Holding and hoping means taking every crash at full size. Every day we publish a quantified read on Bitcoin's regime — a 0–100% exposure reading, from a model tested on years of data it had never seen. It's research; what you do with it is yours to decide.
Simulated / out-of-sample, net of costs · research, not advice · past performance does not guarantee future results.
You didn't get hurt being wrong about Bitcoin — you got hurt holding the same size through everything. You took the full drawdown down, then, too burned to re-enter, watched the recovery without you. One mistake, made twice — a sizing problem, not a conviction one.
just to get back to even — not to profit. Years of waiting most people don't survive.
to recover a −11.2% drawdown — versus +200% after a −66.7% one. Smaller holes are easier to hold through; past behaviour doesn't guarantee the future.
No single indicator survives every market. Three independent models, each checking the others, resolve to one 0–100% exposure reading.
Reads where Bitcoin sits in its long-horizon cycle and filters out noise. The slow, steady hand.
The most influential model. Detects regime transitions as they develop — before the drawdown fully lands.
Built to disagree with the Tactical model when the evidence is thin. When it contradicts, confidence drops and exposure eases — a brake on overreaction.
A hard rule that overrides the models: if Bitcoin's drawdown breaches a fixed threshold while the ensemble still reads Bull, exposure is automatically halved — a seatbelt for when the data is lying and price is telling the truth.
Every layer is run through dedicated stress tests, and outputs are archived daily — an immutable, version-controlled record, not a screenshot.
Powered by a multi-source data pipeline covering on-chain, derivatives, and cross-asset feeds.
Two windows the model was frozen before it saw: a nested walk-forward (2022–2024) and a blind holdout (2025–2026) it had never touched — both well above buy-and-hold.
Holdout 2025-01-01 to 2026-05-31 — current production model (net of 5bp execution cost)
Current production model · CAGR is annualised, not cumulative.
| Metric | Vantegio | BTC B&H |
|---|---|---|
| CAGR | +30.7% | −16.3% |
| Max drawdown | −16.4% | −49.5% |
| Calmar ratio | 1.87 | −0.33 |
| Sharpe ratio | 1.21 | −0.18 |
Nested walk-forward OOS 2022-01-01 to 2024-12-31 — current production model (net of 5bp execution cost)
Current production model · CAGR is annualised, not cumulative.
| Metric | Vantegio | BTC B&H |
|---|---|---|
| CAGR | +76.4% | +25.6% |
| Max drawdown | −11.2% | −66.7% |
| Calmar ratio | 6.83 | 0.38 |
| Sharpe ratio | 2.15 | 0.69 |
| Avg. BTC exposure | 36% | 100% |
The model was sealed before this window opened and untouched since. A blind holdout reduces — but doesn't eliminate — overfitting. Not a promise; a record of what the model output when it couldn't know what was coming. Past performance does not guarantee future results.
On the walk-forward: Calmar 6.83 vs 0.38 for buy-and-hold. A −66.7% crater needs +200% to recover; the model's −11.2% needs +13%.
Is the edge just Bitcoin exposure? Held at the model's own average exposure, passive BTC lost −4.9% on the holdout while the model made +30.7% (beta 0.32–0.40). The return is timing, not participation.
De-risking shows in the shape, not just the average: holdout skew +2.2 — 46% of BTC's up moves, only 35% of the down. A 5,000-path Monte-Carlo puts a losing 17-month stretch at 5%.
Timing-sensitive: within ~2h of the daily brief the holdout keeps a Sharpe of 0.96 (of 1.13); a 4–6h delay roughly halves it. The brief publishes early each day so a prompt subscriber can act in that window.
The figures above assume idle cash earns 0%. Because the model is defensive it holds a large cash share — BTC exposure averaged roughly a third across the test windows. For completeness, the table below shows the model's economically-complete result if that idle capital were instead carried at the prevailing risk-free rate — the 3-month U.S. Treasury bill — for each date in the window.
| Test window · CAGR | Idle cash 0% | Idle cash at risk-free rate |
|---|---|---|
| Blind holdout 2025-01 → 2026-05 | +30.7% | +34.0% |
| Nested walk-forward OOS 2022 → 2024 | +76.4% | +80.4% |
Risk-adjusted metrics also rise modestly (e.g. holdout Calmar 1.87 → 2.08), since risk-free carry adds return with negligible added volatility.
Illustrative and hypothetical. This measures the model's economically-complete return under a stated benchmark assumption — it is not subscriber returns and not a projection. The risk-free rate (3-month U.S. Treasury bill, published rate for each date) is used purely as a neutral accounting benchmark; it is not a recommendation to hold cash in any particular instrument, and Vantegio neither provides nor controls any yield on cash. Simulated results; past performance does not guarantee future results.
