Leverage and profitability are negatively rank-correlated among Polish firms in every year from 2008 to 2010 and in every robustness check run against it — but the size of that link collapsed by nearly three-quarters even as its statistical significance never wavered.
A Statistically Airtight Result That Barely Moves the Needle
In the pre-registered 2010 cross-section of 3,581 Polish firms, financial leverage (total debt / assets) and profitability (ROA, net profit / assets) are negatively rank-correlated: Spearman's rho = −0.058, 95% CI [−0.092, −0.024], p = 5.03e-04. The confidence interval sits entirely below zero — more debt tracks with lower profitability, and the test says so with unambiguous statistical confidence.
Turn that rho into a variance-explained number, though, and the picture changes. Rho-squared is 0.00338 — leverage rank and profitability rank share only 0.34% of variance, and even at the most generous edge of the 95% confidence interval, the shared variance tops out at 0.86%. A finding this statistically airtight is, in practical terms, close to describing nothing about any individual firm.
The point sits reliably left of zero — but only barely. rho² = 0.34% of shared variance.
That gap — a p-value small enough to convince a skeptical referee, an effect size small enough to be almost invisible — is the real subject of this piece. It only gets stranger once you watch the number move over time.
Where This Number Comes From, and Why 2008-2010 Isn't a Neutral Window
This result comes from a specific, real dataset: 14,421 firm-year rows compiled by Grzegorz Michalski, a corporate-finance professor at Wroclaw University of Economics and Business, from audited Polish firm financial statements spanning 2008-2010. After dropping the one row with a missing year label and the handful of rows with non-positive assets, 14,412 rows enter at least one of four year-labeled cross-sections: 2008, 2009, an oddly-behaved separate batch labeled "2009b," and the primary year, 2010.
Michalski built this panel as reusable raw material — he has drawn on the same underlying Polish-firm data for a series of his own published research on corporate liquidity and cash management — rather than as a single-purpose dataset built to prove one theory.
And 2008-2010 is not an arbitrary window. It is the acute phase of the global financial crisis, and Poland was the only European Union economy to avoid recession outright — posting roughly 2.6-2.8% GDP growth in 2009 while Germany contracted more than 5% and the UK more than 4%, then accelerating to about 3.2% growth in 2010. Whatever this dataset shows about Polish firms, it shows it inside an economy that was, unusually among its peers, still expanding throughout the crisis.
The Effect That Nearly Vanished
Inside the three primary-family years, the correlation doesn't just stay negative — it collapses. Rho falls from −0.2118 in 2008 to −0.0683 in 2009 to −0.0581 in 2010: the 2010 effect is only 27.4% the size of the 2008 effect, a 72.6% shrinkage in magnitude over two years. Most of that drop happens in the first step — a 67.8% fall from 2008 to 2009 — with a smaller further 14.9% fall into 2010.
Poland avoided recession through 2009-2010 while the rest of the EU contracted. By 2010, do you think the leverage-profitability link got stronger, stayed about the same, or nearly vanished? Drag to place your guess for 2010, then reveal the real number.
Volume and length = strength of the leverage-profitability link, year by year. 2008 plays loudest and longest; 2010 plays quietest and shortest.
What makes this genuinely strange is that every one of those three years still clears Holm-Bonferroni correction by an enormous margin — 2008 at p_holm=5.44e-38, 2009 at p_holm=1.01e-04, 2010 at p_holm=5.03e-04. None of that is possible without a sample bigger than 3,500 firms in every year: a sample that large can make even a nearly-vanished effect statistically undeniable. The p-values stayed extreme while the thing they were measuring shrank by three-quarters.
Something else was happening underneath that decline. The typical firm in this sample cut its leverage ratio by about a quarter between 2008 and 2010 (median 0.4985 to 0.3731, −25.2%) while its ROA rose by about 15% (median 0.0423 to 0.0485, +14.5%) — both changes point the same direction the correlation implies. As the whole distribution of firms shifted toward less debt and more profit together, there was mechanically less room left for leverage rank to distinguish profitability rank within any single year.
