This is the first of multiple blog posts concerning the addition of collaterals to the Keyring vaults.

Curators seemingly rarely publish their quantitative analysis publicly, but understanding the mechanics behind asset onboarding decisions provides valuable insight into risk management. We hope to document the process of bringing new assets onboard with modelling and market microstructure analysis.

This is an attempt at a more rigorous approach to onboarding new assets - specifically the "LTV Limits" (LLTVs) and "Supply Caps" (Caps).

The core objective is to estimate the price slippage $S$ for a given liquidation volume $L$ which is written as $\mathbb{P}(S \leq s \mid L = \ell)$ and it's conditional probability density $f_{S|L}(s|\ell)$. Then, we will calculate the expected slippage, $\mathbb{E}[S \mid L < X, D]$ where $X$ is a percentile of liquidation volumes $X = F_L^{-1}(0.99 \mid D)$ for a given pool depth $D$, or determined via scenario analysis based on liquidation sizes from other platforms.

The analysis proved more complex than initially anticipated due to numerous unusual market characteristics. Trader Joe is not a well documented trading venue and has several idiosyncrasies that create great benefits for users but also great doom-loops for analytics. Notably, the data revealed instances of large trades that moved markets 6% in favour of traders - a phenomenon that defies conventional market mechanics. The following examples illustrate these data anomalies.

Another anomaly revealed what appears to be probability wave interference effects in the exchange, similar to quantum mechanics (except not in infinite dimensional Hilbert space!). In LFJ they have discretised (quantised) the liquidity bins with specific 10 basis point blocks and the result is a variety of harmonics around the "ground state". This unexpected phenomenon showed interference patterns that deviated dramatically from traditional "bell shaped" return distributions.

This analysis covers wAVAX / USDC, representing the most liquid DEX pair on Avalanche. The dataset encompasses the entire history of swaps until the previous evening, with block, $b$, returns calculated by $r(b) = \text{VWAP}(b) - \text{VWAP}(b-1)$ using Swap events from HyperSync. The exact cause of this strange return distribution remains unclear, but this anomaly alone delayed the wAVAX pool go-to-market by approximately 2-3 days due to the additional analysis required.

To validate the interference theory hypothesis, we modelled a Poisson distribution with triangular convolution of multiple swaps per block causing interference. This approach looked similar to the empirical observations (subject to proper parameter fitting), enabling a resemblance of sanity.

With a robust model of price movements for various sell sizes on the primary venue for wAVAX/USDC trading, stress testing under various market conditions becomes possible.

Kyle's Lambda is a core concept in market microstructure, measuring how trades affect prices in financial markets. It helps estimate execution costs, particularly for high-frequency trading (HFT) in single-state executions. The model uses regression to quantify price changes based on trade volume, expressed as $r_b = \lambda \cdot \sqrt{|V_b|} \cdot \text{sign}(V_b) + \epsilon_b$, where for a given block, $r_b$ represents slippage, $\lambda$ is the price impact coefficient, $V_b$ is the signed trade volume, and $\epsilon_b$ is the random noise term.

The returns were calculated as the change in VWAP between consecutive transactions $r_{x} = \text{VWAP}_x - \text{VWAP}_{x-1}$ where $x$ denotes the transaction index.

The analysis revealed that atomic liquidations can be performed with minimal slippage on collateral assets. While general collateral degradation may occur, the likelihood of generating bad debt through the liquidation process itself remains very low.

The data demonstrates that liquidating $40,000 worth of $wAVAX from an unhealthy position typically results in only $12 slippage on the AMM. This finding proved surprising and requires further individual transaction analysis, though the data seems to consistently support this conclusion.

The bin-model architecture of the LFJ exchange appears responsible for this favourable slippage profile. Analysis uncovered instances of $600,000 orders executing with zero slippage. While LFJ's pricing can appear confusing to users due to dust-related price distortions that may not reflect actual liquidity, the Volume Weighted Average Price (VWAP) typically provides fair execution.

In the next post we will dive into liquidity data and analyze liquidation sizes across DeFi vaults. We’ll explore Just-In-Time (JIT) liquidity patterns, testing the hypothesis of JIT activity and then analyse the liquidation sizes for various DeFi vaults.