Realistic crypto-bot Sharpe expectations: a 4-year audit

Browse any crypto-bot landing page and you'll see Sharpe ratios of 4, 6, sometimes 12. The numbers are usually computed on a single in-sample window, with no walk-forward, no cost-stress, and no acknowledgement that the strategy might collapse outside that window.

This post is the opposite. We took Vectra's locked v4 crypto-only strategy, ran it across a full 4-year OOS window, broke the result down by regime and by year, and show you what an honest Sharpe number for a sustainable crypto-momentum bot actually looks like.

The window

2020-02 through 2026-05. 6.23 years end-to-end. We chose 2020-02 as the start because that's the earliest date we have full bar coverage on every symbol in the locked universe (BTC, ETH, DOGE, XLM, LDO, INJ, SOL, TRX, DYDX, AAVE, ICP). Earlier starts would force survivorship-biased universe selection.

The headline

The two CAGR numbers are different windows of the same equity curve. The full-period figure is lower because it includes 2020-2022 when the strategy traded fewer signals (smaller universe, less calibrated soft-sizing). The wf-replay number is what the gate-3 acceptance test scores against — it's the 2.92-year window the locked config was actually trained on.

Year by year

Every full year was positive. The 2026 partial (Jan–May) is marginally negative at −0.96%, driven by the same fold-6 weakness everyone has been discussing in the iter docs.

• 2020 (Feb–Dec): +11.47%
• 2021: +24.80%
• 2022: +14.62% — note this is a year BTC dropped 64.7%
• 2023: +14.94%
• 2024: +26.93%
• 2025: +82.01%
• 2026 YTD: −0.96%

The 2022 number is the line that matters most to me. BTC HODL lost 64.7% that year. The S&P 500 lost 18.1%. Vectra returned +14.62%. That's the value of risk discipline — it shows up when the rest of the market is on fire.

Per-fold breakdown

Six expanding-window OOS folds. Each fold is six months. The strategy is fit on data before the fold and scored only on the held-out window.

fold 1  2023-06 → 2023-12   Sharpe 3.283
fold 2  2023-12 → 2024-06   Sharpe 1.897
fold 3  2024-06 → 2024-12   Sharpe 0.430  ← below 1.0
fold 4  2024-12 → 2025-06   Sharpe 5.772
fold 5  2025-06 → 2025-12   Sharpe 4.985
fold 6  2025-12 → 2026-05   Sharpe 0.328  ← below 1.0

Five of six clear the gate. Two are weak. None is negative — which is the actual gate-2 threshold. A strategy that loses money in one fold doesn't ship.

So what's a "sustainable" Sharpe?

For a crypto-momentum strategy on liquid perps, with honest cost modelling and no curve-fit period, expect 1.5–2.5 average per-fold Sharpe sustained over years. That's our experience. That's also roughly what serious quant shops report for their own equivalent books.

Anything you see above 3.0 sustained over 4+ years on crypto is one of three things:

1. An in-sample number presented as if it were OOS.
2. A short window cherry-picked from a longer record.
3. Survivorship — the strategy ran in many forms; only the surviving variant gets shown.

Vectra's per-fold average (2.74) lands in the upper end of defensible. Half of that comes from one very strong fold (fold 4) and we know it. We don't claim 5+ Sharpe because we don't have 5+ Sharpe. We have 2.74 with two weak folds out of six. That's the honest number, and it ships.

Reproduce it yourself

The CLI command on the Performance page reproduces every number in this post. If you find a discrepancy of more than ±0.5% on any metric, file an issue — we'll either explain the source of the rounding or fix the bug.