U.S. Total Nonfarm Payroll Employment

USPAYROLL is the seasonally adjusted US total nonfarm payroll employment level, recorded in thousands of persons exactly as published and without transformation. The series counts employment carried on the payrolls of nonfarm establishments, so its population and counting unit differ from the household-survey concept of employed persons, which assigns each individual a single activity status (Hussmanns, Mehran, and Verma 1990), and the two counts can move apart.

Because the unit counted is a payroll job rather than a person, someone paid by two employers is counted twice, while the self-employed and unpaid family workers fall outside the count. A change in the level is therefore the net outcome of gross flows of hiring, separation, and the opening and closing of establishments (Clark and Summers 1979).

As a count of aggregate real activity, payroll employment belongs to the family of monthly series roughly coincident with the general state of business and sits at the center of reference-cycle dating (Mitchell and Burns 1938; Burns and Mitchell 1946). Such comoving aggregates are recast as indicators of a single unobserved state of the economy (Stock and Watson 1989; Stock and Watson 1991), the count takes a codified role in diffusion and composite cyclical-indicator systems and is read in real time as a current-conditions signal (Moore 1961; Shiskin 1961; Zarnowitz 1992), and the turning points of specific cycles are identified by a reproducible algorithm (Bry and Boschan 1971).

A headcount is not an output measure, so employment and production readings can diverge where productivity and hours absorb a shock, which distinguishes this count from the deflation-and-quantity-relative tradition of industrial output measurement (Fabricant 1940). Such an atheoretical count acquires economic meaning only against a model of the labor market (Koopmans 1947).

KRED stores the published value unchanged and computes nothing from it, applying no rescaling, deflating, smoothing, or annualizing. Every operation that produces the number therefore occurs in the measurement system that generates the series rather than in KRED.

That measurement rests on a survey of establishments that tabulates employment paid for the reference pay period, and its counting unit and status rules are defined in contrast with the person-based operational definitions of labor-force statistics (Hussmanns, Mehran, and Verma 1990). The net effect of establishment births and deaths not yet captured in the sample is estimated by a model and later replaced when administrative records settle the level, so the series must be read as subject to revision. The survey-design logic that carries reporting units into a quantitative aggregate stands in the tradition that first treated activity questions scientifically (Anderson 1952).

Raw monthly headcounts move with calendar and seasonal regularities, so the published level is already seasonally adjusted at source, and the series KRED carries is consequently untransformed by KRED and adjusted upstream at once, both true without contradiction. The adjustment removes the recurring within-year pattern in the manner of the moving-average decomposition operationalized for monthly indicators, leaving a series read for cyclical rather than seasonal movement (Shiskin 1961).

The accounting that decomposes a change in the level into flows separates hiring, separation, and duration (Clark and Summers 1979), and interpreted as a cyclical quantity the count inherits the indicator-selection logic, the reference-cycle measurement, the algorithmic turning-point dating, and the diffusion and composite construction of the cyclical-indicator tradition (Mitchell and Burns 1938; Burns and Mitchell 1946; Bry and Boschan 1971; Moore 1961). The employment level is read as a dynamic factor that proxies a latent activity state (Stock and Watson 1989; Stock and Watson 1991), the historical usefulness of that signal has been evaluated (Zarnowitz 1992), and its interpretation carries the qualification that it presupposes an explicit model (Koopmans 1947).

As a coincident measure of real activity, total nonfarm payroll employment is among the series whose clustered turning points define expansions and contractions in the reference-cycle scheme, and it was selected precisely for that rough coincidence with the cycle (Mitchell and Burns 1938; Burns and Mitchell 1946). Such a count enters the single-index reading of the economy's current state, the empirical basis for treating the monthly increment as a signal of the cyclical position (Stock and Watson 1989; Stock and Watson 1991).

The series sits inside diffusion and composite systems that identify turning points in near-real time (Moore 1961; Shiskin 1961), the historical reliability of that signal has been evaluated (Zarnowitz 1992), and specific cycles are dated by algorithm (Bry and Boschan 1971).

The labor-market interpretation reads the employment level through gross entry and exit flows rather than as a static stock (Clark and Summers 1979), and the definitional gap between a payroll basis and a household basis fixes what an increment represents (Hussmanns, Mehran, and Verma 1990). A single month's increment can be small relative to the eventual revision, so a moving average over several months summarizes the hiring trend more stably.

