Openai/6973721c-28b0-8007-a991-09211709d274
This conversation was exported from ChatGPT[1]
gpt-5-2-thinking edit
User: What’s the economic risks of pendle PT looping on morpho or Euler edit
What’s the economic risks of pendle PT looping on morpho or Euler
Also let’s do optimal LTV settings calculation
Assistant: Below is how I’d think about economic not just smart-contract risk when you loop Pendle PT as collateral and borrow on … edit
Below is how I’d think about economic (not just smart-contract) risk when you loop Pendle PT as collateral and borrow on Morpho (Blue) or Euler, and then how to set an “optimal” LTV (really: optimal vs liquidation risk + execution risk + funding risk).
1) Economic risks of Pendle PT looping (Morpho / Euler) edit
A) You’re short funding + long a “zero-coupon bond” edit
A Pendle PT is like a zero-coupon bond: you buy it at a discount and it converges toward redeemable value at maturity. docs.pendle.finance[2] Looping = levering that convergence trade, but you’re also short the borrow rate (variable funding).
Failure mode: borrow APR spikes (utilization shock) while PT implied yield compresses → carry turns negative, you de-lever into bad liquidity.
B) Liquidation is driven by oracle + LLTV/LT, not “fundamentals” edit
Even if PT “should” converge, your position lives/dies by the money market’s collateral valuation rule.
- Morpho Blue: liquidation eligibility is when LTV > LLTV, and health factor is computed directly from collateral value, debt, and LLTV. Morpho Docs[3]
- Euler: liquidation is triggered when risk-adjusted collateral value < debt, and can happen due to price moves or debt growth. Euler Docs[4]
Key economic risk: oracle design (spot vs smoothed vs PT-specific risk oracle). If the oracle can be pulled by secondary-market dislocations, you can be liquidated even though “maturity value” is fine.
C) “Liquidity gap” risk (the silent killer) edit
Liquidators must sell seized PT somewhere (or hedge it). If the PT market is thin / out of range / stressed, liquidation execution can fail or clear at huge discounts → bad debt risk for the market and forced liquidation losses for you. This is explicitly called out in PT risk discussions for money markets. Aave[5]
Looping amplifies this because everyone is crowded on the same unwind path.
D) Maturity + roll risk (cliff events) edit
As maturity approaches, PT volatility typically declines and convergence strengthens, but your loop introduces a “roll decision”:
- close before maturity (market liquidity)
- hold through maturity (redemption mechanics, settlement timing, post-maturity oracle handling, etc.)
If many players roll at once, PT liquidity + borrow liquidity can both gap.
E) Underlying asset + wrapper risk edit
PT inherits risk from:
- Pendle contracts (tokenization + redemption flow)
- the underlying yield-bearing token / protocol (LST/LRT, stable yield token, etc.) A failure in either can reprice PT sharply and instantly. docs.pendle.finance[6]
F) Reflexivity + “loop convexity” edit
Looping makes your liquidation distance nonlinear:
- small adverse move → higher LTV
- higher LTV → less buffer
- any further move → liquidation cascade
This is why “2–3x is fine, beyond that gets hairy” is a common heuristic in PT looping guides. Medium[7]
2) “Optimal LTV” calculation (practical framework) edit
There isn’t one universal optimum because it depends on:
- oracle type + update frequency
- liquidation bonus / close factor / auction mechanics
- PT liquidity + slippage under stress
- your rebalance frequency (do you monitor hourly, daily, weekly?)
- borrow rate volatility (util shock probability)
So the right way is: choose an acceptable liquidation probability over a horizon and back out a safe LTV.
