Factor Research

Alpha factors on Hyperliquid

An alpha factor ranks the Hyperliquid cross-section by expected forward return. This page explains how to judge one on its own — information coefficient, decay, half-life, quantile spread — and how Keel's 262 registered components express it inside a strategy you can backtest and run live.

By Keel Research Team · Updated May 18, 2026

Toolkit, not catalog — and why that matters

Factor libraries come in two shapes. Curated catalogs publish a fixed list of factors with stamped IC, ICIR, half-life, and turnover numbers — usually on one specific universe and cost model. Composable toolkits give you the building blocks so you can build the factor yourself and measure it on your own setup. Keel's component registry is the building-block shape: signals, transforms, regime detectors and aggregators that you compose into a strategy and backtest. It does not stamp IC numbers on them, and that is worth understanding: IC and decay numbers depend on universe definition, cost model, lookback window, and rebalance horizon — a single stamped number reads cleanly but only tells the truth on one specific setup.

The 262 components, by category

The component registry has 262 entries decorated with @register_component. They break down roughly as follows.

CategoryExamples
Momentum signalsCross-sectional momentum at multiple lookbacks; time-series momentum; risk-adjusted momentum.
Trend signalsEMA/SMA crossovers, MACD-style, channel breakouts, trend strength scoring.
Volatility signalsRealized vol, vol-of-vol, vol regime indicators, Garman-Klass, ATR-based.
Volume / liquidity signalsVolume rank, dollar volume, volume z-score, OI-weighted scoring.
Statistical signalsZ-score and percentile rank transforms used as direct signals.
Logic signalsBreakout detectors, mean-reversion triggers, threshold gates.
Regime detectorsFundingLevelRegime, FundingDispersionRegime, plus vol-, trend-, and correlation-based regimes.
Portfolio aggregators (~20)Forecast weighting, rank normalization, sign-split long/short, equal-weight, inverse-vol.
Transforms (~16)Rolling normalization, smoothing, lag, decay, regime gates and scalers, signal-combiners.

The remainder are data loaders, risk overlays (vol targeting, position caps, gross and net leverage limits), execution operators (buffered rebalancer, neutralization), and utility components that wire pipelines together. Every one is composable — you can stack them in arbitrary order, branch them, and store intermediate values for inspection.

Evaluating a factor

Five diagnostics decide whether a factor deserves a place in a strategy. Keel computes rank IC by horizon on a signal in its Research, with ranking skill through time, top and bottom fifths, and turnover and coverage, as in Keel's study on this page; compute any other diagnostic wherever your factor values live. Read them before any backtest, and compute them on exactly the values the strategy will trade, or you are measuring a different signal.

  • Information coefficient (Pearson and Spearman). Per-bar cross-sectional correlation between factor and forward return at a chosen horizon. Spearman default.
  • Rolling IC and ICIR-equivalent stability. Sliding-window mean and standard deviation of IC. Sharpe-style ratio. Distinguishes truly informative factors from lucky ones.
  • Half-life and decay curve. Bars until IC halves; full IC-by-horizon plot. Sets the natural rebalance horizon and reveals whether you have a fast or slow signal.
  • Quantile spread. Forward return of top decile minus bottom decile. Tests monotonicity — many high-IC factors lack a clean monotone payoff, which is a warning sign.
  • Turnover and cost sensitivity. What fraction of the cross-section changes deciles per bar, and what the quantile spread looks like net of realistic fees and slippage. The honest filter that kills the prettiest paper factors.

Hyperliquid-tuned defaults

The components ship with defaults appropriate for HL perp data: price bars from 5 minutes up off a 1-minute grid, 1-hour settled funding, ~200 markets, fee schedule from the HL maker/taker tiers. Funding is a first-class input, not a footnote. Data loaders standardize units across the universe so cross-sectional rankings are comparing apples to apples. The regime detectors are conditioned on HL-specific regime structure, not transplanted from equity research.

The implication: a momentum or carry factor you build with Keel components is HL-shaped by default. You can change the defaults for a different venue if you ever leave HL — but Keel is built for the case where you do not.

Reading someone else's factor numbers

When a catalog does publish IC, ICIR, half-life, decay curve and turnover per factor, ask four questions before trusting it: which universe (and with what minimum history), which cost model, which forward horizon, and which sample period. A momentum factor stamped on a top-20 universe during a trending year tells you little about a top-100 universe in a choppy one. Published factor diagnostics like these may come to Keel in the future; wherever you read them, the four questions come first.

Try it

Browse the 262 components, compose a factor into a strategy, and backtest it against the HL universe with fees, slippage and hourly funding.