One wrinkle worth a footnote: the "2009b" batch — kept deliberately separate throughout because it shares almost no firm-registry matches with 2009 — actually behaves like 2008, not like its same-named neighbor. Its rho of −0.1996 sits just 0.0122 away from 2008's −0.2118, versus 0.1313 away from 2009's −0.0683. Whatever separates 2009b from 2009 doesn't separate it from the older, larger-effect year.
Ruling Out the Obvious Objections
Could this be an outlier artifact — a handful of firms with tiny denominators producing huge, distorting ratios? Two checks say no. Winsorizing the top and bottom 1% of values moves the 2010 rho by just 0.21%, from −0.05811 to −0.05823. Trimming those same extremes outright (dropping 121 of 3,581 rows rather than capping them) moves it a bit further, to −0.05996 — a 3.19% shift, still comfortably inside the primary result's own confidence interval.
Raw, winsorized, and trimmed rho — a 0.21% and 3.19% move. Practically flat.
It also isn't an artifact of aggressive data cleaning. Only 8 of 14,420 primary-plus-batch rows (0.055%) were excluded for non-positive assets, and the 2010 cross-section itself has zero exclusions. These are close to the complete raw files for each year, not curated subsets shaped by exclusion choices.
The more interesting robustness check runs the opposite direction. Collapsing to one mean observation per registry-identified firm — guarding against the same firm appearing, and being double-counted, across multiple years — produces rho=−0.1856 across 5,710 firms (p=1.94e-45). That's 3.19 times the magnitude of the pooled 2010 result, and nearly matches 2008's −0.2118 (0.88x). Correcting for pseudoreplication doesn't shrink this finding. It makes it bigger.
Firm-level aggregation is 3.19× the magnitude of the 2010 pooled result.
Why "Statistically Significant" Isn't the Same Claim as "True"
None of this is a novel discovery. A negative leverage-profitability correlation is one of the most consistently replicated findings in corporate finance, going back to Rajan and Zingales' 1995 study of 3,569 firms across the G7 economies. It is also, at the same time, a known open puzzle: Frank and Goyal's widely cited 2003 survey singles out exactly this negative relationship as the pattern trade-off theory struggles hardest to explain, while survey evidence on how CFOs actually make financing decisions often matches neither trade-off nor pecking-order theory cleanly.
There's a more basic caution too. ROA and leverage are both built by dividing a different numerator by the same denominator — Assets. Any time two ratios share a denominator, part of their correlation can be mechanical rather than economic, a statistical wrinkle documented since Karl Pearson's 1897 work on "spurious correlation" of ratios. That doesn't make the finding wrong. It means "leverage is associated with lower ROA" should be read as a descriptive, ratio-level pattern — not proof that debt is eroding operating performance, or evidence ruling out the reverse story that low performers simply borrow more.
Put those two cautions together and the honest headline isn't "leverage hurts profitability." It's closer to: a well-known, internationally-replicated pattern shows up again here, present with total statistical confidence and almost no practical size, in a form that can't cleanly separate correlation from mechanics, let alone cause from effect.
What Polish Firms Were Actually Doing
Zoom back out to the calendar. 2008 — the year with by far the strongest leverage-ROA link in this dataset — was the year the global financial crisis hit hardest everywhere else in Europe. 2010 — the year with the weakest link — was the year Poland's growth accelerated past 3%, outperforming the entire EU. As the "green island" economy kept expanding while its neighbors contracted, Polish firms broadly deleveraged and grew more profitable together, and the very relationship this dataset set out to measure quietly faded toward zero.
That leaves an open question the 2008-2010 window alone can't answer: was 2010's weak correlation the new normal for a maturing, deleveraging corporate sector, or just the bottom of a cycle that would snap back the next time credit conditions tightened again? The data behind this piece stops at 2010. The pattern it leaves behind — significance without size, replication without resolution — is the more durable finding.
Methodology note: this analysis worked from the underlying research object's derived, independently re-executed record (five result tables, results.json, claims.json) rather than the raw spreadsheet directly — an independent re-execution reproduced 98 of 98 compared statistical values with zero numeric mismatches. This is an associational, cross-sectional finding only; no causal claim is made or supported.