top of page
Image by SJ Objio
Title: Enhanced Wavelet-ARIMA Method for Predicting Metal Prices
Organization

Vale - Cranfield University

Period

July 2022 - July 2024

Status

Completed

Est. Budget

USD 0.7M

Est. Business Benefit

USD 5.5M

Indicative Value-to-cost

8.5x

Keywords

Wavelet analysis, ARIMA, metal-price forecasting, multiresolution analysis, time series, commodities

Benefit basis

Estimated value from improved price forecast accuracy, better commercial decision timing, improved planning assumptions, and reduced forecast error in commodity-price-sensitive decisions.

Business Challenges 
Metal-price prediction supports exploration evaluation, mine planning, sales, procurement, hedging, investment prioritisation, and day-to-day commercial decisions. Metal prices exhibit nonstationarity, cyclical behaviour, structural breaks, and noise. Traditional ARIMA models can underperform when cyclical patterns occur at multiple time scales. The challenge was to improve forecast accuracy by combining time-domain and frequency-domain analysis.
Expanded Technical Solution
The project developed an enhanced Wavelet-ARIMA forecasting method for monthly prices of iron, aluminium, copper, lead, and zinc. Wavelet-based multiresolution analysis decomposed each price series into components operating at different frequencies. ARIMA models were then fitted to decomposed components, improving their ability to forecast cyclicality, trend, and short-term fluctuations. The project also optimised the wavelet transform type, wavelet function, and number of decomposition levels before model fitting.
Technical Work Packages/Methods
  • Collected historical monthly metal-price series and prepared consistent time-series datasets for multiple commodities.
  • Applied stationarity tests, decomposition diagnostics, and transformation logic before model fitting.
  • Used wavelet multiresolution analysis to separate noise, cyclical components, and trend-like structures.
  • Fitted ARIMA models to decomposed components and reconstructed the forecast signal.
  • Used walk-forward validation and error metrics such as MAE, RMSE, MAPE, and directional accuracy for model evaluation.
Key Deliverables
  • Wavelet-ARIMA forecasting methodology for multiple metal products.
  • Model-selection procedure for wavelet type, function, and decomposition level.
  • Validated forecast benchmark against standard ARIMA and other baseline models.
  • Technical report explaining forecast interpretation and commercial use cases.
Results and Value Created
The completed project showed that ARIMA performance can be materially improved by applying wavelet-based multiresolution analysis before model fitting. It provided a more technically robust approach to forecasting metal-price series characterised by multiple cycles and noisy dynamics. The estimated benefit reflects improved commercial planning, price-risk awareness, timing of sales/purchases, and reduced error in economic evaluation assumptions.
bottom of page