The Anatomy of a Compounder: How 7 Great Companies Actually Created Shareholder Returns.
Seven public-company five-year windows decomposed into the five drivers of equity return — volume, price/mix, margin, multiple, and leverage. What actually creates durable shareholder returns?
The Anatomy of a Compounder
How 7 Great Companies Actually Created Shareholder Returns — a decomposition of NVIDIA, Apple, Costco, AutoZone, TransDigm, Constellation, and Eli Lilly into the five drivers of equity return
Every great stock return has a story hiding underneath the headline number. This article tears apart seven famous compounders to answer a deceptively simple question: where did the return actually come from? The answer is almost never what it looks like from the outside.
Before the math: why returns are almost always misread
Here is the simplest way to think about equity returns. Every stock that goes up over time does so for one of five reasons — and only five:
- selling more units (volume growth)
- charging higher prices (price & mix)
- earning more margin on each dollar sold
- getting a higher valuation multiple on those earnings
- changing the capital structure (debt, buybacks, dividends)
- luck, timing & macro — often hiding inside the above four
That's it. Every compounder, every 10-bagger, every "why did this stock go up so much?" story — it's some combination of those five things. The trouble is that headline returns hide which lever did the work.
A 25% annual return from "the market suddenly paid twice as much for earnings" is structurally different from a 25% return from "the company sold five times as many units." The first is fragile. The second tends to repeat.
Why decomposition matters
- Prevents false attribution — separates what management did from what the market gave them
- Identifies repeatability — operational returns recur; multiple re-ratings are one-shot
- Improves forecasting — you can't project what you haven't measured
- Reveals hidden risk — a 3× driven by leverage ≠ a 3× driven by volume
- Sharpens pattern recognition — you start to see archetypes, not just individual cases
- Separates luck from skill — the hardest and most important question in investing
The hierarchy — from gross return to operating drivers
The framework used here is Reinard's Private Equity Value Creation Analysis, restricted to single-company nodes. Think of it as a tree that starts at "equity return" and branches into every independent driver that could have produced it.
Three branches grow from the trunk. The operating branch captures what the business actually did: unit growth, pricing, and margin. The multiple branch captures what the market decided the business was worth. The leverage branch captures everything that happened to the capital structure.
The key property of the logarithmic version of this model is that every leaf's contribution is independent of every other leaf. There are no cross-terms to allocate by hand. The percentages sum cleanly to 100%.
The operating tree in full
The leverage layer — what converts enterprise returns to equity returns
The leverage effect is the bridge between "how much did the enterprise value grow" and "how much did the equity holder earn." A company that grows enterprise value 2× but carries 4× net debt will give its equity holders a return very different from 2× — it could be 5× if things go well, or negative if things go badly.
There are two distinct sub-drivers inside the leverage effect. The first is the debt paydown and recap component — changes in the absolute level of net debt relative to enterprise value. TransDigm manages this one deliberately through periodic leveraged recapitalizations. The second is buyback compression — reducing the share count means each remaining share owns a larger fraction of the equity base.
Share-count compression is mathematically equivalent to leverage on equity. It shows up nowhere on the income statement and is thus consistently underestimated by investors focused on operational metrics.
EBITDA growth: the most misread branch
EBITDA growth is itself a product of two independent things: revenue growth and margin change. These must be measured separately. A company that doubles revenue but halves its margin delivers the same EBITDA, but the return attribution is completely different — the first is a volume and price story, the second is an operational efficiency story.
Within margin, the model breaks into three sub-leaves. Gross margin captures what happens at the COGS level — input cost dynamics, product mix, pricing power at the unit economics layer. SG&A leverage captures operating leverage — revenue scaling faster than the fixed cost base. D&A scaling bridges EBITDA to operating income and captures capital intensity.
Most analysts conflate "margin expansion" with "gross margin expansion." In practice, the largest contributor in many compounders is SG&A leverage — the business scaling revenue faster than its headcount and overhead. NVIDIA is the clearest example: gross margin expanded from ~62% to ~75%, but SG&A leverage contributed more than 3× as much to the equity return.
Multiple expansion: the most dangerous driver
Multiple expansion is dangerous not because it is rare — it is quite common in bull markets — but because it is routinely misattributed to management skill or business quality improvement. It is often neither.
The model distinguishes between two types. Market multiple expansion is sector-wide re-rating: when the S&P 500 moves from 18× to 24× earnings over a bull cycle, almost every company in it participates. This has nothing to do with individual management teams. Intrinsic multiple expansion is company-specific: when the market reprices a specific business because its perceived quality, durability, or growth runway has genuinely changed. Eli Lilly's GLP-1 re-rating is the clearest example in this set.
