We run this research to answer a practical question: based only on what’s disclosed in an IPO’s pre‑IPO filing, which fundamentals help us avoid the worst aftermarket outcomes or tilt toward better ones?
For the August 2026 update, we test dozens of filing‑ and accounting‑derived variables against sector‑neutralized forward returns at 3 months and 6 months, using IPOs priced in the last 12 months (look‑back = 1Y; look‑ahead = 3M/6M). “Sector‑neutralized” means we subtract each IPO’s subsector cohort mean return before scoring a factor, so we are not just rediscovering “this sector did well.”
Here’s the takeaway: nothing survives Benjamini‑Hochberg false‑discovery‑rate (FDR) correction, and the multivariate models have negative out‑of‑sample fit. In other words, we do not have a statistically robust, tradeable “fundamentals factor” for recent IPOs right now. What we do have is a short list of variables that show plausible relationships (especially filing red flags). We treat those as diligence triage, not as an alpha screen.
What, exactly, counts as “predicted returns” in this run?
We use the Information Coefficient (IC), which is the Spearman rank correlation between a factor and the forward return that followed. Plain English: if we rank IPOs by a metric, does that ranking line up with the later return ranking? In cross‑sectional equity work, |0.05| is worth a look and |0.10| is strong, because returns are mostly noise.
We also show a quintile spread (Q5–Q1): we sort IPOs into five buckets by the factor, then subtract the bottom bucket’s average return from the top bucket’s. This is a quick check for economic size and whether the relationship looks reasonably monotonic.
Because we test many variables at once, we only call something robust if it survives Benjamini‑Hochberg (FDR) correction, a standard multiple‑testing adjustment that reduces false positives from “lucky” significance. In this update, robust after FDR correction: 0 for both horizons.
If you want the method details behind IC and multiple‑testing, see: /faq#information-coefficient and /faq#multiple-testing.
Which pre‑IPO fundamentals looked strongest for 3‑month returns, and how should we read them?
Even though nothing is robust after FDR correction, the univariate table is still useful as a map of what wanted to work in the last year of IPOs. When you read it, focus on IC sign/magnitude, sample size (n), and whether the Q5–Q1 spread is meaningfully sized.
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 |
|---|---|---|---|---|---|
| Goodwill / assets % | 23 | 0.481 | 0.016 | 34.002 | |
| SG&A growth % (YoY) | 31 | 0.465 | 0.005 | 192.577 | |
| Capex / revenue % | 33 | 0.408 | 0.614 | 42.678 | |
| SG&A / revenue % | 33 | -0.382 | 0.079 | -32.958 | |
| EBITDA growth % (YoY) | 25 | 0.355 | 0.041 | 125.672 | |
| Gross profit growth % (YoY) | 28 | 0.350 | 0.126 | 68.174 | |
| EBITDA margin % | 40 | 0.301 | 0.190 | 72.188 | |
| Op. cash flow / revenue % | 41 | 0.294 | 0.154 | 49.214 |
The chart below shows the same thing in one glance: which factors posted the largest ICs vs 3M returns.
Strongest pre-IPO factors vs 3M return (IPOs priced in the last 12 months)
What these factors mean in plain English, and why they might connect to 3‑month outcomes:
-
Goodwill / assets % (IC 0.481, n=23): goodwill divided by assets signals an acquisition‑heavy balance sheet and potential impairment risk. Here it correlates positively with 3M returns, which is counterintuitive. We read this as “the market recently rewarded certain consolidator stories,” not as a stable premium. With n=23, it’s fragile.
-
SG&A growth % (YoY) (IC 0.465, n=31): year‑over‑year growth in overhead. Normally, SG&A rising faster than revenue can be a cost‑discipline warning, but early aftermarket trading can reward “spend into growth” narratives. The very large Q5–Q1 spread (192.577) suggests a handful of extreme outcomes may be driving the relationship.
-
SG&A / revenue % (IC -0.382, n=33): overhead burden. The negative sign is the intuitive one. Very high SG&A intensity often reads like an unscaled organization, and that can get punished once the first public‑company reality check hits.
-
EBITDA growth % (YoY) (IC 0.355, n=25) and gross profit growth % (YoY) (IC 0.350, n=28): “momentum in the profitable core” measures (computed only when the prior year was already positive). They can look good partly because they select for cleaner financial trajectories. The trade‑off is smaller n, which raises noise.
Do the multivariate models confirm any of that at 3 months?
No. In the multivariate step we use a cross‑validated Elastic Net. Elastic Net is a regularized regression that shrinks and selects coefficients, and cross‑validation means we repeatedly train on one slice of the sample and score on held‑out IPOs to see whether it generalizes.
The key diagnostic is CV R². If cross‑validated R² is negative, the model predicts worse than a naive baseline of “assume every IPO will be average” on the hold‑out folds.
Multivariate (3M): Cross-validated Elastic Net (n=134, α=1.93, CV R²=-0.013), standardized coefficients:
| Variable | Coefficient |
|---|---|
| Going-concern flag (0-1) | -8.3542 |
| Deal-structure flag (0-1) | 4.6117 |
| Forensic risk score (0-100) | -4.4219 |
| UW syndicate: small/micro-cap specialist (0/1) | 4.2735 |
| Market cap ($M) | -3.0150 |
| Sales yield (S/P %) | -2.2488 |
We still take directional intuition from the signs, but we do not treat the weights as a reusable formula:
- Going‑concern flag (-8.3542) and forensic risk score (-4.4219) both push returns down. That matches the basic idea that explicit survival and gatekeeper red flags are hard to “narrative” away once the stock trades on fundamentals.
