---
title: "Nowcasting with Mixed Frequency Principal Components When Not All Months Are the Same"
authors: "Patrick Higgins"
source: "Federal Reserve Bank of Atlanta (Working Paper 2026-15)"
source_url: "https://www.atlantafed.org/research-and-data/publications/working-papers/2026/09/25/15-nowcasting-with-mixed-frequency-principal-components-when-not-all-months-are-the-same"
paper_id: "Atlanta Fed WP 2026-15"
harvested: 2026-09-28
year: 2026
keywords: [nowcasting, principal components, factor models, mixed frequency, temporal aggregation, GDPNow, missing data, signal extraction]
pdf_url: "https://www.atlantafed.org/-/media/Project/Atlanta/FRBA/Documents/research-and-data/publications/working-papers/2026/09/25/15-nowcasting-with-mixed-frequency-principal-components-when-not-all-months-are-the-same.pdf"
---
## Abstract

This paper develops algorithms for estimating a principal components (PC) model for monthly macroeconomic data mixed with either weekly or daily data. The algorithms' utilization of high frequency data allows it to account for differences in the temporal aggregation patterns among the monthly series that are simple aggregates or averages and those that have an alternative aggregation pattern such as data from the monthly employment report where the aggregation pattern is often series- and month-specific and depends on the timing of the twelfth day of the month. The algorithms are applied to three respective models with mixed frequency price related, labor market related, or economic activity related data, the latter of which is similar to the factor model used by the Atlanta Fed's GDPNow nowcasting model to forecast GDP source data. When aggregated appropriately, the mixed frequency activity model PC closely resembles both the Chicago Fed National Activity Index and GDPNow's dynamic factor. Moreover, the PCs from all three models can be well approximated using the subset of data that would remain available during a widespread government shutdown.

JEL classification: E37, C38, C53. Key words: nowcasting, forecasting, macroeconometric forecasting, principal components, factor models. DOI: 10.29338/wp2026-15

## Why it's in the StratMac library

- **This is the SM pipeline's own problem, solved in a Fed working paper.** The SM signal set mixes monthly, weekly and daily FRED series, and the project's canonical method is dimensionless-ratio construction followed by ARIMA(1,1,0) residual extraction. Higgins formalises what to do when the aggregation pattern of the *monthly* series is itself series- and month-specific (the payroll-report twelfth-day rule is the clean example) rather than a plain monthly average.
- **A constructive alternative to filling gaps.** The paper's headline robustness result is that the activity-model PC can be well approximated from the subset of source data that survives a widespread government shutdown — i.e. a principled treatment of missing releases instead of interpolation by assumption.
- **External validation anchor for an activity factor.** The mixed-frequency activity PC closely tracks both the Chicago Fed National Activity Index and GDPNow's dynamic factor, which gives SM a published benchmark to check its own activity composite against.
- **Method reference for the `forward` section**, alongside the existing ARIMA-vs-ML and macro-signal-detection references.