The most important question to ask about any historical return that includes significant multiple expansion: was this market-wide or company-specific? Market-wide re-rating is mean-reverting and says nothing about future returns. Company-specific re-rating can persist if the underlying competitive position change that triggered it is real.
Revenue growth (Volume + Price = 31.1% + 10.2% = 41.3% of total return) translates to 16.8 pp of the 40.6% pa CAGR. Multiple expansion (45.4%) translates to 18.4 pp. The two together account for ~35 pp of the 40.6 pp annualized return; margin and leverage make up the rest. Use this same map to read every bar in the explorer below.
Return attribution explorer
Click any company tab to see its full five-driver decomposition
NVIDIA
Every lever on"The cleanest 'everything firing at once' decomposition in modern public markets."
Revenue grew roughly 12× — half from datacenter GPU shipment volume on the A100→H100→B100 cycle, half from ASP and mix shift to datacenter SKUs at $25–40k versus prior-generation $5–10k. What makes NVIDIA unusual is that every major return lever moved positively at the same time — volume, price, gross margin, operating leverage, and multiple expansion. That combination is extraordinarily rare.
Operating performance alone accounts for roughly 104% of the total return in log terms. The multiple expansion adds ~19%, and dilution and exit-multiple drift give ~23% back. The hidden detail: the majority of margin expansion is SG&A leverage (+21.8%), not gross margin (+6.1%). Revenue scaled 12× into a roughly fixed cost base.
| Component | Detail | Value / Δ | % of return |
|---|---|---|---|
| Revenue growth | $10.9B → ~$130.5B | ~12.0× | — |
| · Volume / units | A100→H100→B100 shipment cycle | ~3.5× | +40.2% |
| · Price / mix | H100 ASPs ~$25–40k vs prior-gen $5–10k | ~3.4× | +39.5% |
| EBITDA margin Δ | ~30% → ~63% | +~33 pts | — |
| · Gross margin | ~62% → ~75% | +~13 pts | +6.1% |
| · SG&A leverage | OM ~26%→~62%; fixed base scaled by 12× revenue | +~23 pts | +21.8% |
| · D&A leverage | D&A grew slightly faster than capacity | −~3 pts | −4.1% |
| · Multiple Δ | ~22× → ~40× EV/EBITDA | +~18 turns | +19.2% |
| · Leverage Δ | Net cash both periods; minor buyback | ~0 | ~0% |
| · Residual / dilution | SBC dilution + multiple drift vs stated peak | — | −22.7% |
| Total equity return · Jan 2020 → Jan 2025 | ~22.5× | 100% (~86% pa) | |
NVIDIA is the only case in this set where operating performance alone exceeds 100% of the equity return. All five levers pulled in the same direction simultaneously. That is extraordinarily rare and should not be used as a template for future expectations.
Partially repeatable. The SG&A leverage mechanism is structural — but only if revenue keeps growing. A plateau in datacenter demand would compress margins quickly. AI capex continuation and ASP holding are non-trivial assumptions for the next five years.
Apple
Financial engineering"Financial engineering amplified a stable business — ~70% of the return came from capital structure and re-rating, not operations."
Almost the opposite shape from NVIDIA. Revenue 1.44×, margin essentially flat. The bulk of the equity return came from two financial levers — the multiple roughly doubled (12× to 22×), and the share count shrank ~21% from sustained buybacks. Apple's operating story was primarily about price and mix, not volume. iPhone units were roughly flat; the lift came from Services growing its share of the mix.
| Component | Detail | Value / Δ | % of return |
|---|---|---|---|
| Revenue growth | $265.6B → $383.3B | ~1.44× | — |
| · Volume / units | iPhone units flat; Services & Wearables grew | ~1.05× | +4.1% |
| · Price / mix | iPhone ASP up; Services mix shift | ~1.37× | +26.7% |
| EBITDA margin Δ | ~32% → ~33% | +~1 pt | — |
| · Gross margin | ~38% → ~44% | +~6 pts | +12.5% |
| · SG&A leverage | SG&A grew slightly faster (negative leverage) | −~3 pts | −3.5% |
| · D&A leverage | D&A drag offsets most of the OM lift | −~2 pts | −6.4% |
| · Multiple Δ | ~12× → ~22× EV/EBITDA | +~10 turns | +51.4% |
| · Leverage Δ | Share count −~21%; gross debt rose | ~1.27× | +20.0% |
| · Residual / dilution | Approximation noise + remaining bridges | — | −4.8% |
| Total equity return · Sep 2018 → Sep 2023 | ~3.25× | 100% (~26.6% pa) | |
Multiple + leverage = 71% of the return. The operating story was real but modest. The question for the next five years is not whether Apple's business keeps working — it's whether the multiple can double again from 22×. That's a very different question.
