Skip to content
DeFi Reference

Collateral, LTV, and health factor

1. TL;DR

Section titled “1. TL;DR”

Loan-to-value (LTV) caps how much you can newly borrow against a given collateral value, while a separate, always-higher liquidation threshold defines the point at which that position becomes eligible for liquidation — the gap between the two is a deliberate safety buffer against price moves between checks. A single number, the health factor, compresses collateral value, liquidation threshold, and debt value into one figure that must stay above 1; on Solana, marginfi expresses the same idea as an “Initial” weight for borrowing power and a stricter “Maintenance” weight for liquidation, applied per asset. Whether that collateral is walled off to one asset (isolated) or pooled across everything a user has deposited (cross-margin) is a separate, protocol-level design choice with its own risk trade-offs.

2. Explain it simply

Section titled “2. Explain it simply”

Analogy

Section titled “Analogy”

Getting a mortgage, a bank might let you borrow 80% of your house’s appraised value (your LTV cap), but the loan only gets called in if the house’s value falls so far that your debt exceeds, say, 95% of it (the liquidation threshold) — the 15-point gap between 80% and 95% is breathing room for the housing market to wobble without an immediate margin call. A DeFi health factor is the same idea compressed into one dial: as long as it points above 1, you’re fine; below 1, someone else can legally step in and force the sale.

ELI5

Section titled “ELI5”

When you put a toy up as security to borrow coins, the piggy bank only lets you borrow up to a certain fraction of what the toy is worth — that’s how much you can take out today. But it lets your toy’s value drop further before it actually takes the toy away, giving you a cushion. There’s one number that tells you exactly how safe you are: as long as it’s above 1, your toy is safe; the moment it dips below 1, anyone can pay off part of what you owe and take your toy at a discount.

Step-by-step walkthrough

Section titled “Step-by-step walkthrough”

Scenario: Bob (from /lending/money-markets/) deposits 5 ETH as collateral at $3,000/ETH and borrows USDC.

  1. Before borrowing. Bob’s collateral is worth $15,000. The reserve’s ETH parameters are LTV = 80%, liquidation threshold (LT) = 85%. His maximum new borrow is $15,000 × 80% = $12,000.
  2. Bob borrows $9,000 USDC. His debt is $9,000. Health factor HF=collateral×LTdebt=15,000×0.859,000=1.417HF = \dfrac{\text{collateral} \times LT}{\text{debt}} = \dfrac{15{,}000\times0.85}{9{,}000} = 1.417 — comfortably above 1.
  3. ETH drops to $2,200 (state: price shock, no new action from Bob). Collateral value falls to $11,000. HF=11,000×0.859,000=1.039HF = \dfrac{11{,}000\times0.85}{9{,}000} = 1.039 — still above 1, but the buffer has shrunk sharply.
  4. ETH drops further to $1,900. Collateral value is $9,500. HF=9,500×0.859,000=0.897HF = \dfrac{9{,}500\times0.85}{9{,}000} = 0.897 — below 1. Bob’s position is now eligible for liquidation (mechanics in /lending/liquidations/).
  5. What Bob could have done at step 3. Deposit more ETH, repay part of the debt, or swap to a safer E-Mode/Elevation-Group category if his collateral and debt are correlated assets — any of these raises HFHF back above a safe margin.

