Forecasts That Answer for Themselves: A Sealed, Prospective Test of Enrollment Forecasting for Higher Education Planning in Chile
Year: 2026
Type: preprint
Abstract: Stage 1 protocol, written in the Registered Report format; the results will be reported in a separate article after the 2027 enrollment data are released (July 2027). Institutions and governments plan admissions capacity and budgets with predictive analytics whose validity is rarely tested before the outcome is known. This Registered Report tests it with first-year enrollment in Chile, prospectively replicating a study that applied zero-shot time-series foundation models to one university. On October 29, 2026, we sealed 2027 forecasts for 7,424 undergraduate programs and 135 aggregate series, with 50%, 80%, and 95% intervals calibrated by conformal prediction. In the retrospective evaluation (2015-2025), every model reached nominal coverage on average and no foundation model beat the naive forecast in any year. In July 2027 we will test that coverage, against bands that allow for dependence across programs, and the absence of an advantage for foundation models. The results will show whether a simple forecast with calibrated intervals is enough, and for which programs, institution types, and regions it fails. Files: the Stage 1 manuscript and two supplementary files (the detailed disclosure of AI use and an English translation of the preregistration protocol). The sealed forecasts are in Zenodo, https://doi.org/10.5281/zenodo.23768424.
Author: Velásquez Araya, Alberto
Institution: Universidad Católica del Norte
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7225890962
