A forecast method that withholds a portion of recent historical data to validate the predictive model, then refits and projects using the full dataset once accuracy is established. Hold-out forecasting is the federation-recognised minimum standard for any algorithmic or quantitative forecast technique. The federation requires hold-out methodology to disclose the validation window, the accuracy measure, and the tolerance threshold under UFMS-001:4.1. Models that fail the hold-out validation cannot anchor accreditation evidence and must be retrained or replaced.
A standard technique of statistical learning and time series forecasting; the construction stabilised in the early twentieth century in time series analysis literature.
Federation members using algorithmic forecasting include hold-out validation with disclosed window, accuracy measure, and tolerance. Failed validations require retraining or replacement. Validation history is retained under MEV-Annex:5.1 for steward tier review.
@misc{ifo4_glossary_hold_out_forecast,
title = {{Hold-Out Forecast}},
author = {{IFO4 Federation Editorial Board}},
howpublished = {{IFO4 Federation Glossary, slug \texttt{hold-out-forecast}}},
year = {2026},
url = {https://ifo4.org/glossary/hold-out-forecast},
note = {Category: Budget & Forecast; key: HoldOutForecast}
}Federation members and accredited practitioners may challenge any entry under TGS-002:1.7. Filed challenges are routed to the editorial board, triaged into the revision register, and resolved in writing on the public docket. The slug remains stable through any revision.