Skip to content

alexeinazarov/stableunit-risk-framework

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 

Repository files navigation

1. Aave’s official risk outputs and why they matter
The two external risk managers that serve Aave—Gauntlet and Chaos Labs—publish the same headline solvency metrics:

Chaos Labs’ Aave V3 Risk Parameter Methodology (https://chaoslabs.xyz/resources/chaos_aave_risk_param_methodology.pdf) specifies 10⁴–10⁶ Monte-Carlo paths, a 24 h horizon, block-level health-factor tracking, and a Bayesian search that maximises E[revenue]/iVaR while forcing iVaR ≤ protocol reserves.

A practical enhancement would be to publish

  • Expected Shortfall (ES) = average bad-debt conditional on breaching the 99-th percentile, and
  • Liquidity-adjusted VaR, i.e. iVaR + κ·(position / on-chain depth) with κ≈2, following Almgren–Chriss impact theory (https://ideas.repec.org/a/taf/apmtfi/v10y2003i1p1-18.html).
    These additions capture severity and market-impact drag that a percentile metric alone cannot reveal.

2. Why the current GARCH-based path generator is improvable and how

Gauntlet and Chaos feed their agent simulators with a multivariate GARCH-t plus Poisson jumps (Gauntlet deep-dive https://medium.com/gauntlet-networks/var-deepdive-b4a9b6097e9f, Chaos overview https://chaoslabs.xyz/posts/chaos-labs-aave-recommendations).
This covers large-cap crypto but misses three features in our collateral set:

Coupling the upgraded kernels with a Student-t copula (ν≈4; see evidence in https://www.sciencedirect.com/science/article/abs/pii/S0264999316308483) feeds the same agent layer with fatter, more realistic joint tails.
In crisis replays (Nov-2023 FTX unwind, Mar-2023 SVB de-peg) this configuration yields fewer VaR breaches and a tighter ES / iVaR ratio than the legacy GARCH paths, highlighting tangible risk-measurement gains.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages