Yield tokenization — Pendle's PT/YT split and fixed rates

1. TL;DR

Yield tokenization splits a yield-generating deposit into two separate, tradable claims: a Principal Token (PT), a zero-coupon-bond-like claim redeemable 1:1 for the underlying asset at a fixed future expiry, and a Yield Token (YT), which captures every unit of variable yield the deposit generates until that same expiry. Buying PT below its redemption value locks in a fixed annualized rate; buying YT is a leveraged, decaying bet that the underlying’s floating yield will outpace what the market currently implies. Pendle V2 built a custom automated market maker specifically for trading PT against the underlying, because the two generic curve shapes DeFi already had — a constant-product-style curve and the “YieldSpace” power-sum curve — both become dramatically capital-inefficient as a pool nears its expiry date, exactly when trading activity tends to matter most.

2. Explain it simply

Analogy

Imagine a bank certificate of deposit that pays a variable interest rate you can’t predict in advance. Yield tokenization is like a service that lets you split that certificate into two separate paper slips: one slip is redeemable for exactly your original deposit on the maturity date, no more, no less — you can sell that slip today at a discount to someone who wants a guaranteed, fixed return — and the other slip entitles its holder to collect every interest payment the certificate makes between now and maturity, whatever that ends up totaling. Someone confident that rates are about to spike will pay you for that second slip; someone who just wants certainty will buy the first one from you at a discount instead of waiting.

ELI5

Some crypto deposits earn interest that changes over time, going up or down depending on market conditions. Yield tokenization takes one of those deposits and splits it into two separate tokens: one token you can trade in today for a fixed, guaranteed amount later — like a savings bond — and a second token that collects however much interest actually gets paid out along the way, which could be more or less than expected. If you want certainty, you buy the first token at a discount and know exactly what you’ll get back. If you think interest rates are about to rise, you buy the second token instead, paying up front for the right to collect whatever the real, changing interest turns out to be.

Step-by-step walkthrough

Scenario: a Pendle-style pool holds a yield-bearing token derived from lending USDC (cUSDC-style), with 2 years left until expiry, and the pool’s current price implies an annualized fixed rate of 9%.

1. Pool opened (state: pool holds a mix of the yield-bearing asset and its Principal Token, priced so 1 asset unit today is worth 1.1881 PT at expiry — a 9% annualized fixed rate over the 2 years, since $1.09^2=1.1881$). Anyone can deposit the underlying yield-bearing asset and split it into one PT plus one Yield Token (YT) per unit deposited.
2. Alice buys PT (state: Alice pays less than 1 unit of the underlying per PT she buys). She is locking in the 9% fixed annualized rate for two years — at expiry, each PT she holds redeems for exactly 1 unit of the underlying, regardless of what the underlying’s floating rate actually did in the meantime.
3. Bob buys YT instead (state: Bob pays up front for a claim on all future variable interest). Bob is betting the underlying lending market’s floating rate will average more than 9% annualized over the next two years; if it does, his YT payouts exceed what he paid for it.
4. A large trader sells 10,900 PT into the pool (state: implied fixed rate moves from 9% to 11%). Because more PT supply is now available relative to the underlying asset in the pool, PT becomes cheaper (a bigger discount to par), which is mathematically the same thing as the implied fixed rate rising from 9% to 11% — exactly the trade size the paper’s own capital-efficiency comparison in §4 works through for three different AMM curve shapes.

Common misconceptions

• Myth: A Principal Token is just a regular token pegged to $1. Reality: “Principal token (PT) is a token that allows users to get back 1 asset worth after the expiry… buying PT is also equivalent to shorting yield, profiting in the short run if yield drops” (Pendle V2 AMM whitepaper, 2022-10) — its price floats below the underlying asset’s value until expiry, precisely tracking the implied fixed rate.
• Myth: Any constant-product-style AMM works fine for trading PT against the underlying. Reality: The whitepaper’s own capital-efficiency study finds the plain constant-geometric-mean model “becomes much less efficient compared to the other two models as it approaches expiry” — the same pool that works fine two years out can become extremely thin and hard to trade in its final months if it uses the wrong curve.
• Myth: Buying YT is a low-risk way to earn yield. Reality: A YT holder pays up front for a stream of future variable payments and can lose the entire premium if realized yield comes in below what was implied at purchase — it is a leveraged, time-decaying bet on the direction of yield, not a low-risk savings product.
• Myth: Yield tokenization protocols are only exposed to their own smart-contract risk. Reality: Because PT and YT both derive their value from a wrapped underlying yield-bearing asset, “underlying-asset risk passes through fully” — a depeg or slashing event in the wrapped asset impairs both PT and YT regardless of the tokenization layer’s own security (industry guides, secondary, 2026).

