Original hypothesis validation
Candidate 002 asked whether a simple asymmetric close-channel rule could avoid enough adverse crypto exposure to reduce drawdown while still establishing positive excess return over identically costed buy-and-hold. Across BTCUSDT and ETHUSDT, every frozen member improved drawdown. Neither market family met the full criterion.
Reading the result
The headline metrics answer different questions. These short definitions are the interpretation used on this page, not additional evaluation criteria.
While FLAT, a member enters LONG only when the daily close is strictly above the highest close of its prior entry lookback. While LONG, it exits only when the close is strictly below the lowest close of the prior 20 days. Signals execute at the next daily open.
Mean daily net log return of the strategy minus its identically costed buy-and-hold benchmark. BTC was negative for all three members. ETH was positive for all three, but that point estimate was not enough.
The one-sided 95% lower bound for mean daily excess log return using the frozen 20-day HAC lag. It had to be above zero. All six member bounds were negative.
A member also had to improve maximum drawdown strictly versus its benchmark. All six frozen BTC and ETH variants satisfied this part of the criterion.
The frozen asymmetric close-channel family consistently reduced drawdown on both BTC and ETH, but did not establish the positive excess return required by the test.
Every one of the six frozen members finished with a shallower maximum drawdown than identically costed buy-and-hold. That part of the hypothesis worked descriptively.
The return evidence did not.
On BTC, all three members had negative mean daily excess log return. On ETH, all three point estimates were positive, but every one-sided 95% HAC lower bound remained below zero. The primary 55/20 member therefore failed on both markets, no robustness member rescued either family, and the frozen cross-market hypothesis closed DID_NOT_MEET_CRITERIA.
That distinction is exactly why the criterion had more than one condition.
Candidate 002 was an ORIGINAL_HYPOTHESIS_VALIDATION. It was not chosen by scanning Strategy Lab’s own historical results for a profitable combination.
The mechanism was deliberately small:
The primary member used a 55-day entry / 20-day exit. Two nearby entry-horizon perturbations were frozen as robustness checks:
| Member | Role | Entry lookback | Exit lookback |
|---|---|---|---|
| breakout_50_20 | robustness | 50 days | 20 days |
| breakout_55_20 | primary | 55 days | 20 days |
| breakout_60_20 | robustness | 60 days | 20 days |
The horizon asymmetry was Turtle-derived, not a Turtle-system reproduction. The historical Turtle System 2 convention used a 55-day breakout entry and a 20-day opposite-channel exit. Candidate 002 borrowed only that pre-Bitcoin horizon asymmetry.
The implementation here is materially narrower: completed daily closes rather than intraday highs/lows, LONG/FLAT exposure, Strategy Lab’s next-open execution and accounting, no volatility sizing, no protective stop, no pyramiding, and no futures portfolio.
That distinction mattered for the research design. The 55/20 primary was anchored outside these BTC and ETH evaluation results. The 50/20 and 60/20 members were only a small ±5-day perturbation around the entry horizon, while the 20-day exit remained fixed. We did not open a two-dimensional parameter grid.
The same family was frozen independently for BTCUSDT and ETHUSDT. The two markets were not combined into a portfolio and the six members were not treated as one family.
Both market specifications used the same evaluation interval, execution convention, accounting, benchmark, costs, and statistical criterion. The evaluation interval was [2023-09-01, 2026-09-01), or 1,096 evaluation days.
The primary transaction cost was 15 basis points per executed leg, with 20 and 25 bps retained as adverse-cost diagnostics. The benchmark was identically costed buy-and-hold.
A member had to satisfy all three parts of excess_return_drawdown_v1:
Each market then used the same primary_plus_robustness family rule:
The broader Candidate 002 hypothesis was frozen to require both BTC and ETH families to succeed. One completed market failure was therefore enough for the cross-market claim not to meet its criterion.
BTC buy-and-hold returned +202.02% cumulatively after the same primary cost assumption and experienced a maximum drawdown of −52.97%.
All three breakout variants reduced that drawdown materially:
| BTC member | Cumulative return | Mean excess | HAC 95% lower bound | Max drawdown | Outcome |
|---|---|---|---|---|---|
| 50/20 | +120.34% | −2.88 bps/day | −11.23 bps/day | −35.93% | DID_NOT_MEET_CRITERIA |
| 55/20 primary | +112.35% | −3.21 bps/day | −11.65 bps/day | −37.38% | DID_NOT_MEET_CRITERIA |
| 60/20 | +125.29% | −2.67 bps/day | −11.03 bps/day | −38.37% | DID_NOT_MEET_CRITERIA |
| Buy-and-hold | +202.02% | benchmark | benchmark | −52.97% | benchmark |
The defensive effect is visible: the strategy spent far less time exposed and avoided part of the benchmark’s drawdown.
But the research question was not merely whether the filter could reduce risk. Every BTC member had a negative mean excess-return estimate, so the family failed before uncertainty could rescue anything. The negative HAC lower bounds reinforce the same conclusion.
ETH produced the more tempting result.
