US Employment: breakevens, benchmarking and blindspots

Introduction

US employment is perhaps the single most consequential economic statistic for markets, but it is also one of the messiest. Policymakers also pay close attention as maximum employment is formally part of the Federal Reserve’s dual mandate. Non-farm payrolls, the household survey and administrative records each tell a slightly different story. All are highly volatile series with employment statistics also subject to frequent, large and persistent revisions. That reality complicates any attempt to extract signals from initial readings, as market movements and policy action are often driven by data that will later be materially changed. This article summarises recent work (Chiu and Wales, 2025) which develops a practical, model-based framework to cut through that noise and deliver timely, internally consistent inference for investment professionals and policymakers.

At its core our approach delivers three things. First, we create a harmonised vintage database that preserves the full revision history across employment series and links those vintages to a broad set of auxiliary indicators. Second, we build a factor-based state-space model that explicitly accounts for revision uncertainty while separating slow structural trends (migration, demographics) from cyclical demand and supply forces. This model structure is well suited to extracting unknown or unobserved trends and cycles. Third, we demonstrate three concrete applications: estimating the employment breakeven, automating the annual benchmarking correction and producing an unofficial employment read when official releases face blindspots for instance during the recent Federal government shutdown. Together these contributions produce a continuously updated, revision-consistent view of the US labour market.

The remainder of this article focuses on the most relevant applications (breakevens, benchmarking and blindspots) and concludes with some practical takeaways for investors. Detailed analysis is available through the link to the full paper at the end of the article, below.

Breakevens

Our framework formalises the employment breakeven as the pace of payroll gains required to hold the unemployment rate steady and implements that concept via a model-consistent conditioning exercise. Concretely, we impose a soft demographic endpoint that guides the trend component toward roughly 30k job gains per month by December 2026, while letting short-run indicators determine the cyclical path. The conditional breakeven therefore isolates the employment growth required to keep unemployment stable given current demographic expectations.

Figure 1 plots the model’s conditional median breakeven (thousands of job gains per month) from January 2025 through December 2026 together with an approximate credible set. Under this conditioning, the breakeven declines steadily over the forecast horizon as population and migration contributions normalise after accelerating in 2023-24. The conditional median begins close to 150k per month at the start of 2025, has recently fallen below 100k per month, and continues to drift lower towards 45k per month by end-2026. This represents a rapid fall. The credible band reflects uncertainty around both the trend endpoint and the near-term cyclical dynamics. Although the central path is materially lower the credible set leaves room for plausible upside or downside around that endpoint.

Figure 1: US Employment breakeven estimates

Sources and Notes: Fulcrum Asset Management LLP and Cheremukhin (2025). Figure shows the latest vintage of one-month changes in non-farm payrolls alongside estimates of the employment breakeven.

Relative to external estimates, such as Cheremukhin (2025), which are lower and, in some cases, even below the latest vintage of payrolls, our conditional breakeven appears more consistent with the gradual rise in the unemployment rate observed this year.

Benchmarking

Annual benchmark revisions, are a perennial source of measurement error. The benchmarking process is intended to reconcile non farm payrolls, which are a survey-based sample estimate, with more comprehensive administrative data drawn from unemployment insurance records that account for roughly 97% of all US employment. In this way, estimates and assumptions used in the aggregation of the initial payroll survey are resolved with a long lag. Our work shows how a revision-aware factor model can improve upon canonical early-benchmark approaches (Berger and Phillips, 1993, 1994) by incorporating vintage dynamics, auxiliary indicators and a time-varying wedge. This results in a benchmark adjustment which better aligns to historical revision patterns.

This year, the preliminary March 2025 adjustment (announced in September) currently implies a large downward adjustment to the level of employment. Figure 2 illustrates this, where the majority of the monthly differences arise in March 2025. A mechanical early-benchmark method would imply an even larger reduction in seasonally adjusted payrolls. In contrast, our framework estimates a substantially smaller final revision, of roughly half the magnitude of the simple early-benchmark projection. Empirically, about half of the initial gap is expected to persist in the final data.

Figure 2: US Employment benchmarking estimates

Sources and Notes: Fulcrum Asset Management LLP, Haver Analytics, BLS. The chart shows the latest vintages of the Berger and Phillips (1993) early benchmark estimates for the level of non-farm payrolls. The early benchmark model is represented in teal, while the median estimate from the full model is depicted in red.

Blindspots

Official releases can be interrupted. Federal government shutdowns, collection delays, and other disruptions can create “blindspots” across the universe of employment indicators. Our framework mitigates this by continually estimating the underlying employment cycle within a state-space system that naturally accommodates missing observations. In practice, this allows the model to draw on private indicators, such as Automatic Data Processing (ADP) payrolls, state-level initial claims, and other non-federal series, to produce a timely, internally consistent employment update even when key official data are unavailable.

Two exercises illustrate this. First, during the recent federal government shutdown, when official payroll prints were delayed, our model continued to generate implied employment numbers. The current underlying estimates are shown in Table 1. These point to payroll growth slowing toward roughly 40k per month by the end of the year. Because this is below our conditional breakeven estimates, discussed earlier, the corresponding unemployment rate rises to around 4.5% over the same period. These values represent our projection in the absence of the October official report.

Second, the model also projects the revision path of successive non-farm payroll vintages. Initial prints typically understate the benchmark-adjusted final outcome, so first-release numbers for these months are likely to come in below the underlying medians in Table 1. In other words, the initial official data, once released, may appear softer than the model-implied underlying path, contributing to near-term downside surprises.

Table 1: Unofficial US Employment report

Sources and Notes: Fulcrum Asset Management LLP. Table shows latest model projections in absence of the US employment report for October. Black numbers denote latest official data. Red numbers denote model median estimates. For Current Employment Statistics (CES) non-farm payrolls underlying differs from data due to measurement error and estimated impact of future revisions. For Current Population Statistics (CPS) unemployment and underemployment rates differences are attributed to measurement error away from underlying employment cycle.

Conclusion

This article presents three exercises from our recent work on understanding US employment dynamics. By explicitly modelling revision processes and integrating demand, supply, and structural indicators, our framework provides a coherent real-time signal of underlying labour-market conditions. This helps avoid over-interpreting any single initial release as a definitive read on longer-term forces.

These applications suggest a consistent story. US payroll growth has moderated as demographic headwinds and the normalisation of net migration have pulled the structural breakeven lower through this year. In the near term, our model anticipates further softening, with the unemployment rate drifting toward 4.5% and average monthly payroll gains settling in the low tens of thousands. Annual benchmark revisions are expected to lower the level of payrolls, but by substantially less than implied by the most mechanical early-benchmark calculations.