Model Selection and Evaluation

What does each modeling approach require, and when is it best?

Each of the five approaches has its own data requirements, and selection follows from what the market actually offers for a given commodity and location:

Approach	Requires	Best when
Trade-based	Official trade statistics that are timely and suitable for the targeted specification	Suitable trade data exists — the preferred approach wherever it does
Formula-based	A stable relationship with related commodity prices and economic variables	The price moves predictably with measurable drivers
Freight-based	An existing assessment at another location, plus freight and insurance rates	A value is needed at a location where another assessment already anchors the market
Manufacturing-cost-based	Known production economics — input costs, utilities, labor, overhead	Market prices are scarce but the cost of producing the commodity is known
Compiled	Multiple public-source price series that can be validated	Several public series exist and a validated monthly average represents the market

Selection is made per commodity and location, weighing data availability, timeliness, and suitability for the targeted specification. The trade-offs are conceptual, not scored: there is no published ranking or error rate per approach.

How are models evaluated and updated?

Model evaluation runs continuously alongside modeling itself, with a formal review of all methodologies at least annually. During modeling, the accuracy of the models — including forecasts — is evaluated, and computational algorithms and regression models are adjusted when necessary. When a series shows atypical behavior or an inconsistency is detected, the associated data is flagged for further verification; this can trigger a review and recalibration of the model, or its replacement when it no longer reflects the country's conditions. Analysts also investigate the cause of each inconsistency — from erroneous data entry to changes the model has not yet accounted for — and the insights from these investigations are fed back into the models to improve accuracy and reliability over time.

Normalization — adjusting gathered market information to the assessment's base standard — occurs concomitantly and iteratively with this modeling work; it is detailed on the Comparability and Normalization page.

How do methodology changes reach subscribers?

Methodology changes resulting from a review are announced to customers in advance, before they take effect, together with the duration of the transition period. New series and assessments undergo shadow testing for a period before official release. All changes — methodology modifications and data revisions alike — are then documented each month in the publicly available Intratec Release Notes, keeping the path from review to published change traceable. The revision lifecycle itself is covered in the Publishing and Revisions section.