Does headcount growth predict stock returns?

We tested our own tagline. It did not hold.

Corpus uses the tagline "Employees move before the earnings do." That is a testable claim, so we tested it on our own data: does year-on-year change in disclosed headcount predict the following 12 months of stock returns for Japanese listed companies?

Result: no reliable relationship

Across 3,005 observations we found no dependable link between employment momentum and subsequent 12-month returns. A gradient does appear in raw returns, but it disappears once industry and market period are neutralised, it reverses sign between the two largest cohorts, and it breaks if the entry date moves by one month. The tagline is not refuted — it is untested, because the free price history covers only about two years and therefore very few genuinely independent market periods.

Method

Signal: year-on-year change in consolidated headcount as disclosed in the annual securities report. Each filing carries five years of headcount, so one filing yields up to four (fiscal year, change) pairs — the same figures that were public at the time.
Entry: month end, three months after fiscal year end — after the filing deadline, so the information was already public.
Horizon: 12 months, using split-adjusted closing prices.
Median, not mean. On means the gradient looks far stronger (Q1 +6.1% → Q5 +13.5%), but that is a handful of right-tail outliers among fast-growing hirers.
Prices: J-Quants free tier (2024-05-10–2026-05-10, roughly 12-week delay), 4,511 tickers.

Raw returns suggest an effect; industry-relative ranks do not

Quintiles by employment momentum. The last column replaces the return with the company's rank within its own industry and entry date (0 = worst, 1 = best). A real effect would push that away from 0.5.

Quintile	n	headcount change (median)	12m return (median)	industry-relative rank (median)
Q1 (largest cuts)	601	-5.2%	-0.14%	0.500
Q2	601	-1.1%	-0.57%	0.466
Q3	601	+1.2%	+0.96%	0.537
Q4	601	+4.2%	+1.66%	0.500
Q5 (largest hiring)	601	+13.8%	+1.90%	0.500

Spearman rho on the industry-relative rank is +0.015 (p=0.445) — indistinguishable from zero. Most of the raw spread was industry and period composition.

The sign flips between cohorts

Cohort	n	Window	Q1→Q5 median return (%)	rho
FY end 2024-03	1,841	2024-06-28 → 2025-06-30	-3.2 -3.4 -1.5 -1.6 -1.7	+0.056
FY end 2024-12	434	2025-03-31 → 2026-03-31	+9.3 +12.5 +16.4 +8.4 +2.9	-0.054
FY end 2024-02	152	2024-05-31 → 2025-05-30	-4.8 +2.4 -0.9 -3.3 +7.1	+0.137
FY end 2024-09	132	2024-12-30 → 2025-12-30	+13.7 +4.4 +13.4 +8.0 +16.6	-0.021
FY end 2024-06	113	2024-09-30 → 2025-09-30	+6.1 +6.4 +15.5 +8.5 +21.6	+0.130

In the largest cohort the top hiring quintile trails the middle; in the second largest, the biggest hirers were the worst performers.

Limitations

This is close to a single cross-section. About 60% of observations sit in one 12-month window, so the effective number of independent periods is very small.
Survivorship bias. Only companies still filing today are present; delisted and failed companies are absent, and heavy headcount cutters drop out preferentially.
p-values are indicative only, since observations within a cohort are not independent.

Sources: annual securities reports (EDINET, Financial Services Agency of Japan); prices from J-Quants (JPX Market Innovation & Research), free tier. Full method and every figure: Japanese edition. Corpus does not recommend buying or selling any security.