# The Granularity Paradox: How Temporal Disaggregation Inflates In-Sample Fit and Compounds Out-of-Sample Error

Published: 2026-07-08 · Updated: 2026-07-08 · Category: Research

Source page: https://www.sipply.dev/story/the-granularity-paradox-how-temporal-disaggregation-inflates-d074c711 · 中文: https://www.sipply.dev/zh/story/the-granularity-paradox-how-temporal-disaggregation-inflates-d074c711

## Summary

A recent paper titled 'The Granularity Paradox: How Temporal Disaggregation Inflates In-Sample Fit and Compounds Out-of-Sample Error' explores the complexities of time-series forecasting. The authors reveal that while finer temporal disaggregation can lead to improved in-sample fit, it paradoxically results in poorer out-of-sample accuracy due to the compounding of recursive errors. This finding emphasizes the need for careful consideration of model evaluation metrics, particularly cumulative metrics, when assessing forecasting models across varying levels of granularity.

## Key takeaways

- Finer temporal disaggregation can inflate in-sample fit but degrade out-of-sample performance.
- Recursive error compounding is a significant factor affecting forecasting accuracy.
- Cumulative metrics are essential for evaluating model performance across different granularities.

## Why it matters

Understanding the Granularity Paradox is crucial for improving the reliability of time-series forecasting models, which are widely used across various industries for decision-making.

- **Product managers:** Product managers should be aware of the implications of temporal disaggregation on forecasting accuracy to make informed decisions about product features and development timelines.
- **Developers:** Developers and engineers need to consider the effects of recursive errors in their forecasting models and adopt cumulative metrics to enhance model evaluation and performance.
- **Students and job seekers:** For students and job-seekers, grasping these concepts can provide a competitive edge in fields like data science and analytics, where accurate forecasting is critical.

## Sources

- [The Granularity Paradox: How Temporal Disaggregation Inflates In-Sample Fit and Compounds Out-of-Sample Error](https://arxiv.org/abs/2607.05450) (arXiv cs.LG, 2026-07-08)

## 中文

### 粒度悖论：时间分解如何膨胀样本内拟合并加剧样本外误差

一篇题为《粒度悖论：时间分解如何膨胀样本内拟合并加剧样本外误差》的论文探讨了时间序列预测的复杂性。作者揭示，尽管更细的时间分解可以提高样本内拟合，但由于递归误差的累积，反而导致样本外准确性下降。这一发现强调了在评估不同粒度的预测模型时，特别是累积指标的模型评估指标需要谨慎考虑。

**要点：**

- 更细的时间分解可以膨胀样本内拟合，但降低样本外性能。
- 递归误差的累积是影响预测准确性的一个重要因素。
- 累积指标对于评估不同粒度的模型性能至关重要。

**为什么重要：** 理解粒度悖论对于提高时间序列预测模型的可靠性至关重要，这些模型在各个行业的决策中被广泛使用。

---
Sipply: Sip AI News Simply. https://www.sipply.dev