A new paper introduces the "Granularity Paradox" in time-series forecasting, highlighting how increasing temporal disaggregation improves in-sample fit but degrades out-of-sample accuracy due to compounded errors. The research formalizes this trade-off and benchmarks ten models across six granularities using a 13-year public procurement dataset. Findings indicate that while some models like Holt-Winters perform poorly at daily frequencies, LSTMs show a U-shaped error curve, and Linear Regression remains stable, suggesting the paradox is linked to recursive feedback topology rather than model complexity. AI
IMPACT Highlights potential pitfalls in using standard metrics for evaluating complex AI models in time-series forecasting.
RANK_REASON Academic paper detailing a new paradox in time-series forecasting. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