Because employment drives aggregate labor income, the count is treated within the national-income measurement frame as the real basis on which income arises (Kuznets 1941), and it conditions the spending and confidence channels of household behavior (Katona 1951). Labor and confidence readings carry independent predictive content for consumption (Carroll, Fuhrer, and Wilcox 1994; Ludvigson 2004). A coincident count nonetheless carries cyclical information only against a structural model of the labor market (Koopmans 1947).

Markets read the monthly increment in total nonfarm payroll employment as a high-frequency gauge of the cyclical state that conditions monetary policy and the path of short rates, resting on its established coincident-indicator role and its formalization as a latent-state proxy (Mitchell and Burns 1938; Burns and Mitchell 1946; Stock and Watson 1989; Stock and Watson 1991).

Because employment drives aggregate labor income, the count anchors the consumption channel that underlies household spending behavior (Katona 1951), and its forecasting content has been assessed through the related consumer-sentiment series (Carroll, Fuhrer, and Wilcox 1994; Ludvigson 2004). Since expectations for the US short-rate path set the benchmark for global rates and exchange rates, a surprise in this print transmits into the discount rates applied to Korean assets.

Composite and diffusion frameworks embed the count in the real-time dashboards on which positioning across the cycle is built (Moore 1961; Shiskin 1961), the performance record of the signal disciplines how much weight a surprise deserves (Zarnowitz 1992), and turning points are dated by algorithm (Bry and Boschan 1971).

For credit and duration risk, the labor-market flows and the definitional gap determine whether a change in the level reflects hiring strength or compositional churn (Clark and Summers 1979; Hussmanns, Mehran, and Verma 1990). That a print can be revised afterwards is the reason to treat the count as a signal carrying measurement error (Stone, Champernowne, and Meade 1942), and the same number supports divergent trades depending on the model the trader imposes (Koopmans 1947).

USPAYROLL is a seasonally-adjusted-at-source total nonfarm payroll employment level, and the published level in thousands of persons is the statistical object whose integration order the battery assesses. The sample is 797 monthly observations spanning 1960-01 to 2026-05, and the tests are fitted with a constant and trend.

The augmented Dickey-Fuller regression of Dickey and Fuller (1979), in the lag-augmented form of Said and Dickey (1984), fails to reject the unit-root null with a statistic of −2.138 at p = 0.5247, the nonparametric test of Phillips and Perron (1988) also fails to reject with a statistic of −2.4344 at p = 0.3615, and the stationarity-null KPSS test of Kwiatkowski et al. (1992) rejects trend stationarity with a statistic of 0.6738 at p < 0.01. The three procedures agree in one direction, so the verdict is an unambiguous I(1). The GLS-detrended escalation of Elliott, Rothenberg, and Stock (1996) is reserved for ambiguous outcomes under the house protocol and is therefore not required on this clean reading, and no second-stage escalation aimed at near-unit moving-average errors is called for either (Ng and Perron 2001).

Because the level is integrated, the serial-correlation diagnostic and the mean-break search are run on the first difference, the stationary object those procedures require (Perron 1989; Bai and Perron 1998). The portmanteau statistic of Ljung and Box (1978), refining the form of Box and Pierce (1970), returns Q = 19.8735 at p = 0.0695 for lag 12 on the differenced series, which fails to reject the white-noise null at the 5 percent level while sitting close to that boundary, and Q = 21.6884 at p = 0.5979 for lag 24, further from rejection. The multiple-break procedure computed by the dynamic-programming algorithm of Bai and Perron (2003) finds no break in the mean of the differenced series, so the average monthly employment increment reads as one level across the sample.

The seasonal battery is run with the interpretation that any absence of seasonality reflects the source adjustment rather than an inherent property. The seasonal-unit-root test of Hylleberg et al. (1990), with the monthly mechanics of Beaulieu and Miron (1993), rejects unit roots at the seasonal frequencies of the level with a statistic of 708.6833 at p = 0.000, the seasonal dummies fitted on the first difference are not jointly significant with F = 0.9427 at p = 0.4982, and the QS statistic on that first difference at seasonal lags 12 and 24 is 21.688 at p = 0.5979, detecting no residual seasonality. That reflects the seasonal adjustment already applied at source rather than an inherent absence of a seasonal pattern, consistent with the stationary-seasonality reading of Canova and Hansen (1995) and the seasonal-adjustment literature (Ghysels and Osborn 2001).