Is Your Time Series Even Forecastable?

2026-03-03 · Source: Valeriy’s Substack · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Advanced, medium

Summary

This article introduces the concept of forecastability, a data property defining the theoretical ceiling for prediction accuracy in time series, independent of the model used. It highlights that many time series are inherently unforecastable, leading to wasted effort on complex models. The piece proposes a two-step diagnostic process: first, use the variance ratio test (Lo and MacKinlay) to identify random walks, mean-reverting, or trending series. Second, employ entropy measures like Sample Entropy and Permutation Entropy to detect nonlinear structures and quantify the degree of predictability. For instance, developed stock markets exhibit near-maximal permutation entropy, indicating low forecastability, while emerging markets show more structure. The article emphasizes that understanding a series' forecastability before modeling can save significant computational and engineering resources.

Key takeaway

For data scientists and machine learning engineers evaluating new time series projects, you should prioritize assessing forecastability before model selection. By computing the variance ratio and normalized permutation entropy, you can quickly determine if a series contains exploitable structure. This approach prevents investing GPU hours and engineering effort into complex models for inherently unforecastable data, allowing you to allocate resources more effectively to series with a high predictability ceiling.

Key insights

Forecastability is a data property, not a model property, determining a time series' theoretical prediction ceiling.

Principles

• Forecastability is information-theoretic.
• Variance ratio identifies linear structure.
• Entropy measures capture nonlinear dependencies.

Method

Compute variance ratio; if not a random walk, then calculate normalized permutation entropy. High entropy means low forecastability.

In practice

• Use variance ratio as first diagnostic.
• Apply permutation entropy for nonlinear structure.
• Avoid complex models for near-random series.

Topics

• Time Series Forecastability
• Variance Ratio Test
• Entropy Measures
• Sample Entropy
• Permutation Entropy

Best for: AI Scientist, Machine Learning Engineer, Data Scientist

Related on AIssential

• 7 Entropies, 7 Different Answers — Which One Tells You Your Series Is Forecastable? — Entropy is a family of measures, each sensitive to different data structures for assessing forecastability.
• FAME: Forecastability-Aware Mixture of Experts for Heterogeneous Time Series Forecasting — FAME uses forecastability fingerprints and a sparse router to dynamically select optimal experts for heterogeneous time series.
• How Far Can You Actually Forecast? The Lyapunov Ceiling — Forecast quality in time series models decays predictably with increasing prediction horizon.
• Time Series Forecasting Analysis with Python — This article presents a practical workflow for time series forecasting in finance, emphasizing that successful implementation requires more than just model selection, focusing on data cleaning, robust…
• Autoregressive Models: Predicting the Future Using the Past — Autoregressive models predict future values by leveraging patterns observed in their own historical data.
• Navigating the Safety-Fidelity Trade-off: Massive-Variate Time Series Forecasting for Power Systems via Probabilistic Scenarios — Large-scale power system forecasting requires new benchmarks and models to balance safety and fidelity.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Valeriy’s Substack.