges in forecasting REE prices, particularly due to frequent periods of high volatility. This research contributes valuable insights into developing more accurate and reliable REE price prediction tools, emphasizing the importance of tailoring feature selection strategies to specific forecasting horizons.}}, author = {{Evaldsson, Jonathan and Sigfússon, Jóhann Ari}}, language = {{eng}}, note = {{Student Paper}}, title = {{Feature Selection for Rare Earth Element Price Forecasting - A Hybrid Econometric and Machine Learning Approach}}, year = {{2025}}, }