South Korea Construction Employment

KREMPMINING, KREMPCONST, and KREMPTRADE are three not-seasonally-adjusted monthly counts of employed persons by industry that KRED carries without transformation, each recorded exactly as published. The employed are defined within the standard labor-force framework as those who worked for pay or profit, or were temporarily absent from such work, during a short reference period (Hussmanns, Mehran, and Verma 1990). Industry is attached to the main job held in that reference period, so each person enters exactly one industry count, and the three counts are subsets of total employment rather than an exhaustive partition of it.

These three series are headcounts of labor input rather than measures of output. Each moves with hiring, with separation, and with the composition of jobs, and each stands apart from the measurement tradition that builds industry volume indices by deflation and quantity relatives (Fabricant 1940). An industry employment count and an industry output measure can move in opposite directions when productivity or hours absorb a shock, so the two are read as separate measurements of the same industry.

Each count is the net outcome of continuous flows of hiring, separation, and spell duration (Clark and Summers 1979), so a single monthly reading summarizes the churn beneath a headline stock.

Within the cyclical-indicator tradition industry employment counts are comoving series whose turning points cluster around aggregate reference cycles (Burns and Mitchell 1946). They are classified by their timing relative to the cycle (Mitchell and Burns 1938), their turns are dated by a reproducible algorithm (Bry and Boschan 1971), and the share of industries expanding is the raw material of diffusion-index construction (Moore 1961).

Industry detail of this kind is recast as an observable shadow of a latent state of the economy (Stock and Watson 1989; Stock and Watson 1991). The measurement and timing properties of such series are catalogued systematically (Zarnowitz 1992), and each value is treated as a monthly current-conditions signal (Shiskin 1961). Such atheoretical measurement, however, carries meaning only against an explicit model of the labor market (Koopmans 1947).

KRED reports each figure exactly as published and applies no transformation, neither rescaling, deflating, smoothing, annualizing, nor seasonal adjustment. Because the underlying counts are not seasonally adjusted, each retains its within-year pattern, which is pronounced in construction where weather and the building calendar move hiring month by month, and the raw passthrough leaves that pattern intact.

The measurement basis is the household labor-force survey, in which activity status over a fixed reference period is translated into a discrete category under internationally agreed status rules (Hussmanns, Mehran, and Verma 1990), following the survey-design logic first treated scientifically for activity questions (Anderson 1952). Industry is coded from the activity of the establishment at which the main job is held, so a person who changes industry between reference periods moves between counts with no change in total employment. The flow accounting that links statuses across months decomposes the change in each count into hiring, separation, and duration (Clark and Summers 1979).

The three counts are not exhaustive and do not sum to total employment, since the remaining industries are carried in neither of them. KRED applies no adjustment to enforce an adding-up relation, computes no shares, and rescales nothing, and each count retains whatever revision, industry-classification change, or rounding the source vintage carries. No deflation or quantity-relative step of the kind used to build industry volume indices is applied (Fabricant 1940), because the object measured is a headcount and not a volume.

Within the cyclical-indicator tradition these levels are comoving series whose turning points cluster around aggregate reference cycles (Burns and Mitchell 1946), classified by their timing relative to a turn (Mitchell and Burns 1938), and catalogued systematically for their measurement and timing properties (Zarnowitz 1992). They are treated as noisy readings of a latent state of the economy under the dynamic-factor representation (Stock and Watson 1989; Stock and Watson 1991).

The cyclical reading of the resulting figures follows a reproducible turning-point algorithm that dates their specific cycles (Bry and Boschan 1971), a codified embedding of the expanding share of industries in diffusion and composite indexes (Moore 1961), and a formalized monthly real-time construction (Shiskin 1961). This measurement procedure nonetheless presupposes an economic model for its interpretation (Koopmans 1947).

Taken together the three counts split aggregate employment into sectors with distinct cyclical drivers, where mining and manufacturing responds to external demand and capital spending, construction turns with building activity, and wholesale, retail, accommodation, and food services sits closest to household consumption. Those differing timing properties are what justify classifying indicators by their relation to the cycle (Mitchell and Burns 1938; Burns and Mitchell 1946), and they are the starting point for counting the share of industries expanding in a diffusion index (Moore 1961).