Common misconceptions

Section titled “Common misconceptions”
  • Myth: LTV and liquidation threshold are the same number. Reality: They are deliberately different — Aave V3 calls this “granular borrowing power control,” splitting “the usual collateral factor… in LTV (which defines the borrowing power for new borrows) and liquidation threshold (which defines the maintenance margin)” (Aave V3 technical paper, 2022-01, §4.3).
  • Myth: A health factor of exactly 1 means you’re about to be liquidated at a loss. Reality: It means the position is at the edge of eligibility; whether it’s actually liquidated, and by how much, depends on liquidator activity and the protocol’s close-factor rules (see /lending/liquidations/).
  • Myth: All your deposited assets count toward your health factor. Reality: Only assets you’ve explicitly enabled as collateral, and in isolated designs, sometimes only one such asset per account — Aave’s Isolation Mode and marginfi’s “Isolated” risk tier both restrict this (Aave V3 technical paper, 2022-01, §4.2; marginfi README, 2026-08 fetch).
  • Myth: Oracle price feeds report one single trustworthy number. Reality: marginfi explicitly prices assets at the low end and liabilities at the high end of an oracle’s confidence interval — “if Oracles report A is $20 +/- $1, assets in A are priced at $19, while liabilities are priced at $21” — precisely so that oracle noise cannot itself manufacture false borrowing power (marginfi README, 2026-08 fetch).
  • Myth: Higher LTV is always better for users. Reality: A high LTV means less price cushion before liquidation; Aave’s E-Mode intentionally offers very high LTV (e.g. 97% for correlated stablecoins) precisely because those assets are expected to move together, not because high LTV is free of risk (Aave V3 technical paper, 2022-01, §3.1).

If you only remember one thing

Section titled “If you only remember one thing”

Loan-to-value tells you how much you can borrow today; liquidation threshold and the health factor built from it tell you how far prices can move before someone else can force you to repay.

3. How it works

Section titled “3. How it works”

Health factor, formally

Section titled “Health factor, formally”

For a position with collateral assets i=1,…,ki=1,\dots,k (amounts cic_i, oracle prices pip_i, liquidation thresholds LTiLT_i) and total debt value DD (summed across borrowed assets at their own oracle prices), the health factor is

HF=∑ici pi LTiDHF = \frac{\sum_i c_i\, p_i\, LT_i}{D}

A position is liquidatable once HF<1HF < 1. Borrowing power for new debt uses the same shape but with each asset’s LTV instead of its liquidation threshold: maxDebt=∑ici pi LTVi\text{maxDebt} = \sum_i c_i\,p_i\,LTV_i — note LTVi≤LTiLTV_i \le LT_i always, which is exactly the safety buffer from §2.

marginfi’s asset/liability weight system

Section titled “marginfi’s asset/liability weight system”

marginfi expresses the identical mechanism with different vocabulary: every asset has an Initial weight (used when checking new borrows) and a stricter Maintenance weight (used when checking liquidation eligibility), and every liability has its own Initial and Maintenance weight too, with Maintenance liability weight always lower than Initial. Health is “the sum of assets minus the sum of liabilities, after all applicable weights and confidence intervals have been applied” (marginfi README, 2026-08 fetch):

Health=∑i(priceilow×qtyi×wiasset)−∑j(pricejhigh×qtyj×wjliab)\text{Health} = \sum_i (\text{price}_i^{\text{low}}\times \text{qty}_i \times w^{\text{asset}}_i) - \sum_j (\text{price}_j^{\text{high}}\times \text{qty}_j \times w^{\text{liab}}_j)

A borrow is rejected if Initial-weighted health would fall below zero; liquidation becomes possible once Maintenance-weighted health falls below zero — the same two-threshold structure as LTV/liquidation-threshold, just expressed as a subtraction instead of a ratio. Oracle confidence intervals are folded directly into the price used (assets priced low, liabilities priced high), capped at 5% — “if the Oracle reports a confidence interval higher than 5%, we may clamp it to 5% or abort the transaction” (marginfi README, 2026-08 fetch).

Isolated vs. cross-margin accounting

Section titled “Isolated vs. cross-margin accounting”

Cross-margin (the historical Aave/Compound default) aggregates all of a user’s enabled collateral against all of their debt in one health computation — capital-efficient, since idle collateral in one asset backs a borrow in any other, but a single mispriced or crashing collateral asset threatens every other position the account holds.

Isolated accounting walls collateral off instead. Aave’s Isolation Mode: a borrower using a designated isolated asset “cannot supply other assets as collateral,” and can only borrow specific stablecoins up to a hard USD debt ceiling (Aave V3 technical paper, 2022-01, §4.2). marginfi’s Isolated risk tier: banks in this tier have an asset weight fixed at zero — they cannot count as collateral at all, though they can still be borrowed, and “accounts can only borrow one isolated asset at a time” (marginfi README, 2026-08 fetch). Both let a protocol list a new or volatile asset without exposing every other depositor’s collateral to it.