If you only remember one thing

Splitting a variable-rate deposit into a fixed-redemption Principal Token and a variable-payout Yield Token turns “what will this yield?” into a tradable market price — the discount on PT is the market’s fixed rate.

3. How it works

Wrapping and splitting

Pendle first wraps a yield-bearing position into a Standardized Yield (SY) token, a common interface so any yield source can plug into the same downstream machinery, then splits SY into one Principal Token (PT) and one Yield Token (YT) per unit deposited. PT alone is redeemable for exactly 1 unit of the underlying asset at the market’s expiry; YT alone entitles its holder to every unit of yield the underlying SY generates between now and that expiry. Define normalised time $t\in[0,1]$, with $t=0$ at expiry, and let $\text{price}_p(t)$ be PT’s price in terms of the asset, so $\text{price}_a(t)=1/\text{price}_p(t)$ is the asset’s price in terms of PT. Since holding 1 asset today buys $\text{price}_a(t)$ PT, each redeemable for 1 asset at expiry, the implied annualized interest rate over a remaining period of $\tau$ years is

$$\text{Implied APY} = \big(\text{price}_a(t)\big)^{1/\tau} - 1$$

Buying PT pushes $\text{price}_p(t)$ up and $\text{price}_a(t)$ down, which lowers the implied rate; selling PT does the reverse.

Three ways to shape the PT-asset curve

Let $x$ and $y$ be the pool’s asset and PT reserves, and $p = y/(x+y)$ the proportion of PT in the pool. The whitepaper compares three AMM shapes used across DeFi for this exact problem:

1. Constant geometric mean (used by Pendle V1 and Apwine): $x^{w_x}y^{w_y}=k$ with $w_x+w_y=1$, giving $\text{price}_a = \tfrac{w_x}{w_y}\cdot\tfrac{p}{1-p}$. This is the plain Balancer/Uniswap-v2-style curve, applied with no awareness that PT approaches par as expiry nears.
2. YieldSpace’s constant power sum (Yield Protocol, Element Finance, Sense Finance): $x^{1-t}+y^{1-t}=k$, giving $\text{price}_a = \left(\tfrac{p}{1-p}\right)^{t}$ — the exponent $t$ shrinking toward 0 as expiry approaches flattens the curve near $t=0$, which is more capital-efficient close to maturity than the geometric-mean model but behaves identically to it far from expiry.
3. Notional’s AMM (adopted, with modifications, for Pendle V2): rather than defining reserves via a formula, this model directly specifies $\text{price}_a(t) = \dfrac{\ln\!\left(\frac{p}{1-p}\right)}{\text{rateScalar}(t)} + \text{rateAnchor}(t)$, where $\text{rateScalar}(t)=\text{scalarRoot}/t$ controls capital efficiency and $\text{rateAnchor}(t)$ is recalibrated before every trade so the pre-trade implied rate always matches the rate left by the previous trade.

Why Pendle V2 chose the Notional shape

The whitepaper’s capital-efficiency study measures, for a fixed pool size, how large a PT trade is needed to move the implied rate by a set amount under each model. Across three simulated scenarios (a 2-year cUSDC-style pool, a 3-month high-yield pool, and a 1-year stETH-style pool), “the Notional AMM model is consistently the most capital efficient… around 9-35 times more capital efficient than geometric mean, and 8-9.4 times more capital efficient than YieldSpace’s,” with the gap “the most pronounced for yield bearing assets with a tighter range of possible interest rates” because the Notional curve’s two free parameters (scalar root and rate anchor) let each pool be custom-tuned to its own asset’s realistic rate range, unlike the two generic curves that use the same shape for every asset regardless of its typical volatility.

4. Worked numeric example

This example reproduces the whitepaper’s own Scenario 1 exactly. Setup: a pool trading fixed yield for cUSDC, expiry in 2 years, average expected annualized rate 9% ($\text{rateExpected}=1.09$), maximum plausible annualized rate 20% ($\text{rateMax}=1.2$), and 1,000,000 USDC-worth of PT and USDC combined in the pool. Following the paper’s heuristic for choosing Notional-AMM parameters from these assumptions, the pool starts (at $t=1$, the full 2 years remaining) with an initial rate anchor of $1.09^{2}=1.1881$ and $\text{rateScalar}(t{=}1)=8.7226$.

