We tested a simple question: do pre‑IPO fiscal‑year fundamentals (numbers you can read in the prospectus at pricing) line up with post‑IPO forward returns in this month’s factor run, as of 2026-09-01.
Why this matters: a lot of “IPO factor” commentary is really just sector exposure (owning what was hot) or multiple comparisons (when you test many variables, a few will look significant by luck). We’re trying to isolate fundamentals that predict within-sector, within-vintage returns, and then check whether the apparent winners still hold up after correcting for how many factors we tested.
We look at two horizons because the first months after an IPO are usually driven by positioning, liquidity, and narrative repricing, while the 6‑month window is more likely to reflect the market’s read on balance-sheet risk and profitability.
Bottom line from this run:
- At 3 months, no factor survives our false-discovery-rate filter, and the multivariate model keeps no signal.
- At 6 months, several “financial quality” measures (especially interest coverage and margins) do survive the higher bar, but the multivariate model still posts negative out-of-sample R². We treat this as evidence for ranking/triage, not a forecasting engine.
What exactly did (and didn’t) predict 3‑month returns in this September 2026 run?
At the 3‑month horizon (returns over the first three months of trading) for IPOs priced in the last 12 months, several fundamentals show large Information Coefficients (ICs). The Information Coefficient (IC), the Spearman rank correlation between a factor and the forward return that followed is a simple “did higher values tend to coincide with higher subsequent returns?” score.
However, none of the factors below survive Benjamini–Hochberg false-discovery-rate correction, which is the procedure we use to reduce false positives when we test lots of variables see /faq#multiple-testing for why this matters.
Use the table like this: IC gives direction and strength; the Q5−Q1 spread shows how far apart the top and bottom quintiles were.
3M forward return — IPOs priced in the last 12 months Sample window: 1Y. Robust after FDR correction: 0.
| Variable | n | IC | p | Q5−Q1 spread | Robust |
|---|---|---|---|---|---|
| SG&A / revenue % | 28 | -0.546 | 0.024 | -54.821 | |
| SG&A growth % (YoY) | 29 | 0.436 | 0.024 | 181.528 | |
| Goodwill / assets % | 22 | 0.425 | 0.056 | 36.860 | |
| EBITDA margin % | 33 | 0.417 | 0.071 | 67.380 | |
| Effective tax rate % | 24 | -0.389 | 0.947 | -44.841 | |
| Asset turnover | 33 | 0.331 | 0.117 | 32.171 | |
| Operating margin % | 33 | 0.331 | 0.098 | 67.386 | |
| Net margin % | 33 | 0.301 | 0.246 | 67.386 |
Strongest pre-IPO factors vs 3M return (IPOs priced in the last 12 months)
How we read this (and how we don’t read it):
- SG&A / revenue % (SG&A overhead ÷ revenue) is strongly negative (IC -0.546): lower overhead names did better in this sample.
- EBITDA margin % (EBITDA ÷ revenue) and operating/net margin also point in the “profitability helps” direction.
- But with zero factors robust after multiple-testing correction, this remains a watchlist of candidates, not a set of tradeable 3‑month signals.
Multivariate (3M):
Elastic Net shrank every coefficient to zero.
Elastic Net is a regularized regression that penalizes complexity; if it zeroes out everything, it’s effectively telling us the data can’t justify a stable model at this horizon.
What predicted 6‑month returns—and which fundamentals were genuinely robust?
At 6 months, more of the variance looks tied to “quality” fundamentals, and 8 variables are robust after FDR correction.
In the table, focus on the checkmarks: those are the factors that still cleared the statistical bar after accounting for how many tests we ran.
