Firm-Worker Matches: Experience or Inspection Goods? -- by Victoria Gregory, Guido Menzio, Giovanni M. Topa

URL SCAN: NBER Working Paper w35236
FIRST LINE: "We propose a novel empirical strategy to infer the extent to which firm-worker matches are inspection or experience goods."

THE DISSECTION

This is a technical labor economics paper from NBER studying information asymmetry in firm-worker matching — specifically whether match quality is revealed before hiring (inspection goods) or during employment (experience goods). Using German administrative data, the authors quantify signal strength, demographic variation, and aggregate productivity/output effects under counterfactual scenarios.

The paper is a genuine piece of empirical work. The methodology is sound. The data is high-quality. The counterfactuals are numerically explicit.

But read it with DT eyes and it becomes an autopsy of a system that doesn't know it's already on the table.

THE CORE FALLACY

The paper's entire framework presupposes that labor market matching is a permanent feature of the economic architecture. It treats the problem of information asymmetry between firms and workers as a design flaw to be optimized — not as a transitional friction in a system undergoing mechanical obsolescence.

The paper's core question — how efficiently can firms and workers learn about match quality before committing? — is only salient if human labor remains the primary mechanism of value production. If cognitive work is automated, the question dissolves entirely. The matching friction becomes as irrelevant as the optimal placement of whale oil lamps after electrification.

The 67% variance reduction figure is treated as a description of labor market reality. Under DT logic, it is better understood as a quantification of the remaining human labor problem's complexity — and the 2-4% output variations are the margin that makes AI replacement economically justified.

HIDDEN ASSUMPTIONS

1. Labor market matching remains the primary allocation mechanism for productive work. The paper never questions this premise.
2. Workers are the variable being optimized. The framing is firm-worker dyads, but the efficiency gains flow to firms (who bear search costs, bear training costs, bear mis-match costs). Workers are subjects of information asymmetry, not beneficiaries of reduced friction.
3. Increasing signal informativeness over time is a sign of market maturation. Under DT logic, this could be read as: firms increasingly cannot rely on signals and thus require longer "inspection windows," i.e., tenure data. This is a symptom of increasing matching complexity, which is exactly what AI eliminates.
4. The counterfactual output loss of 4% for pure experience goods is an inefficiency to be corrected. Under DT logic, this is a cost of human labor that machines don't incur. The "loss" is the economic case for replacement.

SOCIAL FUNCTION

This paper serves prestige economics signaling — it is technically rigorous, data-rich, and entirely irrelevant to the actual structural transition underway.

It also serves academic copium for the labor economics guild — a demonstration that their tools can quantify what AI will soon render moot. The authors are studying the friction of human matching with the precision of an engineer cataloging the drag coefficient of a sinking ship.

THE VERDICT

The paper is a precision measurement of the precise parameter that determines how quickly AI makes the measurement irrelevant.

The 67% signal informativeness figure is a description of remaining human uncertainty. The 2-4% productivity variation is the economic margin between human and AI labor efficiency. The demographic variation (gender, education in signal quality) is a documentation of which workers are most information-disadvantaged — and thus most automation-displaceable.

This is not a paper about labor markets. It is a paper about the specific magnitude of friction that makes AI replacement economically attractive. It quantifies with elegant rigor exactly how inefficient human labor remains — and hands that number directly to every AI developer building the replacement.

The paper is excellent microeconomics. It is irrelevant macroeconomics. The authors are measuring the half-life of their own subject matter with clinical detachment.