No dashboard to babysit. Each day a concise brief lands in a private Telegram channel — regime, exposure reading, confidence, and the full model breakdown.
Anyone can post a green curve. The question a serious person asks is how was it produced — and that's exactly where most signal services fall apart.
The core ensemble has run live since launch; a macro risk-appetite overlay has since passed the holdout gate and been folded in — a validated improvement, not a patch. New layers ship automatically.
We've rejected more strategies than we've deployed. Anything that fails a blind holdout is archived, not shipped — you only see what earned its place.
No card is charged during the trial. Cancel in one message and you're never billed. If you stay, this is what it costs.
Fair question — the category earned it. So we built the opposite: models frozen before each test window, a blind holdout the models had never seen, every model vote published daily, an immutable timestamped archive that can't be revised after the fact, and figures reproducible from a version-tagged release. You don't have to take our word for anything — the record is designed to be checked. We never touch your funds, and this is general research, not advice.
Two independent windows. First, a walk-forward OOS (2022–2024) where the model was frozen before the window opened and run forward unchanged, no refit: +76.4% CAGR, −11.2% max drawdown, Calmar 6.83 at 36% average BTC exposure. Buy-and-hold over the same period: +25.6% CAGR, −66.7% drawdown.
Second, the holdout (2025-01-01 to 2026-05-31) — data the models had genuinely never seen: annualised CAGR +30.7%, Calmar 1.87, −16.4% max drawdown. Bitcoin's annualised return was −16.3% over the same period with a −49.5% drawdown. These are annualised figures, not cumulative returns. Model Calmar 1.87 vs −0.33 for buy-and-hold. A blind holdout reduces but does not eliminate overfitting risk. Past performance does not guarantee future results.
The model targets ~36–49% BTC on average rather than 100%. Less raw exposure, but a −11.2% walk-forward worst case requires just +13% to recover — versus +200% from a −66.7% hole. Calmar — return divided by max drawdown — is one risk-adjusted measure used in this analysis. Model (OOS): 6.83 OOS vs 0.38 for buy-and-hold in this simulation.
Fair question — it's the right one to ask, because a good-looking backtest is trivial to manufacture.
The numbers on this page are walk-forward out-of-sample. The model's parameters were frozen before the test window opened, then run forward across multiple years of data — spanning bull runs, crashes, and everything in between — with no re-fitting, no cherry-picked window, no parameter nudging after the fact. It's the difference between predicting tomorrow and "predicting" yesterday.
It's not a promise about the future — nothing is. A blind holdout reduces but does not eliminate overfitting or selection bias, and results from one period do not predict another.
It doesn't ship. We've tested and rejected more strategies than we've deployed.
The ones that didn't survive a blind holdout test — where the model had genuinely never seen that data — get documented and archived. Each research question receives its own separate holdout window, opened once, and not reused. That discipline reduces, but does not eliminate, selection bias. The performance history shown on this page was produced by the model versions identified in the methodology notes, not by candidates that were tested and rejected.
No. Vantegio publishes quantitative model outputs — regime classifications and model exposure readings — as general research. This is not personalised financial advice and nothing on this channel should be construed as a recommendation to buy, sell, or hold any asset. We are not financial advisers, we don't know your personal situation, and we never touch your funds. You are solely responsible for any investment decisions you make. If you need advice tailored to your circumstances, consult a licensed financial professional.
About once a week on average. Historically the model changes regime classification roughly 55–115 times a year (~80 on average). Most days the output is unchanged — the research brief confirms the existing classification. Subscribers independently decide what, if anything, to do with the research. Built for people with jobs, not a screen-watching tool.
The model output is a standardized 0–100% research variable — it is not tied to any specific platform or implementation. Subscribers who choose to use it may do so on any venue they already use — Kraken, Coinbase, Bitstamp, Binance, or elsewhere. We never custody your funds or connect to your account. Any decision a subscriber makes is their own independent decision, in full control.
Send /billing to the bot and cancel through the Stripe portal — no email, no retention call. Cancellation takes effect at the end of the period you've paid for. Full cancellation details are available on the cancellation page.
Still have a question? Email contact@vantegio.com — we usually reply within one business day.
Annualised CAGR +30.7% vs Bitcoin −16.3% · walk-forward MDD −11.2%, holdout MDD −16.4% · cancel renewal at any time.
Get the daily regime read →The model's daily regime read is posted free on X and Discord — condensed and on a delay. Subscribers get it first thing each morning, in full: the exact exposure, the per-model split, and the forward forecast.