Low. The multiple can't double again from 22×. Buybacks can continue, but the equity-base compression is already baked in. The business is excellent — the specific return driver is largely exhausted.
Costco
Market re-rating"Ordinary operations, extraordinary re-rating — strip the 62% multiple contribution and the return drops to 11% pa."
Costco is the purest re-rating story in this set. Revenue grew 1.67×. Margin moved a few tenths of a percent. The multiple moved from ~22× to ~50×. That ~28-turn re-rating explains roughly 62% of the total return. Strip it out and the equity return is ordinary: about 1.7× over five years, or 11% annually. Costco stopped being valued like a retailer and started being valued like infrastructure.
| Component | Detail | Value / Δ | % of return |
|---|---|---|---|
| Revenue growth | $152.7B → ~$254.5B | ~1.67× | — |
| · Volume / units | Membership growth + warehouse openings | ~1.30× | +19.8% |
| · Price / mix | Inflation pass-through + ancillary mix gains | ~1.28× | +18.7% |
| EBITDA margin Δ | ~4.0% → ~4.3% | +~0.3 pts | — |
| · Gross margin | ~13.1% → ~12.6% (slight compression) | −~0.5 pts | −2.9% |
| · SG&A leverage | OM ~3.4% → ~3.7%; modest SG&A leverage | +~0.8 pts | +9.3% |
| · Multiple Δ | ~22× → ~50× | +~28 turns | +62.1% |
| · Leverage Δ | Net cash; one large special dividend | ~0 | ~0% |
| · Residual / dilution | Multiple drift + SBC dilution + special-div timing | — | −7.0% |
| Total equity return · Sep 2019 → Sep 2024 incl. special dividend | ~3.75× | 100% (~30.2% pa) | |
A mediocre operational story wrapped in an extraordinary re-rating. The business will keep executing — but the re-rating has already happened. The next five years run from a 50× baseline, not 22×.
Low. A 50× multiple on a near-zero-margin retail business is already pricing in extreme durability. Can't re-rate from 50× in the same way it re-rated from 22×.
AutoZone
Leverage without an LBO"Buyback compression explains ~32% of the return — more than multiple expansion. Operating margin moved less than a point."
AutoZone engineered an LBO without leaving public markets. The operating story is genuinely modest. The share count went from 24.4M to 17.0M — a ~30% reduction — with structural net leverage maintained throughout. Buyback compression (~32%) accounts for as much return as all revenue growth combined, and more than multiple expansion (~30%). Operating margin moved less than one percentage point across the entire window.
| Component | Detail | Value / Δ | % of return |
|---|---|---|---|
| Revenue growth | $11.86B → $18.49B | ~1.56× | — |
| · Volume / units | Modest store growth; commercial expansion | ~1.20× | +16.1% |
| · Price / mix | Pricing held; commercial mix higher | ~1.30× | +23.2% |
| EBITDA margin Δ | ~22% → ~23% | +~1 pt | — |
| · Gross margin | ~53.7% → ~53.0% (slight compression) | −~0.7 pts | −1.2% |
| · SG&A leverage | OM ~19.4% → ~20.1%; modest SG&A leverage | +~1.4 pts | +4.3% |
| · Multiple Δ | ~10× → ~14× | +~4 turns | +29.7% |
| · Leverage Δ | Share count −~30%; persistent net debt | ~1.43× | +31.6% |
| · Residual / dilution | Approx. noise + net-debt drift | — | −3.7% |
| Total equity return · Aug 2019 → Aug 2024 | ~3.1× | 100% (~25.4% pa) | |
AutoZone is running a permanent LBO with no exit clock and no PE sponsor taking the carry. Share-count compression is one of the most underestimated drivers of public-market compounding — it shows up nowhere on the income statement.
High repeatability as long as FCF stays strong. The buyback mechanism is structural and continuous. Main risk: adverse demand compressing cash flow against a levered balance sheet.
TransDigm
PE on public markets"Recap dividends alone (~32%) equalled all operating performance combined (~52%)."