Category-correlated boosts: E-Mode and Emode

Section titled “Category-correlated boosts: E-Mode and Emode”

When collateral and debt are known to move together (stablecoin-for-stablecoin, ETH-for-stETH), both ecosystems offer a higher-LTV override. Aave’s E-Mode overrides LTV/liquidation threshold/bonus (and optionally the oracle) whenever a borrower’s entire debt is within one category (Aave V3 technical paper, 2022-01, §4.1). marginfi’s Emode does the same at the asset-pair level, but “an Account’s Emode benefit for a given token being lent is always based on the worst benefit across all the assets they are borrowing” — mixing an eligible borrow with an unrelated one can silently forfeit the boost (marginfi README, 2026-08 fetch).

4. Worked numeric example

Section titled “4. Worked numeric example”

Reusing Bob’s position from §2 at the point HF=0.897HF = 0.897 (ETH at $1,900, 5 ETH collateral, $9,000 USDC debt, LTV 80%/LT 85%): a liquidator can now legally intervene (mechanics and payout math in /lending/liquidations/).

Now the marginfi-style version of the same idea, using the exact worked figures from marginfi’s own README (2026-08 fetch) to show the weight-based method reaches the same conclusion as the ratio-based method: Token A is priced at $10 with a $0.212 confidence band (so $9.788 low / $10.212 high), USDC at $1 with $0.0212 confidence ($0.9788 low / $1.0212 high). A user holds 2 Token A (asset) and owes 5.05 USDC (liability).

  • Using Initial weights (Token A asset weight 50%, USDC liability weight 100%): Health=(2×9.788×0.5)−(5.05×1.0212×1)=4.631\text{Health} = (2\times9.788\times0.5) - (5.05\times1.0212\times1) = 4.631 — positive, so new borrows are still allowed.
  • Using Maintenance weights (Token A asset weight drops to 10%, USDC liability weight stays 100%): Health=(2×9.788×0.1)−(5.05×1.0212×1)=−3.199\text{Health} = (2\times9.788\times0.1) - (5.05\times1.0212\times1) = -3.199 — negative, so the position is liquidatable even though the user still holds more dollar-value in assets ($19.58) than debt ($5.05); the weights, not the raw values, decide eligibility. A partial liquidation seizing 0.2 Token A at a 2.5% liquidator fee and 2.5% insurance fee repays 0.2×(1−0.025−0.025)×9.788=1.8600.2\times(1-0.025-0.025)\times9.788 = 1.860 USD of the debt and brings Maintenance-weighted health to −1.496-1.496 — improved, but still short of healthy, illustrating why liquidations are often partial rather than one-shot (marginfi README, 2026-08 fetch).

5. Where it’s used

Section titled “5. Where it’s used”

Ethereum

Section titled “Ethereum”
  • Aave V3 — per-reserve LTV, liquidation threshold and liquidation bonus, with Isolation Mode and E-Mode overrides as described above. Aave V3 overview docs
  • Compound and Morpho — Compound popularized the same LTV/collateral-factor structure on a single aggregated pool; Morpho Blue instead lets each isolated market’s creator set its own LTV — full treatment in /lending/modular-lending/.

Solana

Section titled “Solana”
  • marginfi (rebranded Project 0 / P0) — Initial/Maintenance asset and liability weights plus oracle confidence-interval adjustment, as detailed above; considered by this page’s sources the most granular public documentation of collateral risk parameters among the Solana protocols surveyed. docs.marginfi.com
  • Kamino Lend — Obligations track “position health (LTV ratios)” and enforce liquidation thresholds per reserve, with Elevation Groups providing an Aave-E-Mode-like correlated-asset override (Kamino architecture docs, 2026-08 fetch).
  • Save (formerly Solend) and Jupiter Lend — both implement collateralized borrowing with per-asset risk parameters per their own docs, but neither fetched source spells out exact LTV/threshold figures (Save docs; Jupiter developer docs, 2026-08 fetch) — treat specific numbers as unverified pending a deeper docs read.