Step 1 — Define variables (per market) edit
- LLTV (Morpho) or LT (Euler equivalent trigger)
- h = risk horizon (e.g., 1 day if you monitor actively; 7 days if you want “sleep well”)
- σ = volatility of collateral value in loan terms (PT priced in borrowed asset)
- r_b = borrow APR (use a stressed value, not today’s)
- LB = liquidation bonus / penalty (how much collateral you lose when liquidated)
- s = stress slippage haircut for unwinds (PT liquidity + DEX depth)
- z = quantile (e.g., 2.33 for 99%, 3.09 for 99.9%)
Step 2 — Compute a conservative “survival buffer” edit
Two things push LTV up over horizon h:
- collateral value can drop
- debt grows via interest
A simple (usable) bound:
- Collateral shock factor:
q=exp(−z⋅σh)q = \exp(-z \cdot \sigma \sqrt{h})q=exp(−z⋅σh)
- Debt growth factor:
g=exp(rb⋅h)g = \exp(r_b \cdot h)g=exp(rb⋅h)
- Execution haircut for liquidation mechanics + slippage (rule of thumb):
η=1−(LB+s)\eta = 1 - (LB + s)η=1−(LB+s) (If LB=7% and you assume 5% slippage in stress, η≈0.88\eta \approx 0.88η≈0.88)
Step 3 — Safe initial LTV edit
For Morpho Blue, you want to stay below LLTV even after shocks:
LTV0,max≈LLTV⋅qg⋅η\textbf{LTV}_{0,\max} \approx LLTV \cdot \frac{q}{g} \cdot \etaLTV0,max≈LLTV⋅gq⋅η This gives you an LTV that has roughly “(1 − tail probability)” chance of staying solvent over horizon h, under your σ/r assumptions, and acknowledges that liquidation is lossy.
3) Translating LTV into “how many loops” (quick) edit
If you loop by borrowing and buying more PT repeatedly, leverage roughly scales like:
Leverage≈11−LTV\text{Leverage} \approx \frac{1}{1-\text{LTV}}Leverage≈1−LTV1 So:
- LTV 0.70 → ~3.33x gross exposure
- LTV 0.80 → 5x
- LTV 0.85 → 6.67x This is why tiny LTV changes massively change liquidation fragility.
4) What I need from you to compute your optimal LTV numbers (no fluff) edit
Send:
- Which PT (e.g., PT-sUSDe-YYYYMMDD) and chain
- Borrow asset (USDC? DAI? ETH?)
- Market + parameters (Morpho: LLTV + oracle type; Euler: LT / risk params if shown)
- Your desired monitoring horizon: 1d / 3d / 7d
- Your stress assumptions (or I’ll apply defaults): - σ (daily vol of PT/borrow) - stressed borrow APR (e.g., 2× current or p95)
If you don’t want to pull vol data, I can still give you a parameterized LTV curve (LTV as a function of σ and horizon) so you can plug numbers in instantly.
Default “founder-safe” settings (if you want an immediate starting point) edit
Without your specifics, a conservative institutional heuristic is:
- target liquidation probability ≤ 0.1% over 7d
- assume stressed borrow APR = 2× current
- assume stress slippage 5–10% on PT
- then set LTV 10–20 points below LLTV/LT for looping (bigger buffer if oracle is spot-like)
This aligns with how PT-as-collateral risk is discussed: liquidation execution + oracle behavior dominate outcomes, not the “it converges at maturity” story. Aave[8]
If you reply with the exact PT + market, I’ll compute:
- recommended LTV band (aggressive/base/conservative)
- max loops / target leverage
- break-even borrow APR vs PT implied yield
- liquidation distance under 1d/7d shocks
gpt-5-2 edit
User: Thanks edit
Thanks
Assistant: Anytime 🙂 edit
Anytime 🙂 If/when you want, we can turn this into a one-pager risk memo (Morpho vs Euler side-by-side, LTV bands, stress table) or run exact numbers for a specific PT. Just say the word.
References edit
- ↑ "Economic Risks". ChatGPT. Retrieved 2026-01-23.
- ↑ "docs.pendle.finance". docs.pendle.finance. Retrieved 2026-01-23.
- ↑ "Morpho Docs". Morpho Docs. Retrieved 2026-01-23.
- ↑ "Euler Docs". Euler Docs. Retrieved 2026-01-23.
- ↑ "Aave". governance.aave.com. Retrieved 2026-01-23.
- ↑ "docs.pendle.finance". docs.pendle.finance. Retrieved 2026-01-23.
- ↑ "Medium". medium.com. Retrieved 2026-01-23.
- ↑ "Aave". governance.aave.com. Retrieved 2026-01-23.