Sole-source aerospace SKUs with pricing power + sustained 6–7× EBITDA leverage + periodic recap dividends. Roughly $73 per share in special dividends crossed in multiple recaps over the five-year window. The equity holder gets PE economics without the exit. Multiple expansion is almost an afterthought at +16%.
| Component | Detail | Value / Δ | % of return |
|---|---|---|---|
| Revenue growth | $5.22B → ~$7.94B (organic + bolt-ons) | ~1.52× | — |
| · Volume / units | Aerospace aftermarket recovery + acquired SKUs | ~1.20× | +18.0% |
| · Price / mix | Sole-source SKU pricing; aftermarket mix | ~1.27× | +23.6% |
| EBITDA margin Δ | ~47% → ~52% | +~5 pts | — |
| · Gross margin | ~58% → ~60% | +~2 pts | +3.4% |
| · SG&A leverage | OM ~42% → ~46% | +~2 pts | +5.7% |
| · D&A leverage | Small D&A leverage | +~1 pt | +1.0% |
| · Multiple Δ | ~17× → ~20× | +~3 turns | +16.1% |
| · Leverage / recap Δ | 6–7× EBITDA leverage; ~$73/share special divs | large + | +32.2% |
| Total equity return · Sep 2019 → Sep 2024 incl. specials | ~2.75× | 100% (~22.4% pa) | |
The mechanism: take a business with durable pricing power, lever it permanently at 6–7×, return cash through periodic refinancings. The equity holder gets PE economics without the exit.
High repeatability — pricing power + leverage are structural. Risk: demand downturn in aerospace compressing cash flows against permanent high leverage. Refinancing risk if rates stay elevated.
Constellation Software
M&A is the strategy"~90% of the return came from one lever — acquired revenue growth. The multiple barely moved. M&A is not a supplement to the strategy — it is the strategy."
Revenue grew 2.9×, with ≥80% coming from acquired vertical-software businesses. Margin barely moved. Multiple barely moved. Almost all the return came from one lever — buying small vertical-software businesses at single-digit revenue multiples and adding their ARR to the consolidated entity. Strip out M&A and the company is essentially flat.
| Component | Detail | Value / Δ | % of return |
|---|---|---|---|
| Revenue growth | C$3.49B → ~C$10.07B (≥80% acquisition-driven) | ~2.89× | — |
| · Volume / units | Hundreds of acquired vertical-software businesses | ~2.60× | +81.0% |
| · Price / mix | Modest organic price; maintenance mix | ~1.11× | +8.8% |
| EBITDA margin Δ | ~25% → ~28% | +~3 pts | — |
| · Gross margin | ~30% → ~30% (flat) | ~flat | ~0% |
| · SG&A leverage | OM ~13% → ~14%; modest SG&A leverage | +~1 pt | +6.3% |
| · D&A leverage | Acquired D&A light vs cash margin | +~2 pts | +3.3% |
| · Multiple Δ | ~24× → ~26× | ~+2 turns | +6.8% |
| · Leverage Δ | Modest leverage; Topicus spin-out FY21 | small + | +1.0% |
| · Residual / dilution | Share issuance + Topicus distribution | — | −7.3% |
| Total equity return · Dec 2019 → Dec 2024 (incl. Topicus) | ~3.25× | 100% (~26.6% pa) | |
The alpha is entirely in the deal engine. Underwriting Constellation means underwriting capital allocation skill. The model is durable as long as the fragmented vertical-software universe keeps replenishing.
High repeatability as long as fragmented vertical-software targets remain available at 1–2× revenue. Risk rises if target multiples inflate or deal flow dries up.
Eli Lilly
Platform shift"One breakthrough repriced the entire pipeline — in 2019, Lilly was a pharma company. By 2024, a platform."
Within revenue, roughly 31 percentage points is GLP-1 volume and only 10 points is mix. This is not a price story — it's a units story. The mechanism that won was getting drugs into patients at scale. The multiple roughly doubled, pricing a decade of pipeline optionality.
| Component | Detail | Value / Δ | % of return |
|---|---|---|---|
| Revenue growth | $22.32B → ~$45.0B | ~2.0× | — |
| · Volume / units | GLP-1 volume (Mounjaro/Zepbound) | ~1.70× | +31.1% |
| · Price / mix | GLP-1 mix shift; gross-to-net offset | ~1.19× | +10.2% |
| EBITDA margin Δ | ~30% → ~38% | +~8 pts | — |
| · Gross margin | ~78% → ~81% | +~3 pts | +2.2% |
| · SG&A leverage | OM ~25% → ~33%; large SG&A + R&D leverage | +~5 pts | +14.1% |
| · D&A leverage | D&A scaled with capacity build (small drag) | ~0 pt | −2.4% |
| · Multiple Δ | ~15× → ~30–35× (mid ~32.5×) | +~17 turns | +45.4% |
| · Leverage Δ | Modest debt for capacity; no buybacks | small + | ~0% |
| · Residual / dilution | Approximation noise | — | −0.6% |
| Total equity return · Dec 2019 → Dec 2024 | ~5.5× | 100% (~40.6% pa) | |
Revenue growth (41.3% of log return) → 16.8 pp of the 40.6% pa CAGR. Multiple expansion (45.4%) → 18.4 pp. Together ~35 pp of the 40.6 pp annualized return.
