Paste or upload a returns series, set a window length and step, pick rolling or anchored, and see in-sample vs out-of-sample Sharpe per window. The mean degradation and passing-window count show whether your settings survive more than the regime you fitted them on.
One return per period (daily assumed). Decimal or % accepted. Sample below has a regime shift mid-series.
IS and OOS each use this length in rolling mode; OOS only in anchored.
How far the window slides between iterations.
A rough robustness floor: 60% or more passing is a useful signal; under 40% is fragile.
| every (Days between re-picks) | lookback days (Days looked back) | Sharpe |
|---|---|---|
| 7 | 60 | 0.85 |
| 14 | 60 | 0.60 |
| 30 | 60 | 0.69 |
| 60 | 60 | 0.75 |
| 7 | 90 | 0.96 |
| 14 | 90 | 1.01 |
| 30 | 90 | 0.72 |
| 60 | 90 | 0.86 |
| 7 | 180 | 0.88 |
| 14 | 180 | 0.85 |
| 30 | 180 | 0.86 |
| 60 | 180 | 0.50 |
| 7 | 365 | 0.054 |
| 14 | 365 | 0.17 |
| 30 | 365 | −0.13 |
| 60 | 365 | 0.016 |
Keel's study of Momentum Strategy on Hyperliquid · computed Nov 16, 2026 · data Aug 15, 2024 to Sep 29, 2026
The visualizer runs the canonical walk-forward split: it slices your returns series into IS/OOS window pairs and computes annualized Sharpe on each side. In rolling mode the IS window is fixed-size and slides forward by the step value; in anchored mode the IS window starts at the first observation and expands with each step.
# Per-window annualized Sharpe
sharpe = (mean(returns) / stddev(returns)) * sqrt(365)
# Per-window degradation
degradation = (oos_sharpe - is_sharpe) / abs(is_sharpe)
# Aggregate "passing" rule
passing = count(windows where oos_sharpe > 0.5 * is_sharpe and is_sharpe > 0)The annualization factor (sqrt(365), daily bars of a market that trades every day) is the same on both sides of the degradation ratio, so it cancels out for IS-vs-OOS comparison. The visualizer is built for comparison, not absolute-number reporting; for absolute Sharpe with the right periodicity use the dedicated Sortino vs Sharpe calculator.
Walk-forward is a structural defense against parameter overfit, not a substitute for out-of-sample testing on truly fresh data after deployment. Strategies that survive walk-forward can still degrade live if the underlying market dynamics shift outside historical experience. For window choices, a worked table from this tool’s own sample series and the loop in Keel, read the walk-forward optimization explainer.
Keel is a strategy builder for Hyperliquid. Describe an idea, backtest it with fees, slippage and funding, and run it live on Hyperliquid’s crypto perps. Start free, and connect a wallet when you’re ready to trade.
It splits a returns series into rolling or anchored in-sample / out-of-sample windows, computes annualized Sharpe on each side, and surfaces the degradation. The point is to make parameter overfit visible: a strategy whose OOS Sharpe collapses across most windows is fragile, even if its single-shot backtest looked great.
The visualizer is for ratio comparison: IS vs OOS within the same series. The annualization factor cancels out of the degradation calculation, so a daily default (365 days, since crypto trades every day) works fine for visual inspection even if your series is hourly or weekly. For absolute Sharpe values you trust, use the Sortino vs Sharpe calculator with the correct periodicity.
Rolling drops the oldest in-sample bar each step, keeping the IS window fixed-size. Anchored keeps everything from the start of the series and lets IS grow. Anchored is more sample-efficient (more data per fit); rolling is more honest about regime drift (old data gets discarded). When in doubt, run both and compare degradation.
A pragmatic robustness floor: count the windows where OOS Sharpe > 0.5 × IS Sharpe (and IS Sharpe was positive). 60%+ passing is a useful signal that the strategy parameters generalize; under 40% suggests the IS performance was sample-specific. It is a heuristic, not a formal test. Pair it with Monte Carlo or PBO for stricter overfitting diagnostics.
This visualizer is a standalone browser tool: no Keel account needed, nothing uploaded, computed locally. Keel's Research re-picks a strategy's settings as it goes, using only the days before each pick (its Check a strategy step "Re-picking the settings as you go"), as in Keel's study on this page. For a windowed loop of your own, Keel runs a backtest over any date window (in the app, or `keel backtest run <strategy> --start-date … --end-date …`) and exports each run's trades as CSV; paste the stitched out-of-sample returns here.
The guide: how walk-forward works, window choices, a worked table from this tool, and how to drive the loop on Keel.
IS/OOS sizing for short market histories and rolling vs anchored on regime-shifty HL data.
Compute Sharpe and Sortino side-by-side with the right periodicity. Pairs with the visualizer for full per-window analysis.