Testing Beyond the Backtest
A backtest tells you how a strategy did perform on historical data. Forward testing tries to answer a harder question: how might it perform on data it hasn't seen yet?
There's no perfect way to predict the future. But there are rigorous ways to stress-test whether a strategy's edge is real or just a lucky fit to past data. We ran three independent methods, each designed to probe a different weakness:
Monte Carlo Simulation
Takes the strategy's actual historical trades, shuffles them into thousands of random orderings, and asks: "across all possible sequences of these trades, what range of outcomes do I get?" Tests whether the backtest result depends on lucky ordering.
Walk-Forward Analysis
Splits history into sequential chunks and tests whether the edge holds in each new chunk. If a strategy only works in hindsight, walk-forward exposes it — the out-of-sample windows will show degraded results. This tests for overfitting.
Synthetic Price Path Simulation
Generates entirely new, never-before-seen Bitcoin price histories using a statistical model, then runs the full strategy logic against each one. Tests whether the strategy works on market conditions that haven't happened yet.
If History Repeats (In Any Order)
The Monte Carlo takes the strategy's 40 historical trade cycles — each with its return, duration, and position size — and randomly resamples them 10,000 times across multiple time horizons. Think of it as asking: "if I keep getting trades that look like these, but in random order, where do I end up?"
Three different resampling methods were used to make sure the results aren't a fluke of one approach. All three produced consistent results, which is a good sign.
The Probability Table
This is the core output. Each cell represents the percentage of simulations that hit that threshold:
| Metric | 1 Year | 2 Years | 3 Years | 5 Years |
|---|---|---|---|---|
| Median Return | 43% | 146% | 327% | 1,099% |
| Median CAGR | 43% | 57% | 62% | 64% |
| Chance of Profit | 82.7% | 93.3% | 97.2% | 99.4% |
| Chance of 2x+ | 31.2% | 60.3% | 79.4% | 94.6% |
| Chance of 5x+ | 5.9% | 22.8% | 43.5% | 75.9% |
| Bear Case (Bottom 5%) | −15% | −5% | +17% | +93% |
| Bull Case (Top 5%) | +429% | +1,124% | +2,838% | +12,893% |
| Median Max Drawdown | 8% | 11% | 14% | 17% |
| Worst-Case Drawdown (Top 5%) | 21% | 27% | 31% | 35% |
What the Numbers Mean
The strategy is profitable in the vast majority of scenarios. Over a 3-year horizon, 97% of all simulations ended in profit. Even in the worst 5% of outcomes (the "bear case"), the strategy was still up 17% after three years.
The bear case is remarkably mild. The worst realistic 1-year scenario is a 15% loss. For a Bitcoin-based strategy, that's extremely compressed downside — BTC itself routinely drops 50–70% in bear markets.
Time is your friend. The longer you run the strategy, the better it gets. The chance of doubling your money goes from 31% at one year to 95% at five years.
Drawdowns stay controlled. The median max drawdown is only 14% over three years. Compare that to Bitcoin buy-and-hold, which has historically experienced drawdowns of 65–73% in bear markets.
If the strategy keeps producing trades similar to its historical profile, the probability of a profitable 3-year outcome is ~97%. The position-sizing framework (25% on unconfirmed entries, 99% on confirmed) creates a heavily right-skewed distribution: you lose small and win big.
Does the Edge Persist?
The biggest risk with any backtest is overfitting — the strategy might look great because it was (consciously or not) tuned to fit the historical data perfectly, but would fall apart on new data. Walk-forward analysis is the standard test for this.
The idea is simple: split the 40 trade cycles into sequential windows. Train on the first chunk, then test on the next chunk the strategy has never "seen." Slide the window forward and repeat. If the strategy is overfit, the out-of-sample (OOS) windows will show significantly worse results than the in-sample windows.
Anchored Walk-Forward Results
We ran 6 expanding windows, each testing on the next 5 trade cycles:
The out-of-sample win rate held at 88% of the in-sample rate — meaning the strategy's hit rate barely degraded when tested on data it hadn't been optimized against. All 6 out-of-sample windows were profitable. This is a strong pass.
Cohort Analysis
We also split the full history into 5 non-overlapping 8-trade cohorts and checked whether every era was profitable:
| Cohort | Period | Win Rate | Avg Return | Best Trade | Worst Trade | Total Multiplier |
|---|---|---|---|---|---|---|
| #1 | May 2014 – Apr 2016 | 50% | +3.7% | +41% | −13% | 1.23x |
| #2 | Apr 2016 – Jan 2018 | 88% | +45.2% | +178% | −5% | 11.54x |
| #3 | Apr 2018 – Apr 2021 | 62% | +64.7% | +384% | −14% | 10.27x |
| #4 | Jul 2021 – Jul 2023 | 50% | +1.7% | +12% | −8% | 1.12x |
| #5 | Oct 2023 – Jan 2026 | 50% | +15.6% | +48% | −9% | 2.73x |
Every single cohort was profitable. The weakest period (Cohort #4, mid-2021 through mid-2023 — the post-crash chop zone) still returned 1.12x. The strategy didn't have a single era where it went sideways-to-down for an extended period. This is important because it means the edge isn't dependent on any single market regime.
OOS win rate holds ≥75% of in-sample. All OOS windows and all cohorts are profitable. No evidence of catastrophic overfitting. The strategy's core behavior — enter fast at low risk, confirm with momentum, size up when confirmed — persists across every tested time window.
Markets That Never Happened
This is the hardest test. Monte Carlo shuffles your existing trades. Walk-forward re-slices your existing data. But Synthetic Path simulation generates entirely new Bitcoin price histories that have never existed, and runs the full strategy logic against each one.
We built a statistical model that generates BTC-like daily prices with three regimes (bull markets, bear markets, and sideways chop), fat-tailed returns (sudden large moves), and occasional jump events (5–15% single-day spikes or crashes). The model was calibrated to match Bitcoin's historical volatility characteristics but produces completely novel sequences.
The entire GVTS + VATS strategy was reimplemented in Python — HMA smoothing, Gaussian filter, ATR/SD volatility envelopes, EMA confluence, regime detection, VATS z-score, and the full 25%→100% position-sizing state machine — and executed against 1,000 synthetic paths, each covering 3 years.
Strategy vs. Buy & Hold — Head to Head
The strategy beats buy-and-hold in 56% of all synthetic paths. That might sound modest, but the important context is where it wins and how it loses.
Where the Strategy Wins vs. Loses
| Market Regime | Paths | Strategy | Buy & Hold | Winner |
|---|---|---|---|---|
| Bull-Dominated | 62% | 2.3x | 4.1x | Buy & Hold |
| Bear-Dominated | 14% | 1.1x | 0.2x | Strategy |
| Chop-Dominated | 18% | 1.3x | 0.7x | Strategy |
In bull markets, buy-and-hold wins. When BTC goes straight up, being 100% invested 100% of the time beats any system that occasionally moves to cash. The strategy captures most of the upside (2.3x vs 4.1x) but not all of it. This is the known cost of the insurance.
In bear and chop markets, the strategy dominates. When BTC crashes (bear-dominated paths), buy-and-hold ends at 0.2x — an 80% loss. The strategy ends at 1.1x — roughly breakeven. In sideways chop, buy-and-hold loses 30% while the strategy gains 30%.
This is the same pattern from the backtest deep dive: the strategy's value isn't in beating buy-and-hold during rips. It's in surviving bear markets with capital intact so you compound from a higher base going into the next cycle.
Performance Statistics
| Metric | Strategy | Buy & Hold |
|---|---|---|
| Chance of Profit | 77.5% | 59.0% |
| Chance of 2x+ | 51.4% | 46.7% |
| Bear Case (Bottom 5%) | −51% | −94% |
| Median Max Drawdown | 43% | 61% |
| Worst-Case Drawdown (Top 5%) | 67% | 98% |
| Median Win Rate | 33% | N/A |
| Avg Trades per Path | 12 | N/A |
The Monte Carlo shuffles trades the strategy actually produced on real BTC data. The synthetic paths include market conditions much harsher than BTC has historically delivered — 18-month bear markets, extended chop with no clear trend, and whipsaw regimes that generate more false signals. The synthetic test is intentionally adversarial. Think of it as the "what if things are harder than they've ever been" scenario.
Even on never-before-seen price paths — including brutal bear markets and extended chop — the strategy remains profitable in 77.5% of simulations and cuts max drawdowns by roughly a third compared to buy-and-hold. The edge is real and isn't dependent on BTC always going up.
Putting It All Together
"Three different methods. Three different questions.
The same answer: the edge is real."
Why the Position Sizing Is the Edge
Every analysis pointed to the same conclusion: the 25% → 100% position sizing mechanism is the single most important piece of the strategy. It creates an asymmetric payoff structure where the cost of being wrong is capped at roughly a quarter of what a standard system would risk, while the upside on confirmed trends is uncapped.
The historical trade data shows this clearly. The average winning trade returns +47.6% at the portfolio level. The average loser costs only −6.1%. That's a 7.85x win/loss ratio. You can be wrong almost 8 times for every 1 time you're right and still break even. With a 60% historical win rate on top of that ratio, the math compounds aggressively in your favor.
What Each Test Tells You
Monte Carlo: "Is the backtest sequence-dependent?"
No. Shuffling the trades into thousands of random orderings still produces a profitable outcome 97% of the time over 3 years. The result isn't dependent on one lucky stretch — the trade-level edge is strong enough to survive any ordering.
Walk-Forward: "Is the strategy overfit?"
No evidence of catastrophic overfitting. The out-of-sample win rate retains 88% of the in-sample rate, and every sequential cohort from 2014 through 2026 was profitable. The edge persists across every tested era.
Synthetic Paths: "Does it work on markets we've never seen?"
Yes, with caveats. The strategy is profitable on 77.5% of entirely novel price paths and dramatically reduces drawdowns compared to buy-and-hold. It underperforms B&H in strong bull markets (the known cost) but dominates in bears and chop.
The Honest Caveats
No forward test can predict the future. All three methods assume, to varying degrees, that the future will resemble the past. They cannot model regulatory black swans, a fundamental change in Bitcoin's market structure, or the possibility that the strategy's edge degrades as more people adopt similar approaches.
Use the bear case as your planning baseline, not the median. The P5 number (bottom 5% of outcomes) is the one you should size your portfolio around. If you can survive the bear case comfortably, the median and bull case take care of themselves.
What to Expect
Across all three methods, the consistent finding is that the 8th Rule has a genuine, robust edge that isn't an artifact of curve-fitting or lucky sequencing. The edge comes from one place: asymmetric position sizing that makes being wrong cheap and being right very profitable.
The strategy won't beat buy-and-hold during every bull run — that's the cost of the insurance. But it will protect capital during the crashes that destroy compounding. And over a multi-year horizon, that protection — compounding from a higher base after every drawdown — is worth more than capturing the last 10% of a parabolic rally.
The math says: stay patient, trust the sizing framework, and let time work for you. A 97% chance of profit over 3 years — with a median 3.3x return and a worst-case floor of +17% — is as strong a forward outlook as any systematic strategy can produce.