Unlike Costco's sector re-rating, this was intrinsic — GLP-1 changed what the market believed Lilly could do next. ~45% of the return is the multiple paying for pipeline optionality. The question is how much of that is already in the current multiple.
Moderate. Operating performance is real and substantial — but the market is pricing what comes next. Pipeline optionality is partially priced in; returns depend on delivery against that expectation.
Synthesis: six archetypes of compounding
Looking across all seven decompositions, a few things become clear that would have been impossible to see from the headline returns alone. Every compounder fits one of six structural archetypes — and the archetype determines what the next five years look like far more than the headline return does.
The synthesis table
| Company | 5y MOIC | Primary driver | Operating* | Multiple | Leverage | Repeatable? |
|---|---|---|---|---|---|---|
| NVIDIA | 22.5× | Volume + price | ~104% | +19% | ~0% | Partially |
| Apple | 3.25× | Multiple re-rating | ~33% | +51% | +20% | Low |
| Costco | 3.75× | Multiple re-rating | ~44% | +62% | ~0% | Low |
| AutoZone | 3.1× | Leverage / buyback | ~43% | +30% | +32% | High |
| TransDigm | 2.75× | Leverage + ops | ~52% | +16% | +32% | High |
| Constellation | 3.25× | Acquisitions | ~92% | +7% | +1% | High |
| Eli Lilly | 5.5× | Multiple + ops | ~55% | +45% | ~0% | Moderate |
* Operating = volume + price/mix + margin contributions combined.
Durability of return drivers
Volume growth and margin expansion driven by real business activity. Tends to persist as long as competitive position holds. AutoZone, Constellation, and TransDigm have structural mechanisms here.
Buybacks, leverage, recaps. Mathematically repeatable but operationally fragile under stress. Leverage amplifies downturns as much as upturns.
Sector re-ratings are almost always one-shot. Costco's 50× and Apple's 22× are starting points now — not tailwinds for the next five years.
Most investors see outcomes. This article breaks down the machinery underneath — and asks which parts of that machinery are still running.
The conclusion most investors don't reach
The best compounders are not necessarily those with the highest growth. They are the ones whose return composition is durable and repeatable.
Some of the greatest stock returns in this set came from businesses with modest operational performance. Costco grew revenue 1.67× over five years — hardly extraordinary. Apple's iPhone unit count was flat. Yet both produced returns that beat the S&P by a wide margin. The market did not reward growth alone. It rewarded durability — and paid a very high multiple for it.
The most reproducible archetypes in this set are AutoZone (financial structure runs continuously), TransDigm (pricing power + leverage is permanent), and Constellation (M&A machine has replenished consistently). These are businesses where the mechanism that drove returns in the last five years is structurally intact for the next five.
Investors routinely confuse earnings growth with return drivers. Buying back stock ≠ operational growth. Multiple expansion ≠ management skill. Commodity cycles ≠ competitive advantage. The biggest mistake: projecting past returns forward without asking which lever drove them — and whether that lever is still available to pull.
References
Michael David Reinard, Private Equity Value Creation Analysis — Volume I: Theory (2023). The framework underpinning this post. Table 2.15 (Logarithmic Model) restricted to single-company drivers.
Michael David Reinard, Private Equity Value Creation Analysis — Volume II: Applications (2024). Worked diligence and post-close attribution examples.
Tim Koller, Marc Goedhart, David Wessels (McKinsey), Valuation: Measuring and Managing the Value of Companies, 7th ed. (Wiley, 2020). DCF and multiple-arithmetic foundation.
Bogdan (Dan) Baciu, Measuring Returns — IRR, MOIC, and Attribution in a PE Roll-Up (Apr 2026). Companion piece on fund-level mechanics: IRR, MOIC, DPI, TVPI; four-factor attribution bridge; dividend-recap effect on IRR.
All case-study figures are approximate, drawn from public 10-Ks / annual reports / market data and rounded for readability. Verify against primary filings before making investment decisions. Nothing in this article constitutes investment advice.