Variant Perception

Where we sit off the market

The one-line variant view. The market is paying $3.17 for a stub-equity it has written off as a sub-scale, founder-captured, cash-burning China ADR with a forced related-party convertible on top — but the same three institutional cohorts that know the business best (founder, Trustar, Hillhouse) wrote checks at $4.63 inside the last sixty days, and the only published FY2026 consensus EPS quietly assumes the FY2025 non-GAAP turn extends through this year — an assumption Q1 FY2026 has already broken. Our variant view sits on top of those two specific gaps: the smart-money mark is a real floor that the tape has not priced, and the FY2026 non-GAAP operating margin path the Street's lone EPS print implies is 3-5pp too high. Both close in the next 6-9 months, and they close in opposite directions — which is why "no edge" is the wrong read, and "lean long, sized after the Q2 print" is the right one.

The variant strength is medium, not high, because the disagreement is between two cohorts of believers — the smart-money cohort (founder + Trustar + Hillhouse, all underwriting at $4.63) and the tape (which has marked the stock 46% off in three months) — and the resolution path runs through one earnings print and one 13G/A. Consensus clarity is low-medium on purpose: only one analyst publishes the FY2026 EPS line, no major broker covers the name, independent contributor sentiment has rotated to "Hold/Skeptical/Convoluted" (per the Web Research tab). When the market view is thin, "the market believes X" needs to be earned by signals, not asserted by vibe — and the signals we have all point the same way.

What the market actually believes — the consensus map

The right way to read a thin-consensus name is to triangulate. Six observable signals tell you what the marginal buyer of YSG at $3.17 is underwriting. Five out of six are bearish-to-neutral; the sixth — the lone published price target — is the outlier the variant view leans against.

Three things to read out of this map. First, the lone published FY2026 EPS line (Row 3) is the one piece of explicit Street math available, and it implies the FY2025 non-GAAP turn extends — which is exactly the assumption Q1 FY2026 already contradicted. Second, the price action (Rows 1 and 2) and the smart-money rotation (Row 5) tell a different consensus story — the marginal seller is treating the convertible as forced and the FY2025 turn as bookkeeping. The variant view has to land between those two: the Street's lone EPS print is too high; the tape's read of the $4.63 mark as forced is too low. Third, the illiquidity overhang (Row 6) is real and limits the variant view's upside even if the disagreement is correct — this is not the kind of disagreement that resolves to a multiple re-rating; it resolves to a basis-points-of-AUM trade, not a sizing call.

Where we disagree with the market — the ledger

Three measurable disagreements survive the five tests (consensus check; evidence check; materiality; resolution path; what would break it). They are ranked by expected value to the PM: the sharpest is the smart-money mark because it carries a 46% gap to the floor a known long-only China specialist (Hillhouse) just paid for the same equity. The margin variant is second because it closes in 62 days on a hard date. The moat / gross-margin variant is third — directionally important but resolves on the FY2026 20-F nine months out.

Variant 1 — The $4.63 strike is a floor the tape has not priced

Consensus reads the March 11, 2026 Trustar/Huang convertible as a forced related-party self-deal: balance sheet drained from ¥2.59B (FY22) to ¥1.05B (FY25), CEO on both sides of the table, 4% IRR put floor only the founder vehicle gets, and a price tape that traded the stock down -46% over the period spanning the announcement, the FY2025 20-F, and the May 21, 2026 first-tranche close. The frame collapses on one fact: Hillhouse, the largest outside holder at 13.8% Class A and a multi-year long-only China specialist, formally objected to the structure in writing and then forced inclusion on equal terms [1], not the other way around. A holder fighting to participate at $4.63 is a holder marking the equity at $4.63. The People and Verdict tabs make this point in passing; the Variant Perception read is that the smart-money cohort agreement is the most credible third-party endorsement in the public record, and the market has not priced it because the tape conflates "founder needs the money" with "everyone in the deal got the price they wanted."

What would the market have to concede if we are right? That a name a global China-strategy long-only is adding to at $4.63 cannot simultaneously be a name on a permanent $3.17 stub-equity discount. The cleanest disconfirming signal is the next 13G/A — Hillhouse trimming below 10% beneficial ownership flips the smart-money read to forced inclusion; continued accumulation at or above $4.63 confirms it.

