In December 2025 the SEC rescinded the 2003 Global Research Analyst Settlement — the rulebook written after the dot-com bust to wall off equity research from investment banking. Within weeks, every underwriter of SpaceX's IPO published a buy. The obvious question is whether the old conflict has come back.
The classic study is Michaely & Womack (1999) [1]: buy recommendations from a company's own underwriters underperformed buys from independent analysts. That paper used 1990–91 IPOs. We rebuilt it on 3,581 firm-attributed ratings across 259 US IPOs from 2024–2026, and found something the original could not show — the same bank, rating the same kind of company, says "buy" 15.7 percentage points more often when it underwrote the deal.
How post-IPO ratings actually get issued
The process is heavily choreographed, and the choreography is what makes it measurable.
Nothing is published before the IPO. During the SEC quiet period no research on the issuer may be distributed. There is no such thing as a pre-IPO analyst rating, so any test of analyst accuracy necessarily starts after listing.
The underwriters go first, together, on a scheduled day. Under FINRA Rule 2241 (which replaced NASD 2711) banks that underwrote the offering must wait before publishing. That window shrank from 25–40 days to 10 days, and emerging-growth companies — most modern IPOs — are exempt entirely under the JOBS Act. The result is a synchronised burst. In our data underwriter initiations cluster at day 25 (25th percentile day 25, median day 25, 75th percentile day 26). Quantinuum drew 14 initiations on a single day.
Independents arrive much later, and individually. No quiet-period obligation, but also no client relationship, no management access and no fee to justify the coverage cost. Median independent initiation: day 217 — roughly seven months after the underwriters have spoken.
An initiation is not automatically a buy. Of 1,382 initiations, 1,150 (83%) are buys, 217 (16%) are holds and 12 (1%) are sells. A hold at initiation is, in sell-side dialect, usually a polite negative.
There is no standard rating vocabulary. Our ratings use 46 distinct grade strings. Morgan Stanley says Overweight / Equal-Weight / Underweight; RBC says Outperform / Sector Perform / Underperform; Weiss publishes letter grades like "Sell (D-)". Others use Speculative Buy, Accumulate, Top Pick, Peer Perform, In-Line, Sector Weight, Cautious. All of it collapses to three stances — buy, hold, sell. Note that Overweight contains Underweight's root, so a careless text match silently miscounts an entire house's output.
Method, and what we do with holds
- The unit is one rating, not one IPO. Returns run from each rating's publication date forward 3 or 6 months. A call on day 4 is measured day 4 to day 4+H; a call on day 200 from day 200. No price information from before the call is used.
- A rating is "underwriter" if the issuing firm sat in that deal's syndicate, and "independent" otherwise. The same bank is an underwriter on some deals and an outsider on others — the basis of our cleanest test.
- Holds do not count as buy initiations. The buy sample requires an initiation and a buy rating. The 217 hold initiations and 12 sell initiations are analysed separately, never folded in. Counting holds would flatter whichever group issues more of them — precisely the bias under investigation.
- The benchmark is matched on age and size. For each rating we take the stock's age since IPO and its market-cap bucket, then average the return of every other IPO in that same size bucket over that same age window — from each peer's own day A to its own day A+H. The rated stock is excluded from its own benchmark.
That last point is the correction that mattered most, and it reversed our own first result. IPOs bleed as they season — mean return is −4.5% in month 1 and negative in 13 of the first 15 months. Underwriters rate at day 25 and independents at day 217. Benchmarking both against "the average IPO on that calendar date" charges the day-25 call for a decay phase the day-217 call already survived.
Finding 1: the same bank is far more bullish when it underwrote
Twenty-six firms rate IPOs in both roles. Comparing each bank against itself removes house style entirely.
Share of ratings that are "buy", same bank, by role
| Bank | As underwriter | As independent | Difference |
|---|---|---|---|
| Barclays | 95% | 38% | +57 pp |
| KeyBanc | 93% | 50% | +43 pp |
| Truist | 100% | 65% | +35 pp |
| Robert W. Baird | 91% | 62% | +29 pp |
| Jefferies | 83% | 57% | +26 pp |
| Guggenheim | 97% | 76% | +21 pp |
| Cantor Fitzgerald | 100% | 80% | +20 pp |
| Citigroup | 79% | 87% | −8 pp |
| Roth Capital | 88% | 100% | −12 pp |
| UBS | 60% | 80% | −20 pp |
22 of 26 banks are more bullish when they underwrote. Mean difference +15.7 pp. Sign test p ≈ 0.0002.
This is the most statistically robust result in the study, and it is a direct measure of the conflict rather than an inference from returns. It also rules out the obvious alternative explanation — that structurally optimistic houses simply win more mandates — because each bank is its own control.
Finding 2: nobody says sell about a deal they underwrote
Ranked by share of buy ratings on underwriter-role calls (minimum 20 rated calls):
| Firm | n | Buy | Hold | Sell |
|---|---|---|---|---|
| Cantor Fitzgerald | 46 | 100% | 0% | 0% |
| BTIG | 22 | 100% | 0% | 0% |
| Needham | 59 | 98% | 2% | 0% |
| Guggenheim | 38 | 97% | 3% | 0% |
| William Blair | 35 | 97% | 3% | 0% |
| Evercore ISI | 50 | 96% | 4% | 0% |
| Raymond James | 71 | 96% | 4% | 0% |
| Barclays | 80 | 95% | 4% | 1% |
| Piper Sandler | 97 | 84% | 16% | 0% |
| RBC Capital | 83 | 81% | 19% | 0% |
| Wells Fargo | 77 | 81% | 17% | 3% |
No firm covering its own deals falls below 81% buy, and the highest sell rate on the list is 3%. In their independent-role calls the same population ranges from Bernstein at 48% buy / 19% sell up to 100%. The dispersion exists — it just disappears on deals the bank was paid to sell.