- With negative CV R², we assume the exact weights are unstable and regime‑dependent.
Which pre‑IPO fundamentals looked strongest for 6‑month returns, and what’s the story behind them?
At 6 months, the univariate list shifts toward what we would normally expect to matter with more time: balance‑sheet survivability and filing red flags.
6M 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 |
|---|---|---|---|---|---|
| EBITDA growth % (YoY) | 25 | 0.395 | 0.057 | 166.334 | |
| Forensic risk score (0-100) | 95 | -0.390 | 0.004 | -113.558 | |
| Interest coverage (EBIT / interest) | 25 | 0.385 | 0.043 | 157.426 | |
| Going-concern flag (0-1) | 95 | -0.358 | 0.011 | -330.680 | |
| Asset turnover | 39 | 0.283 | 0.487 | 59.659 | |
| Net income growth % (YoY) | 24 | 0.273 | 0.202 | 169.333 | |
| SG&A growth % (YoY) | 28 | 0.269 | 0.112 | 134.222 | |
| Piotroski F-score (0-8) | 30 | 0.265 | 0.126 | 99.101 |
The chart below highlights the same “top IC” set. The main thing to notice is that the strongest negatives are filing/survivability signals.
Strongest pre-IPO factors vs 6M return (IPOs priced in the last 12 months)
Two relationships here are straightforward:
-
Forensic risk score (IC -0.390, n=95; Q5–Q1 -113.558): this composite score rises when the filing contains more structural red flags (going‑concern doubts, internal‑control material weakness admissions, weaker auditor/gatekeeper signals, and deal-structure flags). The negative sign says: more “smoke” in the filing tends to show up as worse 6‑month performance.
-
Going‑concern flag (IC -0.358, n=95; Q5–Q1 -330.680): a going‑concern disclosure is explicit language that the company may need financing or a turnaround to stay viable. The large spread is consistent with a simple mechanism: downside becomes dominated by financing/dilution risk and confidence shocks, which compound over months.
On the positive side:
- Interest coverage (IC 0.385, n=25): EBIT‑to‑interest is a basic “can you service your debt?” measure. It is only defined for issuers with meaningful interest expense and positive EBIT, so the sample is small.
Do the multivariate models confirm anything at 6 months?
No, not in a way that generalizes.
Multivariate (6M): Cross-validated Elastic Net (n=95, α=18.88, CV R²=-0.016), standardized coefficients:
| Variable | Coefficient |
|---|---|
| Going-concern flag (0-1) | -4.8376 |
| Auditor-quality flag (0-1) | 4.0491 |
| Forensic risk score (0-100) | -2.4725 |
| Market cap ($M) | -2.4451 |
| UW syndicate: high-risk / cross-border flipper (0/1) | 2.4325 |
| UW syndicate: small/micro-cap specialist (0/1) | -2.3385 |
Even with negative CV R², the ranking is directionally coherent: gatekeeper quality and explicit distress flags show up with meaningful weights (going‑concern and forensic risk negative; auditor quality positive). We use this less as “factor investing” and more as “do not ignore filing red flags.”
What changed since the previous run (2026‑07‑23)?
Nothing, which is the point.
Change since the previous run of 2026-07-23 (3M):
- Newly robust in this run: none
- No longer robust: none
- Held up in both runs: none
Change since the previous run of 2026-07-23 (6M):
- Newly robust in this run: none
- No longer robust: none
- Held up in both runs: none
Our read is simple: we still do not have a factor that clears the bar for statistical dependability once we account for how many variables we tested. In a noisy, small, regime‑dependent IPO tape, apparent “signals” can flip just because a few new deals enter the 1Y look‑back or a few large moves roll into the 3M/6M look‑ahead.
What should a reader actually take from this?
Three practical takeaways, none of which are “buy this factor.”
-
Use pre‑IPO fundamentals as a risk filter first, not a return optimizer. The most plausible 6‑month relationships are tied to survivability and disclosure red flags (going‑concern language; forensic risk). Those do not tell us what will outperform; they flag what can blow up.
-
Be skeptical of short‑horizon “winners.” The 3‑month leaders include SG&A growth and goodwill intensity with very large spreads and small n. That profile is consistent with cohort quirks and flow, not a durable premium.
-
If a multivariate model cannot beat the mean out of sample, do not operationalize it. Both horizons show negative cross‑validated R². That is the data saying: a linear combination of these fundamentals does not reliably predict the cross‑section of recent IPO returns under an honest test.
What are the limits of this run (what it does not prove)?
- This does not show fundamentals never matter. It shows that, for IPOs priced in the last 12 months, with these definitions, we do not have factors that survive multiple‑testing and hold up out of sample.
- The highest ICs in the tables are not “the best factors.” They are pre‑FDR results, so some will look good by chance.
- We are not measuring longer horizons. Many accounting quality and profitability effects show up over 12–24 months, and we are not testing that here.
- Several metrics exist only for subsets of issuers. EBITDA growth, interest coverage, and net income growth require positive prior‑year numbers (and other definitional constraints), which shrinks n and increases fragility.
If you want the deeper dashboards behind these tables, see /analytics#factor-summary and /analytics#factor-performance.