Variant 2 — Consensus EPS is 3-5 percentage points of margin too high for FY2026

The lone published FY2026 EPS at RMB1.56 (the data file in data/estimates/analyst_estimates.json from yfinance) is roughly flat to FY2025 — implying the analyst extends the FY2025 -2.0% non-GAAP operating margin into FY2026 even though Q1 FY2026 already printed -8.3%. The CFO did not "guide down" — he said management would "selectively deploy resources" on Douyin in Q2, i.e. spend not retreat [2]. The arithmetic per the Catalysts tab is: even a heroic Q2-Q4 average of -1% non-GAAP OM clears full-year -3%; a more realistic average closer to Q1's run rate puts FY2026 at -4 to -5%. That is a 3-5pp gap to the implicit consensus line, or roughly RMB160-260M of non-GAAP EBIT the Street has not yet acknowledged is missing. The down-revision pressure is already in the EPS-revisions data — 1 downward revision in the last 30 days on the FY2026 line — but the magnitude of the down-revision has not happened yet.

This is the narrowest and most monetizable disagreement of the three because it resolves on a fixed date — Aug 20, 2026, in 62 days — and the price-reaction base rate the Catalysts tab compiled (13.3% average absolute one-day move on earnings, with downside skew on washed-out positioning) frames the magnitude. What would the market have to concede if we are right? That the FY2025 non-GAAP profit was a budget-cycle phenomenon, not a structural rebase — which would close the gap between consensus and the bear read of the Forensics tab.

Variant 3 — The 78%+ gross margin is mix, not moat

A consensus analyst looking at FY2025 sees 78.2% gross margin (peak 80.2% Q1 FY2026), highest in the China-listed beauty peer set, and reads it as evidence of pricing power earning a Proya-style rerate from 0.24x sales toward 2x sales. The Moat tab walked through the disconfirming evidence row by row: both reportable segments lost money at the operating line in the best year of the cycle (Color -¥59M, Skincare -¥31M), Eve Lom has been goodwill-impaired twice for ¥757M cumulative, and the segment-level operating loss persists at a higher gross margin than Proya — which converts 73.3% gross margin to 17.6% operating margin. High gross margin alongside negative segment OI is the fingerprint of mix shift, not a brand moat. Mix shift is mechanical; it stops when skincare crosses ~65-70% of revenue. A brand moat extracts pricing power that flows past S&M; this brand portfolio does not yet do that.

This is the slowest-resolving disagreement of the three — the cleanest test is the FY2026 20-F skincare segment operating profit, in roughly nine months — but it is the variant view that most matters to the multi-year underwriting because it is the entire bridge between EV/Sales 0.24x and EV/Sales 2x. What would the market have to concede if we are right? That the FY2025 gross-margin record is the upper bound, not the leading edge, and the next leg of the multiple has to come from the operating line, not the mix line.

The evidence layer — what each piece does to the variant view

A PM auditing this page should be able to pressure-test the strongest five pieces of evidence. These are the items that actually move probability — not generic facts about the company. Each one carries a fragility column: what would make the evidence misleading, and what condition would prove it.

The two strongest pieces — the $4.63 strike with Hillhouse's forced inclusion and the Q1 FY2026 margin reversal — are the load-bearing items. Both are 100% from the primary record, both are disclosed within the last 90 days, and both are concrete enough that a PM can put them on a watchlist by close of business. The weakest is Row 6 — the segment-OI test — because it permits two equally plausible reads (FY26 segment OI positive because last-year trajectory continues, or skincare loss widens because Douyin S&M lands on the skincare brands disproportionately). Resolution requires the FY2026 20-F.

How this gets resolved — observable signals on a watchlist

The disagreements are useful only if they close on observable evidence inside the PM's underwriting window. Every signal below has a specific source, a clean validation and refutation condition, and a clock. None of these are "execution improves" or "time will tell."

The decisive ordering is signals 1 → 2 → 5 — Q2 FY2026 non-GAAP OM in 62 days, then the next Hillhouse 13G/A, then skincare segment OI in the FY2026 20-F. Signals 3 and 4 are confirmatory layers; signal 6 is the tail-event check that breaks the moat thesis if it triggers. A PM only needs to keep signals 1 and 2 on a working watchlist to update the variant view inside the next 90 days.

