Continuous stress-testing

Keom's R&I team built, researches and maintains an in-house agent-based simulator tasked with continuously stress-testing all user positions. The simulator is also used to perform deep-dives into anomalous lending market events on other protocols as well as to update internal snapshots of user risks taking place on all other major lending markets operating.

The team assesses the market risk of the DeFi lending protocols using a multi-asset agent-based model to simulate ensembles of users subject to price-driven liquidation risk. The model was detailed in a research paper reference below:

A multi-asset, agent-based approach applied to DeFi lending...arXiv.org

Whitepaper

The team ran an extensive amount of simulations on randomly generated portfolio allocations reproducing the risk appetite of Compound users for reference (BTCUSD, ETHUSD, MATICUSD shown below). Users are randomly assigned a portfolio size, starting LTV, and portfolio directionality.

For each pair of liquidation LTV and Liquidation incentive values, the team simulated the performance of 2000 distinct user portfolios across 200 distinct price trajectories. All user portfolios are regenerated at each price run.

• The dots shown on the scatter plot represent pair values at which we begin seeing more than 0.1% of all simulated users becoming undercollateralised.

• The color shading represents the average protocol LTV at the end of the runs (darker=higher).

• Yellow shaded rectangle is the current suggestion for protocol parameters across the assets shown.

Scope:

Every non-pegged asset listed on Keom is subject to rigorous cross-asset undercollateralization risk assessments to fine-tune its protocol parameters. For pegged assets, standard historical price analysis doesn't suffice due to unique depegging risks, so we apply specialized criteria to assess stability.