Model accuracy · Lead · Backtest

Lead price forecast accuracy

How well would MetalAlert's lead probabilities have done? Every dot below is a past month that the model had never seen when it made its call. Change the question and the probability threshold to see the result.

How MetalAlert evaluates its accuracy

Tested on months it has never seen.

  1. 01

    Split the history

    Training, backtest and holdout, so no month is scored by a model that has seen it.

  2. 02

    Expanding-window backtest

    Every month is predicted by a model that has only seen earlier months. The latest years are scored once.

  3. 03

    You set the bar

    The probability threshold decides how many calls you get and how often they are right. See the methodology. The video uses copper as the example.

A backtest shows how the model would have done on past months. It is not a guarantee of future results, and MetalAlert provides probability estimates for informational and decision-support purposes only.

Related

Accuracy for other metals: Copper · Aluminum · Nickel · Zinc · Tin

See the odds behind your buying decisions

MetalAlert estimates the probability of a 2%, 5% or 10% move in each LME base metal over one and three months, with alerts at the level you choose. Book a short walkthrough and we will show it against your own purchases.

Book a walkthrough Get early access
← Model accuracy MetalAlert home