A new study published on arXiv investigates the impact of different calendar feature encoding methods on the accuracy of time series foundation models (TSFMs). Researchers found that the chosen encoding strategy significantly affects forecast accuracy, with differences up to 83% on holidays for certain model-encoding combinations. The study systematically compared seven calendar encodings across four state-of-the-art TSFMs using nine diverse datasets, highlighting that improvements are most pronounced around holidays where calendar information offers unique signals. The findings suggest that current benchmarks may distort model comparisons due to fixed encoding strategies. AI
IMPACT Highlights the importance of feature engineering for time series models, potentially improving forecasting accuracy in various domains.
RANK_REASON Academic paper published on arXiv detailing research findings. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
