Asset Clustering

Assets Clustering is used to tailor risk parameters for different assets, particularly exotic long-tail assets that may lack extensive historical data. By grouping assets with similar volatility and liquidity characteristics, BasisOS can apply more customized risk controls.

Technical Details:

• Methodology: A k-Nearest Neighbors (k-NN) clustering algorithm is applied to historical candlestick data (e.g., from Binance).

• Key Metrics:

  • $Q(5m)_{0.99}$​: The 99th percentile of the 5-minute price change, indicating extreme short-term volatility.

  • Average Volatility $(\bar{v}(5m))$: Computed from the high, low, and open prices.

  • Additional Metrics: Similar measures for 15-minute candles and observed maximum leverage levels (e.g., from Hyperliquid) are also considered.

• Purpose: By grouping assets with similar risk profiles, the protocol can assign more precise risk parameters. This is especially important for managing assets like memecoins, where conventional risk models may not be directly applicable.