Finding 3: both groups beat comparable IPOs, independents by more
Cumulative return from the buy-initiation date, against the age- and size-matched benchmark:
Cumulative return after a buy initiation vs matched IPO benchmark
Sample thins with horizon — 718 underwriter and 380 independent calls at month 1, falling to 206 and 66 by month 12. Treat the right-hand side as indicative.
At the horizons we can measure properly:
| Buy initiations | Underwriter | Independent |
|---|---|---|
| n (3M / 6M) | 535 / 446 | 304 / 244 |
| Matched benchmark, 3M | −4.6% | −5.8% |
| Excess return, 3M | +4.0 pp | +8.9 pp |
| Matched benchmark, 6M | −11.7% | −11.9% |
| Excess return, 6M | +9.7 pp | +16.2 pp |
| Spread (independent − underwriter) | +4.9 pp (p ≈ 0.10) | +6.5 pp (p ≈ 0.26) |
Both groups beat a same-age, same-size IPO basket. Independents lead by 5–7 points — the direction Michaely & Womack found — but the gap is not statistically significant at this sample size. On realized returns we can show that underwriter buys are more optimistic, not reliably that they are more wrong.
Finding 4: the rating itself carries almost no information
If buy/hold/sell means anything, buys should beat sells. Excess return versus the matched benchmark, sample size in brackets:
| Rating (3M) | Underwriter | Independent | Total |
|---|---|---|---|
| Buy | +8.5 pp (1,097) | +8.4 pp (538) | +8.5 pp (1,635) |
| Hold | +9.0 pp (230) | +9.7 pp (147) | +9.3 pp (377) |
| Sell | +17.6 pp (12) | +13.6 pp (24) | +15.0 pp (36) |
Holds do as well as buys. Sells do best, but on 36 observations that ordering is a curiosity, not a result. What is reliable is the absence of the expected gradient across 1,600+ buys and 300+ holds.
The consistent finding is the column, not the row: every rated group beats the benchmark by 8–15 points whatever the call said. But note what the benchmark is — all IPOs, rated and unrated. Tested directly, comparing covered against never-covered names within each size bucket, coverage is worth +9.1 pp at day 90 and +32.4 pp by day 360. The effect is near zero at day 30 and grows with age, which is the signature of selection, not foresight: analysts choose the names worth covering, and by day 360 "has coverage" is close to a synonym for "has worked."
Caveats
- 3,581 ratings over 259 IPOs, 2024–2026. Only the buy-rate difference clears conventional significance; the return spreads are directional.
- One market regime — a favourable tape dominated by large AI and infrastructure listings.
- The benchmark controls for age and size but not market regime; peers contribute their day-A returns from their own calendar dates. Excess return means "beat comparable IPOs at the same stage of life", not "beat the market".
- Coverage is chosen, not assigned, so the covered-vs-uncovered gap mixes skill with selection.
- Sources: FMP historical ratings (action and grade parsed from headlines) plus a daily MarketBeat scrape. About 2% of rows still lack a classifiable action.
The bottom line
The Global Settlement's repeal removed a rule written for exactly this behaviour, and the behaviour is measurable in the first two years of data: banks are dramatically more positive about the deals they were paid to underwrite. Whether that optimism costs you money is, on this sample, not yet provable. But you no longer have to take the conflict on faith — you can count it.
References
- Michaely, R. & Womack, K. (1999). Conflict of Interest and the Credibility of Underwriter Analyst Recommendations. Review of Financial Studies. https://academic.oup.com/rfs/article-abstract/12/4/653/1578416
- Lin, H. & McNichols, M. (1998). Underwriting relationships, analysts' earnings forecasts and investment recommendations. Journal of Accounting and Economics. https://www.sciencedirect.com/science/article/abs/pii/S0165410198000175
- Dechow, P., Hutton, A. & Sloan, R. (2000). The relation between analysts' forecasts of long-term earnings growth and stock price performance following equity offerings. Contemporary Accounting Research. https://onlinelibrary.wiley.com/doi/10.1506/13KT-BW1P-CTM4-8VKA
- Bradshaw, M., Brown, L. & Huang, K. (2013). Do sell-side analysts exhibit differential target price forecasting ability? Review of Accounting Studies. https://link.springer.com/article/10.1007/s11142-012-9216-5
- Bonini, S., Zanetti, L., Bianchini, R. & Salvi, A. (2010). Target Price Accuracy in Equity Research. Journal of Business Finance & Accounting. https://onlinelibrary.wiley.com/doi/10.1111/j.1468-5957.2010.02209.x
- Hong, H. & Kacperczyk, M. (2010). Competition and Bias. Quarterly Journal of Economics. https://w4.stern.nyu.edu/sternfin/mkacperc/public_html/bias.pdf