We ran a simple question that matters for IPO investors: do pre-IPO fundamentals reliably predict which new listings outperform in the first few months after pricing, or do the “signals” disappear once we control for statistical noise?
Run details (as-of 2026-07-23): We tested IPOs priced in the last 12 months and compared each pre-IPO factor to sector-neutralized forward returns over two horizons (3 months and 6 months). “Sector-neutralized” means we adjust returns so we’re testing stock-specific effects rather than broad sector moves.
A quick translation of the key metric: the Information Coefficient (IC), the Spearman rank correlation between a factor and the forward return that followed, asks “if we ranked IPOs by this factor, would that ranking resemble the return ranking?” (More on IC here: /faq#information-coefficient.) We call a factor robust only if it survives Benjamini–Hochberg false-discovery-rate correction, which controls for the fact that we test many factors and some will look “significant” by luck (why that matters: /faq#multiple-testing).
Which pre-IPO fundamentals predicted 3-month returns in the July 2026 run?
None survived multiple-testing (FDR). That is the main result.
A few factors still look strong in a simple one-factor view, but they are not robust. For example, operating cash flow growth (YoY) shows a high IC (0.555) on a small sample (n=24), and SG&A / revenue % is negative (IC -0.326), which matches the intuitive “leaner cost base did better” narrative. The issue is straightforward: once we acknowledge the full menu of factors we tested, these are exactly the kinds of “good-looking” outcomes we expect to see by chance.
Before the table, two columns to interpret:
- IC: direction and strength of the rank relationship.
- Q5−Q1 spread: the return gap between the top and bottom quintiles of the factor (a simple way to express “how different were outcomes at the extremes”).
Look for the same tell throughout: the Robust column is blank on every row.
| Variable | n | IC | p | Q5−Q1 spread | Robust |
|---|---|---|---|---|---|
| Operating cash flow growth % (YoY) | 24 | 0.555 | 0.010 | 116.021 | |
| Capex / revenue % | 38 | 0.436 | 0.300 | 45.671 | |
| Goodwill / assets % | 27 | 0.405 | 0.023 | 33.369 | |
| Gross profit growth % (YoY) | 31 | 0.327 | 0.105 | 43.976 | |
| SG&A / revenue % | 40 | -0.326 | 0.036 | -31.590 | |
| SG&A growth % (YoY) | 36 | 0.312 | 0.039 | 154.292 | |
| EBITDA growth % (YoY) | 28 | 0.277 | 0.027 | 101.723 | |
| Net income growth % (YoY) | 27 | 0.270 | 0.161 | 67.314 |
The chart below shows the same story visually: the “strongest-looking” ICs in this run. Use it as descriptive context, not a screen, because none of these passed FDR.
Strongest pre-IPO factors vs 3M return (IPOs priced in the last 12 months)
Two pattern-level observations (not robust findings):
- More “quality/momentum” than “story.” Cash-generation momentum (operating cash flow growth) and profit scaling (gross profit / EBITDA growth) are economically sensible candidates when they are real.
- Cost discipline flips quickly in small samples. SG&A intensity (SG&A / revenue %) has the expected negative sign, but SG&A growth (YoY) is positive here. That kind of internal inconsistency is common when IPOs are at different scaling stages and the eligible sample is thin.
Multivariate sanity check (3M): We also fit an Elastic Net model (a regularized regression that tends to zero-out weak or redundant predictors). It shrank every coefficient to zero, which is consistent with the univariate “edges” not being stable once we ask the factors to compete with each other.
Split by profitability (3M): Breaking the sample into profitable vs loss-making at IPO did not change the headline result: nothing survived FDR correction in either subgroup.
Which pre-IPO fundamentals predicted 6-month returns in the July 2026 run?
Again, none survived FDR.
At 6 months, the factors that rise to the top are more balance-sheet and gatekeeper-risk flavored. Interest coverage (EBIT / interest) is positive (IC 0.381); in plain English, firms that could more comfortably service debt tended to do better. The forensic risk score (0–100), a composite of filing red flags (for example, going-concern language and control weaknesses), is negative (IC -0.341). The directions make sense, but the robustness problem remains.