Red team — what would make us wrong

The variant view is not high-conviction. The honest disconfirming arguments are below, written by the side that wants to kill the thesis, not protect it.

1. The $4.63 mark is not a floor; it is a coupon trade. Hillhouse may be participating at $4.63 because the structure is a 1.5% coupon plus a 4% IRR put floor — i.e., the worst case for the convertible holder is a 4% IRR, with conversion optionality on top. That is not a vote on $4.63 as fair value of the common equity; it is a vote on $4.63 as the strike price of an instrument that returns 4% IRR regardless. A PM long the common at $3.17 does not get the 4% IRR floor — only the dilution. If we are wrong, the floor evaporates on full conversion and the $4.63 mark becomes a ceiling on the next 12 months instead of a floor.

2. Consensus EPS is thin enough to be a coin flip, not a tradeable mispricing. Variant 2 leans on a single published FY2026 EPS line that almost no buyer actually relies on for pricing. The marginal seller at $3.17 is not selling because their model says +0% OM — they are selling because they cannot underwrite the related-party governance structure at any margin. A Q2 beat does not solve that; the stock can stay at $3.17 even if our margin call is right, because the multiple is structurally capped by the controlled-company NYSE exemption and the dual-class wedge. The Forensics tab said this directly: this is a position-sizing limiter and a valuation-haircut item, not a multiple expansion catalyst.

3. Mix shift may be enough — Proya is the wrong comp. Variant 3 leans on the Proya 17.6% OM as the rerate target. Proya has Tmall category leadership (#1 face cream, #1 mask) that Yatsen does not have and cannot manufacture. The variant view confuses "same business model" with "same outcome." If Yatsen tops out at the Shanghai Jahwa economics (62.6% gross margin / 2.7% OM) — also a domestic, multi-brand, profitable peer — then the FY2025 turn IS the structural answer and there is no 22pp gap to close. Mix shift to break-even, not to Proya-grade. Stock is fairly priced.

4. The platform-pricing mechanic was the right call all along. Failure Mode #1 in the Long-Term Thesis tab was correctly named and is now playing out. Q1 FY2026 is not noise — it is the structural answer to "can a sub-scale brand earn operating margin against a platform with visible pricing power?" The variant view that says "the consensus is too high" ignores that the consensus is already too high because it has not yet been re-cut for the platform-pricing reality, and the post-Q2 print may take the consensus down to where the bear case lives — not where the variant view lives. The 3-5pp gap may be the wrong direction.

5. Liquidity overhang dominates everything else. The Technicals tab put the 20-day ADV at ~$0.7M. A 1% position takes 25 trading days to exit at 20% participation. Any variant view that requires the multiple to rerate is structurally hostage to a market-clearing event the float cannot absorb — convertible second-tranche close, FY2026 20-F print, an actual buy program. Without one of those, the variant view is correct but unmonetizable.

Items 1 and 4 are the most serious. Item 1 is the load-bearing argument against Variant 1: the $4.63 is a strike, not a fair-value mark, and the convertible's 4% IRR put floor is exactly what makes Hillhouse's participation not a vote on the common. Item 4 is the load-bearing argument against Variant 2: the down-revision may go past us, not stop where we sit. The variant view survives both objections only if the smart-money cohort actually adds at $4.63 in the open market, not just inside the convertible — which is the signal to watch.

The single watchpoint

The honest synthesis: this is a "small but real" variant view, not a high-conviction call. The market has marked the stock down for the right operating reasons (Q1 FY2026 margin reversal, FY2025 turn quality, governance friction) and the wrong valuation reason (the smart-money mark sits 46% above spot and is corroborated by a known long-only China specialist that actively demanded inclusion). The right PM action is to keep the position small until Q2 FY2026 resolves the lone-analyst EPS gap, then size on the Hillhouse 13G/A path. The wrong action is to underwrite the multiple to Proya. The wrongest action is to read the $4.63 strike as evidence of fair value without separately watching the $3.17 to $4.63 closing path.

References

1. Yatsen Holding Limited - FY2025 Annual Report (Form 20-F), Item 3.D Risk Factors - March 11, 2026 Trustar/Huang Note Purchase Agreement and significant-shareholder objection - p.40
2. Yatsen Holding Limited - Q1 FY2026 Earnings Release, Operating Expenses and Douyin traffic-acquisition cost commentary - p.2