Tezoro aggregates 2,000+ vaults and always keeps your funds in the top-performing ones.
Tezoro applies quantitative finance methodology to identify top-performing lending markets before yield shifts become visible to the broader market. Using Exponential Moving Average over 81,000+ on-chain data points, the ranking model achieves a Spearman ρ of 0.527 — 52% above the naive real-time APY baseline and well into the "large effect" range by Cohen's conventions for financial time series.
Lending market rates are continuously repriced by on-chain utilization, liquidity inflows, and rate curve mechanics. Across 18 USDC markets tracked hourly over 6.3 months, no protocol held rate leadership for sustained periods — top positions rotated constantly, with individual markets crossing each other hundreds of times. A static strategy anchored to the current highest APY is structurally exposed to rate reversals within hours of allocation.
The EMA halflife parameter governs how quickly historical observations lose influence. Too short, and the model overreacts to noise; too long, and it fails to detect genuine rate regime shifts. To identify the optimal value, we ran a grid search across halflife values from 2 to 48 hours, evaluating each against Spearman rank correlation and Top-5 Overlap simultaneously. Both metrics peak at halflife = 24 hours and remain stable across the 16–48h range. This plateau is a critical validation: it confirms the result is structurally robust and not an artifact of parameter overfitting to the evaluation window.
Evaluated against two reference baselines — random market selection (ρ = 0.000) and naive best-current-APY allocation (ρ = 0.347) — EMA at halflife = 24 achieves a Spearman rank correlation of 0.527. In quantitative finance, rank correlations above 0.3 are considered actionable signals; crossing 0.5 places the result in the top quartile of the theoretical range and above Cohen's "large effect" threshold. The 52% improvement over the naive baseline compounds across 4,500+ hourly rebalancing cycles, translating directly into higher realized yield for depositors.
The hexbin density plot maps EMA-predicted market ranks against realized ranks across all 60,970 market-hour observations from 14 active markets. In the EMA panel, data concentrates along the diagonal — markets ranked highest by the model are, with statistically significant frequency, the markets that actually deliver the highest yield the following hour. In the random baseline panel, no diagonal structure exists: predictions are uniformly distributed, carrying zero information. The contrast is visible at the population level, not merely in aggregate summary statistics — making the model's edge empirically verifiable and reproducible.
Tezoro applies Exponential Moving Average to hourly on-chain yield data, assigning exponentially decaying weights to older observations. Unlike a simple moving average, EMA responds to recent regime changes without lag accumulation, while suppressing short-lived rate spikes that carry no predictive information. Ranking markets by EMA score rather than raw APY converts a high-noise real-time signal into a structured, forward-looking ordinal indicator — the foundation of the allocation model.
Tezoro applies quantitative finance methodology to identify top-performing lending markets before yield shifts become visible to the broader market. Using Exponential Moving Average over 81,000+ on-chain data points, the ranking model achieves a Spearman ρ of 0.527 — 52% above the naive real-time APY baseline and well into the "large effect" range by Cohen's conventions for financial time series.
Lending market rates are continuously repriced by on-chain utilization, liquidity inflows, and rate curve mechanics. Across 18 USDC markets tracked hourly over 6.3 months, no protocol held rate leadership for sustained periods — top positions rotated constantly, with individual markets crossing each other hundreds of times. A static strategy anchored to the current highest APY is structurally exposed to rate reversals within hours of allocation.
The EMA halflife parameter governs how quickly historical observations lose influence. Too short, and the model overreacts to noise; too long, and it fails to detect genuine rate regime shifts. To identify the optimal value, we ran a grid search across halflife values from 2 to 48 hours, evaluating each against Spearman rank correlation and Top-5 Overlap simultaneously. Both metrics peak at halflife = 24 hours and remain stable across the 16–48h range. This plateau is a critical validation: it confirms the result is structurally robust and not an artifact of parameter overfitting to the evaluation window.
Evaluated against two reference baselines — random market selection (ρ = 0.000) and naive best-current-APY allocation (ρ = 0.347) — EMA at halflife = 24 achieves a Spearman rank correlation of 0.527. In quantitative finance, rank correlations above 0.3 are considered actionable signals; crossing 0.5 places the result in the top quartile of the theoretical range and above Cohen's "large effect" threshold. The 52% improvement over the naive baseline compounds across 4,500+ hourly rebalancing cycles, translating directly into higher realized yield for depositors.
The hexbin density plot maps EMA-predicted market ranks against realized ranks across all 60,970 market-hour observations from 14 active markets. In the EMA panel, data concentrates along the diagonal — markets ranked highest by the model are, with statistically significant frequency, the markets that actually deliver the highest yield the following hour. In the random baseline panel, no diagonal structure exists: predictions are uniformly distributed, carrying zero information. The contrast is visible at the population level, not merely in aggregate summary statistics — making the model's edge empirically verifiable and reproducible.
Tezoro applies Exponential Moving Average to hourly on-chain yield data, assigning exponentially decaying weights to older observations. Unlike a simple moving average, EMA responds to recent regime changes without lag accumulation, while suppressing short-lived rate spikes that carry no predictive information. Ranking markets by EMA score rather than raw APY converts a high-noise real-time signal into a structured, forward-looking ordinal indicator — the foundation of the allocation model.
Tezoro is a non-custodial DeFi protocol that aggregates vaults and protocols and rebalances your assets to maximize APY at your chosen risk level.
Tezoro supports 2,000+ markets on Ethereum, including lending protocols, vaults and yield strategies. New markets are added only after they meet our safety and performance checks.
Yes. You can withdraw any time - there are no lockups or fixed withdrawal windows. Tezoro initiates withdrawals immediately, but actual execution can be temporarily delayed or partially filled if an external protocol has high utilization.
Yes. Tezoro covers gas costs for automatic rebalances. You only pay gas when you deposit and when you withdraw. The deposit/withdrawal gas costs are low - under $1 as of March 2026.
Yes. Yield is auto-compounded: earned yield is automatically reinvested, and rewards are converted into the deposited asset and reinvested - no manual steps required.
Yes. When you deposit, you receive Tezoro "t" tokens (for example, tUSDC if you supply USDC) that represent your pro-rata share of the vault. Tezoro tokens are yield-bearing: their exchange rate increases as the vault earns yield.
You can see t-tokens in your web3 wallet and transfer them. Tezoro tokens are redeemable for the underlying asset per the vault rules. We plan to add the ability to use t-tokens as a collateral in other DeFi protocols in the future.
Tezoro uses layered controls:
No. Tezoro is a technology platform that provides algorithmic allocation. We are not a fiduciary, asset manager, or financial adviser and do not provide personalized investment advice.
No. Tezoro is fully non-custodial. Tezoro's contracts can only algorithmically distribute your assets among whitelisted vaults and protocols within the risk tier you selected - they cannot access your private keys or move funds outside those permissions.
Tezoro charges a 15% performance fee on earned yield (profits only). Fees are collected on-chain weekly and are also realized at withdrawal according to the vault rules.