Traditional yield curves rely on opaque, centralised data. As DeFi grows, the onchain economy needs transparent, verifiable term structures derived directly from onchain fixed-rate markets.
A yield curve shows the relationship between interest rates and time to maturity. Traditional finance uses government bonds, we use onchain fixed-rate markets from Pendle Finance.
The curve answers: "What yield can I lock in today for 30 days? 90 days? 1 year?"
We gather fixed-rate yield data from Pendle v2 markets across multiple blockchains:
• Multi-chain coverage: Ethereum, Arbitrum, BSC, Optimism, Base, Mantle
• USD-denominated only: Markets with USDC, USDT, or USD-pegged assets
• Minimum TVL filter: ≥$10M liquidity per market
• Maturity range: At least 1 day remaining to expiration
Why Pendle? Pendle is the largest onchain fixed-rate protocol, offering transparent, verifiable yield data directly from smart contracts. No intermediaries, no estimates. Just raw blockchain data.
Rather than fitting a global curve through all data points, we use a localised estimation approach. For each target tenor (7d, 14d, 30d, etc.), we:
1. Calculate how far each market is from the target tenor
2. Apply a Gaussian kernel to weight markets by proximity
3. Markets closer to the target tenor receive higher weight
4. Markets farther away receive exponentially lower weight
Where d is the distance from market maturity to target tenor, h is the bandwidth parameter controlling smoothness, and K(d) is the kernel weight from 0 to 1.
The Gaussian kernel tells us which markets to trust based on distance, but we also need to account for market size and liquidity. We combine both:
Key insight: A large market far away contributes less than a small market nearby. This prevents extrapolation and keeps estimates locally grounded in actual market data.
Before computing a yield estimate for a tenor, we verify it has sufficient market support. A tenor point is only published if it passes all three checks:
Why strict rules? We'd rather return null than publish unreliable estimates. Institutional users need to trust that published yields have solid market backing.
For tenors that pass all support checks, we calculate the yield estimate as a weighted average of nearby market yields:
Where yτ is the estimated yield at tenor τ, Wi is the combined weight for market i (TVL × kernel), and yi is the fixed APY from market i.
Non-parametric approach: Unlike traditional curve models (Nelson-Siegel, cubic splines), our method makes no assumptions about the curve's functional form. The shape emerges entirely from market data.