Pushing the implied rate from 9% to 11% at full time-to-expiry. A trader wants to sell PT to move the pool’s implied annualized rate from 1.09 (9%) to 1.11 (11%). Under both the constant-geometric-mean model and the YieldSpace model, that trade can only absorb 10,900 PT before hitting the 11% target. Under the Notional AMM, the same size move in implied rate can absorb 102,936 PT — 9.44 times as much (102,936 / 10,900 ≈ 9.44), confirming the paper’s stated multiplier for this scenario.

How the gap changes over time. The same comparison at 1 year left (rate move 1.11→1.13) gives trade sizes of 9,920 PT (YieldSpace) versus 87,671 PT (Notional); at 6 months left (1.07→1.09) the sizes are 9,567 versus 83,300. The geometric-mean model degrades badly as expiry nears — its capacity for the same trade falls from 10,900 PT at 2 years left to just 2,400 PT at 6 months left, roughly a 4.5x worse outcome, exactly the effect the whitepaper flags: “the model does not take into account the fact that PT price will converge to 1 at expiry.” The Notional AMM, by contrast, stays in the 83,000–103,000 PT range throughout, which is precisely the property Pendle V2 needed to keep its pools tradable all the way to maturity.

5. Where it’s used

Ethereum

• Pendle — the dominant yield-tokenization protocol; V1 used a Sushiswap-derived constant-geometric-mean pool, V2 adopted the Notional-derived AMM described in §3–§4; total value locked reached a peak of roughly $8.9 billion in August 2025 during the EigenLayer/liquid-restaking wave and stood at roughly $1.04 billion across 12 chains as of July 2026 (secondary, dated). docs.pendle.finance
• Notional Finance, Element Finance, Sense Finance, Yield Protocol — the earlier protocols whose AMM designs Pendle’s own capital-efficiency study directly compares against; Notional’s rate-scalar/rate-anchor approach is the one Pendle V2 builds on, with modifications, for its own PT market.

Solana

• Exponent — an emerging Solana-native yield-tokenization protocol bringing a PT/YT-style split to Solana lending and staking yield sources (RockawayX / eco.com, secondary, as of 2026). No Solana-native primary documentation for a live, audited PT/YT AMM comparable to Pendle V2’s was found in the sources reviewed for this page, so mechanism-level detail here should be treated as directional rather than verified against a primary source.

6. Risks, attacks, and incidents

• Penpie exploit, 4 September 2024 — approximately $27 million. Penpie, a yield-boosting protocol built on top of Pendle (not Pendle itself), was drained via a reentrancy vulnerability in its PendleStaking contract, which lacked reentrancy guards. Because Pendle’s market creation is permissionless — anyone can deploy a market with a custom SY contract — the attacker created a market backed by a malicious SY contract, then used flash loans and Penpie’s requirement to call external SY contracts when claiming rewards to inflate their credited rewards and drain the pool. Pendle itself was unaffected and stated it had paused contracts in time to prevent an additional “over $100 million” of user funds from being put at risk; the attacker did not return the Penpie funds and laundered them through Tornado Cash (Halborn / The Defiant / QuillAudits, secondary, 2024-09). This incident is a useful illustration that a tokenization layer’s permissionless composability is a double-edged sword: any protocol built on top of it inherits both its flexibility and its attack surface.
• Underlying-asset risk passes through by construction. Both PT and YT derive their entire value from the wrapped SY position, so “a PT-sUSDe holder bears Ethena depeg risk, a PT-eETH holder bears EtherFi and EigenLayer slashing risk” — a de-peg, slashing event, or governance failure in the underlying yield source impairs PT and YT alike regardless of Pendle’s own contract security (industry guides, secondary, 2026).
• YT is a decaying, potentially total-loss position. “YT carries time decay as a structural cost, every day held without realized yield exceeding implied yield is a day of loss,” and in the extreme case where PT trades near par close to expiry, “the APY from YT tokens could fall as low as -100%” — a YT buyer who mis-forecasts the direction of yield can lose their entire premium (industry guides, secondary, 2026).
• AMM curve choice is itself a risk parameter. As shown numerically in §4, a pool built on the wrong curve shape (plain constant-geometric-mean) becomes progressively thinner and more slippage-prone as expiry approaches — precisely when traders most want to exit or roll a position — which is a structural liquidity risk baked into the AMM design choice rather than a bug.

