By Keel Research Team · Updated May 12, 2026
A dollar-balanced long/short index — long decentralized AI compute, short L2 majors. The trade thesis: capture the spread between the active 2026 narrative (AI infrastructure) and the prior-cycle narrative (L2 scaling). +61% YTD with Sharpe 4.47 and only -8% max drawdown on the live Keel backtest engine.
Window: Jan 1 — May 1, 2026 (YTD). Equal-weight per leg, dollar-balanced long/short. $10,000 start → $16,073 end.
Verified 2026-05-12 against the live Keel backtest engine on Hyperliquid price data. Past backtest performance is not indicative of future returns.
Bring this index live on Hyperliquid via Keel, or open the Index Builder to tweak weighting, add a short side, or change the asset list.
Opens the Keel strategy editor with the index pre-built. New users sign in first; the config persists through sign-up.
| Ticker | Category | YTD return | Target weight |
|---|---|---|---|
| VVV | — | +454.8% | 12.5% |
| RENDER | AI | +23.9% | 12.5% |
| TAO | L1 | +14.3% | 12.5% |
| IO | AI | -24.7% | 12.5% |
| Ticker | Category | YTD return | Target weight |
|---|---|---|---|
| −POL | L2 / Scaling | -11.5% | 12.5% |
| −ARB | L2 / Scaling | -37.6% | 12.5% |
| −STRK | L2 / Scaling | -52.4% | 12.5% |
| −OP | L2 / Scaling | -58.9% | 12.5% |
Why drawdown is small: AI Compute alone hit -46% max DD. The L/S version capped DD at -8% because the L2 shorts hedged the AI selloff. That’s the structural payoff of L/S vs long-only.
The pipeline branches the OHLCV stream via Parallel. The long leg selects AI Compute assets, computes a baseline forecast, and normalizes to half the target gross leverage. The short leg does the same with L2 assets and a negative forecast (short positions). WeightConcatenator merges both legs into a single dollar-balanced position; gross exposure sums to the configured target leverage.
This is implemented in the same Keel pipeline framework as the long-only indexes — the only structural change is the Parallel branching pattern. Open the Index Builder and toggle “Add short side” on any basket to construct similar rotations.
Beyond AI-vs-L2, other sector-rotation theses worth considering: long HL ecosystem / short L1 incumbents, long privacy / short stablecoin issuers (regulatory hedge), long memes / short DeFi (high-beta vs yield rotation). The framework supports any 2-leg combination.
About Keel: Keel is a Strategy OS for AI-assisted systematic trading on Hyperliquid. Build, backtest, and deploy index strategies — and anything else systematic — in one place. Learn more →
Keel is a Strategy OS for AI-assisted systematic trading on Hyperliquid. Backtest, optimize, and run live strategies across single-stock perps, indices, and crypto majors — realistic fees, slippage, and funding modeled.
Free to start — connect a Hyperliquid wallet when you’re ready to go live.
A dollar-balanced basket with two legs — long the assets you expect to outperform, short the assets you expect to underperform. Each side normalizes to half the target gross leverage, so the combined position is net-neutral (zero broad-crypto beta) but captures the spread between the two narratives.
AI Compute is the dominant 2026 narrative — VVV, RENDER, IO, TAO have ridden the GPU-and-inference thesis. L2 majors (ARB, OP, POL, STRK) are the 2023-2024 narrative that capital has rotated out of — they are down 12-59% YTD with no obvious catalyst. The trade is the divergence between the active narrative and the prior-cycle narrative.
The long leg uses ForecastWeightNormalizer(target_leverage=0.5) and the short leg does the same. Combined gross exposure is 1.0x, net is 0x (dollar-balanced). On any bar, sum of long weights ≈ sum of |short weights|. WeightConcatenator merges the two sides.
AI Compute alone hit -46% drawdown when the basket pulled back. The L/S version had only -8% drawdown because the short side acts as a hedge — when AI Compute sold off, L2s also sold off, so the basket P&L was largely unchanged. That is the structural advantage of L/S vs long-only.
It is dollar-neutral — equal long and short exposure. Not strictly beta-neutral or vol-neutral. AI Compute is higher-beta than L2s right now, so on big down days the trade has positive crypto beta (loses more on long than gains on short). Dollar-neutral is the simple structural framing; more advanced users would build a vol-equalized version.
L2s ripping while AI cools off. That can happen — token unlock schedules, narrative rotation, or a specific protocol catalyst (Arbitrum Stylus mainnet, OP Superchain expansion). The L/S structure caps your downside on each rotation but doesn’t protect against narrative reversal.
Another L/S rotation — counter-narrative privacy long vs new-gen DeFi short. +22% YTD, Sharpe 1.96.
The long leg of this trade as a standalone index. +69% YTD but with a -46% drawdown.
Build your own L/S sector-rotation index — toggle "Add short side" on any preset.