6. Risks, attacks, and incidents

Section titled “6. Risks, attacks, and incidents”
  • Mango Markets exploit, 11–12 October 2022 (Solana). Attacker Avraham Eisenberg opened a large MNGO-PERP position, then used a second account to pump the illiquid MNGO spot price from roughly $0.038 to $0.91 on Mango’s own market. Because Mango’s risk engine valued his perp position’s unrealized profit at the manipulated oracle price and counted it as collateral, he was able to borrow roughly $100–117.8M in real assets (as of 2022-10) against paper “collateral” that existed only because of his own manipulation — sources disagree on the figure, with early reporting near $100M and later on-chain tracing near $116–117.8M (secondary: Elliptic, 2022-10; Cointelegraph, 2022-10, both accessed 2026-08-29 — no primary Mango post-mortem was retrievable). The Mango DAO, with Eisenberg voting his own (stolen) governance tokens, then let him keep roughly $47M as a “bug bounty” in exchange for returning the rest; U.S. prosecutors later charged and a jury convicted him. This is the canonical illustration of why a health-factor calculation is only as trustworthy as the price feed behind the collateral it’s valuing — see /oracles/oracle-manipulation/.
  • Confidence-interval gaming. Any system, like marginfi’s, that widens or narrows the effective price based on a reported confidence interval is itself an attack surface if an attacker can influence that confidence — mitigated by hard-capping the adjustment at 5% and reserving the right to abort the transaction if it’s exceeded (marginfi README, 2026-08 fetch).
  • Isolation-mode debt-ceiling exhaustion. Because an isolated asset’s debt ceiling is a fixed USD cap shared by all borrowers using it, one user maxing it out can lock out every other borrower of that asset until governance raises the cap (Aave V3 technical paper, 2022-01, §4.2).
  • Cross-margin contagion. In a non-isolated account, a single crashing collateral asset can push health factor below 1 even for a user whose other collateral was perfectly safe, since all collateral and debt share one combined ratio — the structural reason isolated designs exist.
  • E-Mode/Emode category mis-classification. Both systems depend on governance correctly asserting grouped assets are actually correlated; the Aave paper concedes “the correct categorization cannot be enforced on-chain” (Aave V3 technical paper, 2022-01, §4.1) — a wrong assumption (e.g. a de-pegging LST) can silently turn a “safe” category under-collateralized.

7. Open problems

Section titled “7. Open problems”
  • How wide should the LTV/liquidation-threshold gap be? Too narrow and normal volatility triggers unnecessary liquidations; too wide and capital efficiency suffers — Aave’s “granular borrowing power control” treats this as a tunable judgment call, not a solved formula (Aave V3 technical paper, 2022-01, §4.3).
  • Whether category-based correlation assumptions (E-Mode/Emode) are safe long-term. Both designs concede the correlation assumption is maintained off-chain by governance/admins, not enforced by the protocol (Aave V3 technical paper, 2022-01, §4.1).
  • Recourse after an oracle-driven health-factor miscalculation. cc7768’s ethresear.ch framework lists “incorrect liquidations due to manipulated prices” among its motivating examples (ethresear.ch 11182, 2021-11-04) — Mango is the clearest large-scale case of this materializing.
  • Whether confidence-interval-aware pricing generalizes beyond Solana’s oracle stack. marginfi’s adjustment depends on Pyth/Switchboard publishing confidence bands; whether it’s worth adopting where push oracles report only a point price is not addressed by this page’s sources.

8. Ethereum vs Solana

Section titled “8. Ethereum vs Solana”
AspectEthereumSolana
Core mechanismLTV (borrow) vs. liquidation threshold (LT) ratio, health factorInitial vs. Maintenance asset/liability weights, health as a subtraction
Oracle uncertainty handlingNot addressed by a confidence interval in this page’s Aave sourcesmarginfi explicitly prices assets low / liabilities high within a capped 5% oracle confidence band
Isolation mechanismIsolation Mode (hard USD debt ceiling, single collateral asset)marginfi “Isolated” risk tier (zero asset weight); Kamino Elevation Groups
Correlated-asset boostE-Mode (per-category LTV/LT/bonus/oracle override)marginfi Emode (per-pair weight boost, worst-case across positions)
Best-documented incident(see /lending/flash-loans/ for bZx/Cream)Mango Markets, October 2022 — oracle-inflated collateral value