7. Open problems

• Whether Notional-style parameter tuning generalizes safely to novel or volatile yield sources. The whitepaper’s heuristic for setting scalar root and rate anchor assumes the pool operator can reasonably estimate an expected rate and a maximum plausible rate in advance; for a genuinely novel or fast-moving yield source, misjudging those bounds could reproduce the geometric-mean model’s near-expiry inefficiency problem within the “improved” curve.
• Composability risk from permissionless market creation. The Penpie incident shows that letting anyone create a Pendle market with an arbitrary SY contract is a feature for extensibility and a liability for security simultaneously; how much validation a base layer should impose on markets built atop it — versus leaving that entirely to integrators — remains an open design trade-off industry-wide, not specific to Pendle.
• Fair pricing of YT under regime change. Because YT’s payoff depends on the path of a variable yield rate rather than a single terminal price, standard options-style pricing intuition transfers only partially; how best to price and hedge YT when the underlying yield source’s own risk profile shifts mid-life (e.g., a lending market’s utilization curve changing sharply) is not addressed in the reference document reviewed for this page.

8. Ethereum vs Solana

Aspect	Ethereum	Solana
Dominant protocol	Pendle (V1 geometric-mean, V2 Notional-derived AMM)	Exponent (emerging, secondary-sourced only)
TVL scale	~$1.04bn across 12 chains (as of 2026-07); peaked ~$8.9bn in 2025-08	Not found in sources reviewed
AMM maturity	Multiple generations compared and benchmarked in a public whitepaper	No primary AMM whitepaper found in sources reviewed
Composability risk demonstrated	Yes — Penpie exploit (2024-09, ~$27M) shows real integration risk	Not found in sources reviewed
Underlying yield sources tokenized	Lending markets, liquid staking/restaking tokens, stablecoin yield (e.g. sUSDe)	Lending and staking yield sources (per secondary reporting)

Ethereum’s yield-tokenization ecosystem is mature enough to have gone through multiple AMM design generations, a public capital-efficiency benchmarking exercise, and a real integration-layer exploit — Pendle’s TVL swings between roughly $1 billion and $9 billion track the broader liquid-staking and restaking yield cycles on Ethereum. Solana’s equivalent (Exponent) is comparatively nascent, and this page found no primary technical documentation for it, so a like-for-like mechanism comparison is not yet possible from the sources available.

9. Reference doc

The reference

A study on AMMs for trading fixed yield and Pendle V2’s Principal Token AMM — Vu Nguyen (Pendle Finance), 14 October 2022. Pendle V2 AMM whitepaper

Summary of the reference

The paper opens by defining the core vocabulary of principal-token trading — a yield generating asset following the Standardized Yield (SY) model, the expiry at which a Principal Token (PT) becomes redeemable 1:1 for the underlying asset, normalised time $t\in[0,1]$, and the implied interest rate derived from PT’s price relative to the asset. It then surveys three existing models for a principal token AMM: the constant geometric mean formula used by Pendle V1 and Apwine, the constant power sum formula from YieldSpace (used by Yield Protocol, Element Finance, and Sense Finance), and Notional Finance’s AMM, which is defined directly as a formula for the implied interest rate rather than via pool reserves, parameterized by a rate scalar (controlling capital efficiency) and a rate anchor (recalibrated before every trade to preserve continuity with the prior implied rate). A capital-efficiency study compares all three across three realistic scenarios — a 2-year cUSDC-style pool, a 3-month high-yield pool, and a 1-year stETH-style pool — consistently finding the Notional model most capital-efficient, especially for assets with a tight expected rate range, and finding the constant-geometric-mean model degrades sharply as expiry approaches because it has no mechanism to reflect PT’s convergence to par. Based on this study, the paper adopts the Notional AMM model, with modifications, as the basis for Pendle V2’s own Principal Token AMM, and begins defining the V2-specific building blocks — the SY/PT market, total asset value, and time-to-expiry — that the remainder of the (longer) whitepaper builds on.

