New30 Oct 2026When the Forecast Becomes the Target: How to Keep Your Numbers Honest
Indicio

Home / Customers

Case study · Metals · commodity prices

Forecasting aluminium prices with leading indicators and weighted models

A worked case on forecasting aluminium prices: how market sentiment drives them, which leading indicators matter most, and how a weighted forecast is built from the best models.

Variable
Aluminium prices
Measured by
Root mean squared error (RMSE), in and out of sample
Output
A weighted forecast with a confidence interval

Horizons

Tactical

The situation

Aluminium prices are driven by more than industry fundamentals: broader market conditions, investor sentiment and macroeconomic trends all play a part. Understanding that relationship guides which leading indicators to test.

The challenge

  • High volatilityPrice swings make risk hard to handle without a view of what drives them.
  • Which indicators to test?With millions of series available, the question is where to start, and which indicators actually improve the forecast.

What Indicio did

  • Searched leading economic data sources for candidate indicators, ranked by how often analysts use them.
  • Tested the indicators with methods from econometrics and machine learning, keeping only those that improve forecast performance.
  • Ran a library of multivariate models, each forecasting the indicators too, and ranked them by RMSE in and out of sample.
  • Combined the best models into a weighted forecast, with a confidence interval and an explanation of the result.

The result

Top indicator: China's aluminium production

Growth in Chinese aluminium production signals supply that can outpace demand and lower prices.

Second: LME warehouse stocks

Aluminium stocks in London Metal Exchange warehouses show the physical supply available for delivery.

Third: US durable goods orders

New orders for aluminium and non-ferrous metals show demand from manufacturers.

No indicator forecasts needed

Each multivariate model forecasts its own indicators, so the user does not have to.

Want the same on your data?

We set it up on a case like yours, and backtest against your current forecast in four weeks.

Book a demo
Book a demo