Playbook › Playbook
Incremental gross margin below the average locks in future margin decay — it is an identity, not a forecast
Claim
A blended gross margin that has stopped falling is one of the easiest signals to misread, because it can be the average of a repair that is nearly finished and a mix shift that is just getting started. The two partly cancel, the group number looks flat, and both the bull and the bear read their own case into it.
The incremental margin resolves it without a forecast:
incremental gross margin = ΔGross profit / ΔRevenue
While incremental < average, the average must fall. That is an identity — a weighted average pulled by a lower-margin marginal dollar can only decline. No management commentary, guidance, or "stabilisation" narrative changes it.
Evidence — PDD, Q1-2026 vs Q1-2025
| Q1-2025 | Q1-2026 | Δ | |
|---|---|---|---|
| Revenue (RMB M) | 95,672 | 106,229 | +10,557 |
| Gross profit (RMB M) | 54,726 | 59,336 | +4,610 |
| Gross margin | 57.20% | 55.86% | −134bp |
Incremental gross margin = 4,610 / 10,557 = 43.68%, against a 55.86% average.
The implied annual drag at ~10.6% revenue growth is 10.6% × 12.18pp / 1.106 ≈ −117bp/yr, which
reproduces the observed −134bp. The incremental had improved enormously — it was 6.9% a year
earlier and 26.9% across the two-year stack — but it had not crossed the average, so the
average kept falling.
The decomposition that must accompany it
The raw identity tells you the direction. It does not tell you whether the cause is durable. Split the change by holding segment margins constant:
| Component (PDD, Q1-25 → Q1-26) | Effect |
|---|---|
| Mix drag — low-margin segment gaining revenue share | −262bp |
| Within-segment improvement — the high-cost segment repairing its own unit economics | +128bp |
| Net | −134bp |
These two are not symmetric and must never be netted without comment:
- The within-segment repair was Temu moving merchandise cost off its books after the de-minimis change. It is a one-time conversion with a finite endpoint — once the cost is off, it cannot come off again.
- The mix drag is a compounding share shift with no endpoint, and management had just committed RMB 100B to accelerating it.
So the flattering half expires and the damaging half compounds. A projection that freezes both segment margins at their improved levels — i.e. grants the bull case entirely — still produced 490–660bp of gross-margin decline over eight quarters.
Why it beats the usual tests
The blended margin was flat at 55.5–56.7% for five quarters, which read as stabilisation. It was not: Q1 was that company's seasonal gross-margin high, so the "flat" range was comparing a seasonal high against a seasonal low, and the like-for-like Q1-vs-Q1 change was −134bp. The incremental-margin identity is seasonality-robust in a way the level is not, because both terms come from the same pair of periods.
It also catches the case this note was born from: a group margin that hides a decaying profit pool. At PDD the high-margin line was 47% of revenue but 71–80% of gross profit, and it was growing +2.5% while the low-margin line grew +19.9%. The company was swapping ~90%-margin revenue for ~25%-margin revenue one point of mix at a time, and the group margin barely moved.
How to apply
- Compute it on like-for-like periods (Q vs same Q, or FY vs FY) — never sequentially on a seasonal business.
- Compare to the average. Below → the average must fall. Above → it must rise. Crossing the average is the actual inflection to wait for, and it is a cleaner upgrade trigger than any margin level.
- Decompose into mix versus within-segment before concluding, and state which half is finite. Netting them without that distinction is how a temporary repair gets mistaken for a structural fix.
- Watch the gross-profit-growth-to-revenue-growth ratio as the running check: PDD's was 0.76 in the quarter and 0.44 on the two-year stack. Below 1.0 means the average is falling; the trend in the ratio tells you whether the drag is accelerating.
- Check whether the profit pool and the revenue pool are the same thing. If one segment is a minority of revenue but a large majority of gross profit, its growth rate matters far more than the group's, and the group number will hide it.
What would falsify this
Nothing falsifies the identity itself. What can falsify the inference is a mix shift that reverses — the low-margin segment losing share again — or a within-segment improvement large enough to push the incremental above the average. Both are observable in the next print, which is why this pattern comes with a specific line item to watch rather than a general worry.
Related
[[pattern-margin-intact-while-volume-falls-defers-the-damage]] (this is the same failure mode with the variables swapped — there a flat margin hid lost volume, here a flat group margin hides a decaying profit pool) · [[pitfall-growth-ratio-against-collapsing-base]] (the companion error — reading the growth ratio instead of the level) · [[pattern-seasonal-ratio-erasure-detects-structural-decline]] · [[pitfall-logo-retention-masks-dollar-retention]] (same shape in retention metrics)
History
- 2026-08-19 — derived during the Phase 3 adjudication of the first PDD analysis, where two analysts reached opposite conclusions from the same correct margin figures. The identity settled the dispute where neither the level nor the trend could.