Key quotes

“Principal token (PT) is a token that allows users to get back 1 asset worth after the expiry. As the final value of principal token is known, PT holders receive fixed yield on expiry.” (§1.1 Definitions)

“Buying principal tokens will increase its price… and decrease asset price in principal… hence decreasing the implied interest rate. Conversely, selling principal tokens will increase the implied interest rate.” (§1.1 Definitions)

“The Notional AMM model is consistently the most capital efficient… around 9-35 times more capital efficient than geometric mean, and 8-9.4 times more capital efficient than YieldSpace’s.” (§1.3.4 Conclusions)

“The geometric mean model becomes much less efficient compared to the other two when nearing expiry… because the model does not take into account the fact that PT price will converge to 1 at expiry.” (§1.3.4 Conclusions)

How to read the original

Background needed: the concept of a zero-coupon bond (a PT is economically one), and basic AMM reserve-curve notation ($x$, $y$, $k$). Skip the full derivation of the Notional AMM’s heuristic parameter-setting inequalities (equations 7–13 in the source) on a first pass — the conclusion, that scalar root should be set to keep the reasonable trading range (proportion 0.1 to 0.9) covering the expected rate range, matters more than the algebra that gets there. The hardest part is distinguishing $\text{price}_p(t)$ (PT’s price in asset terms, always below 1 before expiry) from $\text{price}_a(t)=1/\text{price}_p(t)$ (the asset’s price in PT terms, always above 1) — the paper switches between them depending on which is more convenient for a given formula, and conflating the two inverts every inequality in the capital-efficiency section.

What changed since

• Pendle V2 shipped in production with the Notional-derived AMM this paper recommends, and later added Boros, a separate product for trading off-chain and cross-market interest rates with leverage (Pendle, secondary, 2026).
• Pendle’s TVL has swung enormously with the broader liquid-staking and restaking cycle, from roughly $230 million in early 2023 to a peak of roughly $8.9 billion in August 2025, back to roughly $1.04 billion across 12 chains by July 2026 (secondary, dated).
• The composability risk implicit in Pendle’s permissionless market creation (any SY contract can be used) was realized in practice via the September 2024 Penpie exploit on a protocol built atop Pendle, not Pendle’s own contracts (see §6).

Secondary references

1. Pendle Finance official documentation (docs.pendle.finance) — read for the current, user-facing explanation of SY/PT/YT and the newer Boros product, though note this page found the live docs difficult to extract stable quotes from due to dynamic content.
2. Penpie exploit post-mortems (Halborn, QuillAudits, Three Sigma) — read if you want the detailed reentrancy mechanics of the most consequential real-world incident touching the Pendle ecosystem to date.
3. YieldSpace whitepaper (Yield Protocol) — read if you want the original derivation of the constant-power-sum curve this paper benchmarks Pendle V2’s AMM against.

10. Sources

• A study on AMMs for trading fixed yield and Pendle V2’s Principal Token AMM — Vu Nguyen (Pendle Finance) — 2022-10-14 — https://raw.githubusercontent.com/pendle-finance/pendle-v2-resources/main/whitepapers/V2_AMM.pdf
• Pendle Documentation (docs.pendle.finance) — fetched 2026-08-30 — https://docs.pendle.finance/
• Explained: The Penpie Hack (September 2024) — Halborn (secondary) — 2024-09 — https://www.halborn.com/blog/post/explained-the-penpie-hack-september-2024
• Decoding Penpie Protocol’s $27M Exploit — QuillAudits (secondary) — 2024-09 — https://www.quillaudits.com/blog/hack-analysis/penpie-protocol-exploit
• Pendle Pauses Contracts After Yield Protocol Penpie Suffers $27 Million Exploit — The Defiant (secondary) — 2024-09 — https://thedefiant.io/news/defi/pendle-pauses-contracts-after-yield-protocol-penpie-suffers-usd27-million-exploit
• Pendle Finance Explained — Eco (secondary) — fetched 2026-08 — https://eco.com/support/en/articles/15002226-pendle-finance-explained
• What Is Pendle Finance? The Complete 2026 Guide — earnpark.com (secondary) — 2026 — https://earnpark.com/en/posts/what-is-pendle-finance-the-complete-2026-guide-to-yield-tokenisation-pt-yt-mechanics-and-boros/
• Exponent is coming for Solana DeFi — RockawayX Insights (secondary) — 2026 — https://www.rockawayx.com/insights/exponent-is-coming-for-solana-defi----at-just-the-right-time