The Unreasonable Difficulty of Time Series Forecasting

Preview

Preview I’ve been thinking recently about what makes time series forecasting problems so difficult compared to other sequence learning tasks or IID Machine Learning problems. First some motivating baselines illustrating the difficulty of the forecasting problem. I ran a series of models statistical (Naive, AutoARIMA, Theta, MSTL, Seasonal-Naive) linear/transformer neural (DLinear, NLinear, PatchTST) a gradient-boosted tree (LightGBM) zero-shot foundation models (Chronos, TimesFM, TTM) “AI” LLM prompted to continue numeric sequence (Claude Opus, Haiku) against the benchmark datasets m4_hourly,

related reading

• Trending Papers - Hugging Facepaperswithcode.com
• Time Series - From Analyzing the Past to Predicting the Future | Towards Data Sciencetowardsdatascience.com
• ML Blog - Everything you need to know about Time Series Forecastingmrtunguyen.github.io
• An introduction to time series forecasting | InfoWorldinfoworld.com
• [2310.07820] Large Language Models Are Zero-Shot Time Series Forecastersarxiv.org
• Granger Causality in Time Series Explained with Chicken and Egg problemanalyticsvidhya.com
• Khoa học dữ liệuphamdinhkhanh.github.io
• Is Capability a Liability? More Capable Language Models Make Worse Forecasts When It Matters Mostarxiv.org
• The Illusion of Knowledgeoaktreecapital.com
• Pitfalls in Evaluating Language Model Forecastersarxiv.org
• Recursive forecasting: Eliciting long-term forecasts from myopic fitness-seekers — AI Alignment Forumalignmentforum.org
• Chapter 2 Time series basics | Time Series with Rs-ai-f.github.io