Answers are grounded in this paper and cite their sources.
Retrospective time-series analysis reveals that three day-type profiles outperform seven weekday profiles across 44 U.S. balancing authorities, highlighting the value of recency and temperature.
Methods that work on electricity load series rely on a notion of the typical daily profile, and practice differs on how many to keep: one per weekday, or one per day type (workday, Saturday, Sunday) with a list of special days. This note compares the alternatives on the public EIA-930 record of 44 U.S. balancing authorities (2015-07-01 to 2026-06-30, 172,741 BA-days), with profiles built causally from earlier days. Seven weekday profiles have a higher out-of-sample shape error than three day-type profiles in every BA; the length and recency of the reference history matter more than the calendar; fixed seasonal classes make things worse while the previous year's analog days help; and a three-class day type from daily mean temperature at one NOAA station per BA (heating below 59 F, cooling above 72 F) reduces the error by 14.6% with the previous day's temperature almost as good, so no forecast is required. Thanksgiving, Christmas and the two eves resemble Saturdays; Memorial Day, Labor Day and Independence Day resemble Sundays; four minor federal holidays are not distinguishable from ordinary weekdays. Code, results and figures are open. Web version and plain-language version: github.com/cardinalgrid/notes; cardinalgrid.com/notes.
No takes yet. Share an insight, caveat, or question.
Ricardo Wagner Correia Guerra Filho (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: