DOI: https://doi.org/10.71734/DP-2026-1

Official statistics increasingly make use of higher-frequency time series. But when users ultimately are interested in a seasonally adjusted temporal aggregate of these data, we have to decide whether to perform seasonal adjustment or aggregation first. We examine the consequences of this decision based on simulated and real-world time series using a battery of diagnostics. We synthesise our findings into practical guidelines that help users choose the aggregation level that balances statistical quality and real-time usefulness.

How should we adjust for seasonality in higher-frequency data?

With the increasing availability of daily and weekly time series, economists and statisticians face a fundamental question: should seasonal adjustment be performed at the original high frequency or after aggregating the data to a lower frequency? This study provides a comprehensive framework to guide practitioners in selecting the optimal temporal resolution for seasonal adjustment, balancing statistical quality and practical considerations.

Methodology: A Comprehensive Diagnostic Framework

We employ a combination of simulated and real-world time series to evaluate the performance of seasonal adjustment at different temporal resolutions. We use leading seasonal adjustment methods, including DSA2 (Daily Seasonal Adjustment), WSA (Weekly Seasonal Adjustment), X-13, and TRAMO-SEATS, to adjust data at daily, weekly, monthly, and quarterly levels. The study applies a battery of diagnostic tools, such as tests for residual seasonality, calendar effects, and revision size, to assess the quality of the adjusted series. Additionally, we examine the relationship between higher-frequency proxies and their lower-frequency target variables to evaluate the practical utility of different temporal resolutions.

Theoretical Insights and Core Arguments

The paper builds on the premise that the properties of seasonal patterns and calendar effects vary with the temporal resolution of the data. For instance, daily data can capture intricate patterns like day-of-the-week effects, while monthly or quarterly data may obscure such details. However, higher-frequency data are often noisier and more prone to outliers, complicating the adjustment process. We argue that the choice of temporal resolution involves a trade-off between preserving informational richness and ensuring statistical robustness. We also highlight the limitations of existing seasonal adjustment methods at different frequencies, such as the difficulty of identifying calendar effects in quarterly data.

Key Findings: Practical Guidelines for Temporal Resolution

We propose a decision tree to guide the choice of temporal aggregation for seasonal adjustment. The process begins by defining the use case and eliminating unsuitable aggregation levels. For instance, if the goal is to use the series as a higher-frequency proxy for a quarterly target, quarterly aggregation is only relevant if the target series is significantly delayed. If the primary objective is to proxy a target, we recommend selecting the aggregation level that maximises alignment with the target, using out-of-sample prediction accuracy metrics when applicable.

If no target series exists or if other objectives are also important, we suggest considering the series' properties. We advise avoiding quarterly aggregation when calendar effects are critical, as identifying these effects at that frequency is often unreliable. For additional effects like cross-seasonal patterns, daily or higher frequencies are preferable, as they allow for more reliable estimation. Finally, we recommend evaluating the remaining temporal resolutions using diagnostics and metrics, summarising them via mean ranks, and selecting the option with the lowest rank.

Ollech, D., M. Stefan (2026), Diagnostic tools for selecting the temporal resolution for seasonal adjustment, Bundesbank Discussion Paper, No 01/2026