How well it ranks, 1 to 30 days ahead
How well it ranks, 1 to 30 days aheadRank IC over 9 horizon (Days ahead) values.−0.0450.00780.060−0.0036−0.00010.00200.0130.0150.0170.013−0.00370.00291235710142030
The numbers
horizon (Days ahead)Rank ICt-statisticDaysEffective daysCoveragedays
1−0.0036 (−0.0214 to 0.0141)−0.4074274262.2%742
2−0.0001 (−0.0239 to 0.0236)−0.01173836961.8%738
30.0020 (−0.0259 to 0.0299)0.1473424461.4%734
50.013 (−0.021 to 0.048)0.7572614560.6%726
70.015 (−0.026 to 0.056)0.7371810259.8%718
100.017 (−0.028 to 0.061)0.757067058.6%706
140.013 (−0.033 to 0.058)0.566904957.0%690
20−0.0037 (−0.0456 to 0.0383)−0.186663354.6%666
300.0029 (−0.0416 to 0.0473)0.136262050.7%626

Keel's study of 28-day momentum · computed Nov 17, 2026 · data Aug 15, 2024 to Sep 29, 2026

How much they move together
How much they move togetherRank correlation for 10 cells, result b (Compared with) by result (Result); colour from −0.43 to 0.43.0.43−0.14−0.110.21−0.31−0.0950.0800.010−0.024−0.247-day momentum1-day reversal7-day funding carrySize, by 30-day dollar volume28-day momentum1-day reversalColour · Rank correlation · −0.43+0.43result b (Compared with) ↓result (Result) →
The numbers
result (Result)result b (Compared with)Rank correlationShared daysTop overlapTop overlap by chanceSkill correlationshared days
28-day momentum7-day momentum0.4374749.8%19.8%0.39 (0.30 to 0.47)747
28-day momentum1-day reversal−0.1474723.8%19.8%−0.062 (−0.148 to 0.028)747
28-day momentum7-day funding carry−0.1174719.9%19.8%−0.072 (−0.153 to 0.010)747
28-day momentumSize, by 30-day dollar volume0.2165330.9%19.9%0.29 (0.19 to 0.39)653
7-day momentum1-day reversal−0.3176817.5%19.8%−0.28 (−0.35 to −0.21)768
7-day momentum7-day funding carry−0.09576819.2%19.8%−0.057 (−0.135 to 0.023)768
7-day momentumSize, by 30-day dollar volume0.08065325.1%19.9%0.059 (−0.034 to 0.152)653
1-day reversal7-day funding carry0.01076921.4%19.8%−0.054 (−0.127 to 0.017)769
1-day reversalSize, by 30-day dollar volume−0.02465319.7%19.9%0.076 (0.007 to 0.147)653
7-day funding carrySize, by 30-day dollar volume−0.2465311.2%19.9%−0.28 (−0.36 to −0.20)653

Keel's study of 28-day momentum · computed Nov 17, 2026 · data Aug 15, 2024 to Sep 29, 2026

FAQ

Common questions

What is an alpha factor?

A numerical score, computed per asset per bar, that ranks the cross-section by expected forward return. Examples: 1-week price momentum, current funding rate, open-interest delta over 12 hours, a z-scored basis between funding and realized volatility. The factor itself is a vector across the universe; the strategy is the wrapping that turns the rank into positions. In Keel, factors are pipeline components — a signal class that consumes data loaders and emits a cross-sectional score.

How is information coefficient computed?

For each bar, compute the rank correlation (Spearman by default) between the factor score across the universe and the realized forward return over a chosen horizon. Average across the sample for the headline IC. Roll across a window for the rolling-IC plot. Mean IC over standard deviation of IC gives the ICIR-equivalent stability metric. Whatever tool computes them, run them on exactly the factor values the strategy will trade, so the number you judge is the number you deploy.

How does this differ from QuantInsti or Hudson & Thames factor libraries?

Different scope. QuantInsti is an educational platform; Hudson & Thames mlfinlab is a Python library of academic implementations. Keel is HL-native infrastructure: the components are tuned to perp trading on Hyperliquid (1-hour settled funding, price bars from 5 minutes up, real fee tiers), and the pipeline you backtest is the one that runs live. Keel is opinionated about the venue but composable about the components — the 262 parts are building blocks, not a curated catalog with stamped IC per factor.

Can I add custom factors?

As compositions, yes. A custom factor in Keel is a combination of registered components — a signal, the transforms that normalize or smooth it, a regime gate or scaler — written in Keel's strategy DSL and backtested like any other strategy. The platform runs registered components only, so a factor that needs a genuinely new primitive is a new component, not a script you upload.

Does Keel publish IC numbers per factor?

Not as a stamped table. IC, ICIR, half-life and turnover numbers depend on universe definition, cost model, lookback window and rebalance horizon, and a single published set can mislead more than it informs. Keel ships the components and a backtester that measures a whole strategy over any window with fees, slippage and funding. Keel computes rank IC by horizon on a signal in its Research, on the window and coins each study states, and publishes its own studies of some signals, such as the one on this page.