6M forward return — IPOs priced in the last 12 months Sample window: 1Y. Robust after FDR correction: 8.
| Variable | n | IC | p | Q5−Q1 spread | Robust |
|---|---|---|---|---|---|
| Interest coverage (EBIT / interest) | 21 | 0.690 | <0.001 | 274.447 | ✓ |
| EBITDA margin % | 32 | 0.520 | <0.001 | 223.001 | ✓ |
| EBITDA growth % (YoY) | 22 | 0.497 | 0.031 | 180.838 | |
| SG&A / revenue % | 27 | -0.496 | 0.177 | -131.732 | |
| Operating margin % | 32 | 0.490 | <0.001 | 217.824 | ✓ |
| SG&A growth % (YoY) | 27 | 0.468 | 0.011 | 236.372 | ✓ |
| Asset turnover | 32 | 0.456 | 0.077 | 193.601 | |
| Net margin % | 32 | 0.444 | <0.001 | 217.824 | ✓ |
Strongest pre-IPO factors vs 6M return (IPOs priced in the last 12 months)
What the robust 6‑month results have in common is straightforward:
- Interest coverage (EBIT ÷ interest expense) is the strongest relationship (IC 0.690, Q5−Q1 274.447, robust). Plain English: it measures how comfortably operating profit can pay interest. It often functions as a proxy for balance-sheet fragility and refinancing risk.
- EBITDA/operating/net margins (all robust) tell the same story in different accounting lines: investors rewarded businesses that already convert revenue into profit.
- SG&A growth % (YoY) is also robust (IC 0.468). We do not read this as “spend more and win.” In this sample, faster SG&A growth coincided with better 6‑month returns, potentially because the spend was associated with scaling without destroying margins (consistent with the margin signals). We would not generalize beyond that without a deeper issuer-level interaction analysis.
Two factors that look intuitive but did not clear the robustness bar here:
- SG&A / revenue % is directionally negative (leaner looks better) but not robust.
- EBITDA growth % (YoY) is positive but not robust.
Net: at 6 months, this run looks more like a solvency-and-profitability tape than a growth tape.
If these 6‑month signals are robust, why does the multivariate model still look bad?
Because robust univariate relationships are not the same as a reliable prediction model.
Here’s the multivariate fit when a cross-validated Elastic Net chooses a sparse model. Cross-validation means the model is repeatedly trained on part of the data and tested on held-out names to estimate how it would perform on unseen IPOs.
Multivariate (6M):
Cross-validated Elastic Net (n=91, α=98.65, CV R²=-0.048), standardized coefficients:
| Variable | Coefficient |
|---|---|
| Auditor-quality flag (0-1) | 0.5149 |
| Going-concern flag (0-1) | -0.1723 |
| UW syndicate: high-risk / cross-border flipper (0/1) | 0.1643 |
A negative cross-validated R² (here -0.048) means that, out of sample, the model did worse than a dumb baseline that just predicts the average return.
Two reasons this is common in IPO data:
- Small samples and high noise. Even at 6 months, issuer-specific events and post-IPO supply/demand can dominate what the prospectus said.
- Collinearity among “quality” metrics. Margins, interest coverage, and disclosure flags can move together. The model can struggle to choose one “winner” consistently across folds, so performance degrades.
It’s also notable that the multivariate selection leans on disclosure/credibility flags (auditor quality, going concern) and a syndicate-type flag. For IPOs, deal mechanics and governance can matter alongside the income statement.
What should an investor actually take from this run (without overfitting it)?
-
Don’t expect pre‑IPO fundamentals to be a 3‑month edge. In this run, nothing clears the multiple-testing bar, and the regularized model finds no stable signal.
-
At 6 months, the market rewarded “quality,” not novelty. Interest coverage and profitability margins were the cleanest robust relationships in the univariate tests.
-
Use these as a screen, not a forecast. The multivariate model’s negative cross-validated R² is the constraint: this is more useful for avoiding weak balance sheets and thin profitability than for picking precise winners.
For more on how we prevent sector moves from masquerading as “fundamentals,” see sector confounding and sector-neutralization. For why we don’t trust raw p-values across dozens of variables, see multiple testing and false discoveries.
What this analysis does not prove (limits to keep in mind)
- No causality. A high interest coverage ratio didn’t “cause” higher returns; it may simply proxy for a broader bundle of quality attributes.
- No guarantee of stability. The sample is a rolling set of IPOs priced in the last 12 months; results can change as cohorts enter and exit.
- Not a tradable forecasting model. The 6‑month multivariate fit is negative out of sample; even with robust univariate signals, prediction remains weak.
If you want the cleanest takeaway: 3‑month IPO outcomes in this run are still dominated by tape and flow, while 6‑month outcomes start to reward pre‑IPO financial quality. The edge is promising but fragile, and it needs repeatability across future runs before we would size it like a durable factor.