Even when aggregate employment is flat, a fall in mining and manufacturing can offset a rise in wholesale, retail, accommodation, and food services and conceal a shift in industry composition, so the three counts read together reveal the reallocation that a single headline hides. Movement between industries and its welfare content are disciplined by the flow decomposition (Clark and Summers 1979), and the boundary of each category is set by the standard labor-force framework (Hussmanns, Mehran, and Verma 1990).

These levels are treated as noisy readings of a latent state of the economy (Stock and Watson 1989; Stock and Watson 1991), their timing and reliability have been assessed systematically (Zarnowitz 1992), turning-point dating separates cyclical from incidental movements (Bry and Boschan 1971), and each is used as a monthly current-conditions gauge (Shiskin 1961).

Employment in the consumption-facing sector governs household income expectations and so links directly to spending behavior (Katona 1951), and labor readings of this kind carry forward-looking content for consumption (Carroll, Fuhrer, and Wilcox 1994; Ludvigson 2004).

An industry employment count is a headcount of labor input, however, and must not be read as a measure of industry output, from which it diverges whenever productivity or hours change (Fabricant 1940). The levels inform but do not identify the mechanism (Koopmans 1947).

For markets the industry employment counts decompose a headline employment print into the sectors that produced it, and their classification by cycle timing governs how a given release maps into expectations for the policy path (Mitchell and Burns 1938; Burns and Mitchell 1946).

Fixed-income and rate markets track them as components of a latent-state estimate (Stock and Watson 1989; Stock and Watson 1991), and how much weight a surprise in any one sector deserves is disciplined by that series' forecasting record (Zarnowitz 1992).

An unexpected move in the consumption-facing count shifts the expected trajectory of growth and inflation, because labor readings carry independent predictive content for household demand (Carroll, Fuhrer, and Wilcox 1994; Ludvigson 2004) and because employment conditions the sentiment and spending channel (Katona 1951).

For credit risk the sectoral split reveals exposure that the aggregate conceals, since construction employment is read alongside building and property exposure while mining and manufacturing employment is read alongside the export cycle. Whether a change in a level reflects hiring strength or reallocation between industries is settled by the labor-market flows and the status accounting (Clark and Summers 1979; Hussmanns, Mehran, and Verma 1990).

Each release is positioned within a diffusion, composite, and real-time signaling frame (Moore 1961; Shiskin 1961), and phase transitions are flagged by turning-point dating (Bry and Boschan 1971). The same figures nonetheless support divergent trades depending on the model the trader imposes (Koopmans 1947).

The construction employment count is published as a not-seasonally-adjusted level, and the appropriate statistical object is its logarithm. The sample is 162 monthly observations spanning 2013-01 to 2026-06, and the unit-root triplet is fitted on the log level 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), does not reject the unit root with a statistic of −0.545 at p = 0.982, the nonparametric test of Phillips and Perron (1988) likewise does not reject with a statistic of −2.357 at p = 0.403, and the stationarity-null test of Kwiatkowski et al. (1992) rejects trend stationarity with a statistic of 0.357 at p < 0.01. All three procedures therefore concur on a clean I(1) classification of the log level without ambiguity. The GLS-detrended escalation of Elliott, Rothenberg, and Stock (1996) is reserved for ambiguous outcomes under the house protocol and is not needed on this clean reading.

Because the log level is not a stationary object, the portmanteau and the break search are run on its first difference, since a level portmanteau on a near-unit-root series reads mechanical near-unit autocorrelation and a level mean-break search spuriously segments a stochastic trend (Perron 1989; Hamilton 2018). The portmanteau statistic of Ljung and Box (1978), refining the form of Box and Pierce (1970), rejects the white-noise null on the differenced log level, with Q = 158.740 at lag 12 and Q = 307.123 at lag 24, each at p = 0.000, and much of that remaining dependence is the deterministic seasonal structure identified below. The multiple-break procedure of Bai and Perron (1998), computed by the dynamic-programming algorithm of Bai and Perron (2003), locates no break in the mean of the differenced series.

Seasonal tests are appropriate on this unadjusted monthly level. 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 with a statistic of 72.600 at p = 0.000, the seasonal dummies are jointly significant with a statistic of 22.808 at p = 0.000, and the seasonal portmanteau at lags 12 and 24 also rejects with a statistic of 309.881 at p = 0.000, so the seasonality is deterministic rather than stochastic and its amplitude is the largest of the three industry counts. This is consistent with the stationary-seasonality reading of Canova and Hansen (1995) and with the not-seasonally-adjusted character of the series (Ghysels and Osborn 2001).