As above, the key check is that Robust is blank for every row.
| Variable | n | IC | p | Q5−Q1 spread | Robust |
|---|---|---|---|---|---|
| EBITDA/EV yield % | 29 | 0.460 | 0.118 | -657.359 | |
| Interest coverage (EBIT / interest) | 26 | 0.381 | 0.047 | 133.625 | |
| Forensic risk score (0-100) | 100 | -0.341 | 0.010 | -78.405 | |
| Asset turnover | 45 | 0.309 | 0.279 | 75.904 | |
| Earnings yield (E/P %) | 92 | 0.272 | 0.073 | 66.064 | |
| PFIC status + no-opinion hedge (0-1) | 100 | -0.259 | 0.058 | -250.579 | |
| Effective tax rate % | 31 | -0.256 | 0.091 | -54.266 | |
| Going-concern flag (0-1) | 100 | -0.237 | 0.064 | -277.982 |
The chart below highlights the top 6M ICs (descriptive only).
Strongest pre-IPO factors vs 6M return (IPOs priced in the last 12 months)
Multivariate sanity check (6M): The Elastic Net fit again shrank every coefficient to zero.
Split by profitability (6M): Profitable-at-IPO and loss-making-at-IPO subsamples also produced no FDR-robust factors.
What changed since the previous run (2026-06-30), and why does it matter?
The only clean message is the one investors least want: what looked robust did not repeat.
For 3M returns, the previous run had robust findings; this run has none.
Change since the previous run of 2026-06-30 (3M): Newly robust in this run: none No longer robust: Operating cash flow growth % (YoY), Going-concern flag (0-1) Held up in both runs: none
For 6M returns, there were no robust factors in either run.
Change since the previous run of 2026-06-30 (6M): Newly robust in this run: none No longer robust: none Held up in both runs: none
Why we care about repeatability is simple: if a fundamental factor is going to be used prospectively, it has to show up across runs, not just once.
If nothing is robust, what should we take from this?
Three conclusions are justified by what we see in the tables.
-
In this 1-year IPO cohort, pre-IPO fundamentals did not produce a stable, repeatable edge over 3–6 months. That is not the same claim as “fundamentals don’t matter.” It means that, with this sample and this framework, we cannot separate signal from noise at a strict robustness threshold.
-
Treat high ICs on small n as leads, not rules. Operating cash flow growth (YoY) at 3M has IC 0.555 on n=24. With that kind of coverage, a couple of additions/removals from the eligible set (including common definitional issues like negative prior-year bases that make growth rates unstable) can swing the result.
-
The most economically plausible themes remain plausible, but they are not statistically cleared here.
- Cash-generation momentum and profit scaling (operating cash flow growth, gross profit growth, EBITDA growth) are real operating attributes, but they can be noisy around IPO timing and inconsistent across issuers.
- Gatekeeper / structural red flags (forensic risk score, going-concern flag) should relate to downside and financing risk; the fact that they are not robust over 3–6 months is a reminder that markets can ignore risk markers for extended periods, and that short windows may not capture when those risks crystallize.
For methodology, the key guardrail is multiple-testing correction: /faq#multiple-testing. For the broader dashboard view of our factor cuts: /analytics#factor-summary.
What does this analysis not prove?
- It does not prove fundamentals don’t matter for IPOs; it shows that, for IPOs priced in the last 12 months and measured over 3M/6M windows, we did not identify factors that survive a strict robustness bar.
- It does not prove the “best-looking” factors are false; only that they did not survive false-discovery-rate control in this run, and the prior run’s robust results did not repeat.
- It does not generalize to longer horizons (1Y/2Y+) or different sampling designs; the look-back here is 1 year and the look-ahead is 3–6 months.
Bottom line: in the July 2026 run, the data does not support a reusable pre-IPO fundamental screen for near-term (3–6 month) aftermarket returns. We treat the factor tables as hypothesis generation and require repetition across runs before elevating any single variable to an investable rule.