The mathematics is interchangeable — a ratio test and a weighted-subtraction test express the same safety condition — but the two ecosystems differ in how explicitly they price oracle uncertainty: marginfi’s confidence-interval adjustment is a direct, quantified response to Solana’s reliance on frequently updated Pyth/Switchboard feeds, while the Ethereum sources here treat the oracle price as a single trusted number and instead invest more heavily in isolating which assets can even become collateral together.

9. Reference doc

Section titled “9. Reference doc”

The reference

Section titled “The reference”

Aave V3 Overview — Aave docs (aave.com), fetched 2026-08-29. aave.com/docs/aave-v3/overview

Summary of the reference

Section titled “Summary of the reference”

This short, current-state docs page (not the 2022 technical paper) walks through Aave V3’s user-facing mechanics in plain language: supplying mints aTokens that “accrue… interest over time from borrowing activity in the pool”; borrowing requires that “positions are always over-collateralized”; and “risk is tracked with a Health Factor and per-reserve liquidation thresholds; when the Health Factor drops below the threshold, collateral can be liquidated.” It briefly restates E-Mode and Isolation Mode as key features (higher LTV for correlated assets; safe listing of new/volatile assets behind a debt ceiling) and adds a feature not in the 2022 paper — “Siloed Borrowing,” a reserve-level flag restricting a borrower of one asset from holding any other simultaneous borrow. It closes by pointing to ERC-4626 “Aave Earn Vaults” as a separate, newer product built on top of the same supply/borrow primitives.

Key quotes

Section titled “Key quotes”

“Positions are always over-collateralized, meaning the collateral value must exceed the borrowed amount.” (§Borrow)

“Risk is tracked with a Health Factor and per-reserve liquidation thresholds; when the Health Factor drops below the threshold, collateral can be liquidated.” (§Borrow)

“Efficiency Mode maximizes capital efficiency for correlated assets.” (§Aave v3 Key Features, Efficiency Mode)

“Siloed borrowing is a reserve-level flag that restricts users who borrow a given asset to borrowing only that asset.” (§Aave v3 Key Features, Siloed Borrowing)

How to read the original

Section titled “How to read the original”

Background needed: none beyond this page’s own §3 — the docs page is intentionally written for a general audience and assumes no prior DeFi knowledge. There is nothing to skip; it is short by design. The one subtlety worth pausing on is the “Siloed Borrowing” paragraph, which is easy to conflate with Isolation Mode: Isolation Mode restricts what a borrower can use as collateral, while Siloed Borrowing restricts what they can simultaneously borrow — the two can apply to the same or different assets independently.

What changed since

Section titled “What changed since”
  • Siloed Borrowing and Aave Earn Vaults are both absent from the 2022 technical paper referenced on /lending/money-markets/ and /lending/interest-rate-models/ — evidence Aave has kept adding reserve-level risk controls incrementally rather than only at major version boundaries.
  • The Mango Markets exploit (October 2022) postdates the Aave V3 paper and this docs page’s core design, and was not an Aave incident, but it became the industry’s most-cited example of exactly the health-factor failure mode both documents’ oracle-price assumptions are vulnerable to in principle.

Secondary references

Section titled “Secondary references”
  1. marginfi-v2 README, “Risk Engine” section (GitHub, 2026-08 fetch) — read if you want the same LTV/health-factor logic expressed with explicit weights and a fully worked liquidation example.
  2. Aave V3 Technical Paper, §4.1 and §4.3 (2022-01-27) — read if you want the underlying invariants and edge-case rules (e.g. the “LTV == 0 withdrawal ordering” exploit Aave itself documents and closes) behind the plain-language docs page.
  3. Elliptic, “Mango Market Exploit” incident writeup (2022-10, secondary) — read if you want the fullest secondary account of a real health-factor/oracle failure at scale, pending a primary Mango post-mortem becoming available.

10. Sources

Section titled “10. Sources”