walk-forward backtest · may 2023 – mar 2026
The backtest, in full.
v018 is a market-neutral long/short crypto basket. Below is the complete evidence — the equity curve, how it behaved when the market crashed, whether it survived out-of-sample, and the limits we won't hide. Every figure is computed from the strategy's real return series against buy-and-hold BTC.
the curve
$1 became $9.42 — uncorrelated to the market
Cumulative return through the walk-forward backtest, against buy-and-hold Bitcoin on the same capital. Drawdowns are shown — nothing is smoothed.
the differentiator
Market-neutral — even when the market breaks
The test most “market-neutral” strategies quietly fail: does correlation to the market spike in a crash? v018's does not. Each dot is one day — the cloud is flat, not diagonal.
each dot = one day · v018 vs BTC
A diagonal cloud would mean “it's just leveraged Bitcoin.” v018's is flat — its daily moves are independent of the market's.
on the market's worst days
On the 23 days BTC fell more than 5%, v018 averaged +0.45%.
not overfit
It survived its own holdout
The regime clusters were selected on training data only; the last ~5 months were held out. Overfit strategies collapse out-of-sample. v018's Sharpe held — it even rose.
Out-of-sample Sharpe held (+3.12 vs +2.67) rather than collapsing — the signature difference between a real edge and an overfit one. This clean holdout is short (~5.2 months); over the full 13-month out-of-sample period Sharpe was +2.59 (vs +2.72 in-sample), so we treat it as encouraging, not conclusive.
every regime
Positive every year — including the years BTC wasn't
Annual return, v018 vs buy-and-hold BTC. 2025 and 2026 are the tell: v018 kept compounding while Bitcoin went sideways-to-down.
the benchmark
Beats buy-and-hold Bitcoin on all three axes
Return, risk-adjusted return, and drawdown — over the same window, on the same capital.
cumulative return
daily-equity Sharpe
max drawdown
how it was tested · and what it can't yet claim
Validated honestly — limits included
The transparency is the asset. A sophisticated buyer should be able to reproduce every number from the conveyed code.
method
- ✓Walk-forward with a held-out fold; clustering fit on train data only (causal — no look-ahead).
- ✓Execution cost modeled at 15bp per leg — above typical live cost on majors.
- ✓Returns are broad, not lucky: 64% of weeks positive; no single week carries the record.
- ✓Genuinely market-neutral: beta −0.02, correlation -0.05 to BTC.
- ✓Robust, not a knife-edge fit: the edge holds across 25+ parameter perturbations, 10 seeds, and 21 rolling out-of-sample windows, with Sharpe staying in a +2.3 to +2.7 band.
honest limits
- •The cleanest out-of-sample window is ~5.2 months — promising, but a single fold, not a long live record.
- •No pre-2023 (prolonged-bear) test exists — the coin universe's newest member listed in 2023. Acute crashes are covered; a year-long grind down is untested.
- •Backtested performance is not a guarantee; forward returns are expected to be lower.
a working demonstration
This is the strategy. The research behind it is what's for sale.
A market-neutral strategy built on a physics-based market-regime model — its codebase, the full research, model checkpoints, the validation library, and this platform are available for acquisition.