Identically costed buy-and-hold returned +49.49% cumulatively and experienced a −67.55% maximum drawdown.
Every breakout member again improved drawdown. Unlike BTC, every ETH member also had a positive mean excess-return point estimate:
| ETH member | Cumulative return | Mean excess | HAC 95% lower bound | Max drawdown | Outcome |
|---|---|---|---|---|---|
| 50/20 | +102.68% | +2.78 bps/day | −11.19 bps/day | −34.83% | DID_NOT_MEET_CRITERIA |
| 55/20 primary | +74.02% | +1.39 bps/day | −12.82 bps/day | −34.83% | DID_NOT_MEET_CRITERIA |
| 60/20 | +99.92% | +2.65 bps/day | −11.31 bps/day | −28.42% | DID_NOT_MEET_CRITERIA |
| Buy-and-hold | +49.49% | benchmark | benchmark | −67.55% | benchmark |
That is precisely where a looser backtest process could become misleading.
Just ’cause you feel it doesn’t mean it’s there.
The 50/20 and 60/20 ETH variants more than doubled capital over the evaluation period while buy-and-hold gained about 49%, and all three strategy paths had dramatically smaller maximum drawdowns. It would be easy to stop there and declare the idea promising.
The frozen test required more. The one-sided 95% HAC lower bound had to be positive. Instead, the bounds ranged from roughly −11.19 to −12.82 bps/day.
So the correct result is not “ETH passed.” It is that the observed ETH excess-return estimates were positive but too uncertain to establish the declared claim.
A failed research criterion does not mean every measured property was unfavorable.
Across both assets, all six members strictly improved maximum drawdown. For the primary 55/20 member:
The 60/20 ETH member reached −28.42%, the shallowest drawdown among the six frozen variants.
Those are real properties of these historical simulations under the stated accounting and cost model. They just were not sufficient for the predeclared hypothesis, which required both downside improvement and convincing positive benchmark-relative return evidence.
That distinction prevents an attractive secondary metric from silently replacing the question fixed before evaluation.
The six results create several obvious temptations:
None of those actions belongs inside Candidate 002 after observing these results.
The family was intentionally narrow so that the result could be interpreted without a large parameter search. Changing the mechanism or widening the grid now would create a new hypothesis informed by Candidate 002, not a robustness check that was somehow missing from this one.
The completed failure is therefore the endpoint of this experiment, not the start of a tuning loop.
The supported conclusion is:
Under the frozen 2023–2026 Binance Spot tests, the asymmetric 50/20, 55/20 and 60/20 close-channel families consistently reduced maximum drawdown on BTCUSDT and ETHUSDT, but neither market family established the combination of positive excess return, positive one-sided 95% HAC lower bound, and improved drawdown required by Candidate 002.
More specifically:
It does not establish that close-channel trend following is universally unprofitable, that another market or period would behave the same way, or that these historical hypothetical simulations imply future execution or profitability.
The evaluation data were historical and publicly accessible before the study was frozen. Freeze-before-evaluation protects the decision rule from this experiment’s result; it does not turn old market data into genuinely unseen future observations.
The BTC normalized dataset is bound by SHA-256:
1a97c0e21d98f02992c3c46d8e82d0e93a3ca363d0bb44e6dfb3e04cfc5bd223
The ETH normalized dataset is bound by:
7474daf56025ab08f2f22327a2ed08c10177c3bb7abdebce4ad55e4e8b6751ec
The frozen family identities are:
Both captures contained 1,158 canonical daily rows and 39 retained provider archives. The two Strategy Lab handoffs below preserve every frozen member in declared order, the exact primary-cost metrics, diagnostic costs, HAC statistics, dataset identities, implementation identities, interpretation limits, and final artifact hashes.
The family members within each market share one price series and overlapping lookbacks, so they are strongly dependent. The family rule is a declared decision rule, not a calibrated family-wise confidence test.
Frozen decision rule
For each market, the primary 55/20 member had to meet excess_return_drawdown_v1 and at least one of the frozen 50/20 or 60/20 robustness members had to meet it too. A member required positive mean excess log return, a positive one-sided 95% HAC lower bound, and strictly improved maximum drawdown.
Chart enhancement unavailable. The exact values are listed in the article table.
Chart enhancement unavailable. The exact values are listed in the article table.
Inspect the evidence
The two public handoffs below are byte-identical copies of the authoritative Strategy Lab publication handoffs supplied for this article. They preserve each market's frozen family, exact member metrics, HAC statistics, dataset and implementation identities, interpretation limits, and result-artifact hashes. Raw Binance archives, normalized bars, binaries, and private runner material are not republished.
These excerpts support inspection of the reported work; they are not a complete package for independently rerunning the experiment. Read the evidence policy.
167075ce20d6ca19fd6fc16db8dc19b27fcae79cc84bdba3ddc0cb1c54c1af5b
ETH publication handoff
Authoritative Strategy Lab handoff for the completed ETHUSDT ORIGINAL_HYPOTHESIS_VALIDATION.
SHA-256
bde798d439210732e83524d51176610e5a203d22a8f6f8b70a34615563a90dad