Making 19× Was Still Not Enough

The result in one sentence

Turning $1 into about $19 would usually make for a triumphant backtest headline. In this validation, it was not enough.

The strongest frozen member, trb_50, returned +1806.95% cumulatively and annualized at 52.32% after the primary transaction-cost assumption. Its log-return Sharpe was 0.913, higher than the benchmark’s 0.709, and its maximum drawdown was materially shallower. But identically costed buy-and-hold returned +2260.99% — about $23.6 from the same $1 — and the 50-day member’s reported mean daily excess log return was negative.

The same relative test failed for the 150- and 200-day members. The frozen family therefore closed DID_NOT_MEET_CRITERIA / COMPLETE, with 0 of 3 members meeting excess_return_v2 under its predeclared 2-of-3 rule.

That is the useful distinction: spectacular-looking absolute compounding can coexist with negative evidence against the benchmark actually named in the research question.

What the original paper reported

Dirk F. Gerritsen, Elie Bouri, Ehsan Ramezanifar and David Roubaud studied seven families of technical trading rules in The profitability of technical trading rules in the Bitcoin market, published in Finance Research Letters.

The paper’s Bitcoin history ran from July 2010 through 31 December 2018. Its trading-range-breakout rule uses support and resistance over the previous n days: a close above the recent maximum is a Buy signal and a close below the recent minimum is a Sell signal. The published horizons are 50, 150 and 200 days.

In the paper’s ordinary long/out strategy, the investor is long Bitcoin on a Buy signal and otherwise holds an asset earning the risk-free rate. In the full-sample table, all three breakout horizons reported higher daily Sharpe ratios than buy-and-hold, with the 150-day version the highest of the three.

The paper also found that breakout performance varied with market conditions, with stronger relative results in several strongly trending subperiods and weaker results in the comparatively stable 2015–2016 period.

Those findings made the family a useful validation target. They did not make a later Strategy Lab run an exact replication.

This was an out-of-sample validation, not a reconstruction of the paper

The frozen research question was narrower:

Do the published 50-, 150-, and 200-day breakout parameterizations retain evidence of excess return versus identically costed buy-and-hold when evaluated on Binance Spot BTCUSDT over [2019-01-01, 2026-01-01) under current Strategy Lab conventions?

The classification was explicitly OUT_OF_SAMPLE_VALIDATION because the evaluation interval begins immediately after the source price sample ends. But this was still a historical evaluation performed after the data already existed. It was not a live forward test and the 2019–2025 observations were not genuinely unseen future data when the research specification was frozen.

Seven substitutions were declared before evaluation:

These are not footnotes to be added after seeing the answer. They define what this validation actually tested.

The accounting substitution is especially important. The paper’s long-only investor earns a risk-free return while outside Bitcoin. Strategy Lab’s LONG/FLAT accounting leaves that capital unchanged. A result under one convention should not be presented as though it were generated by the other.

What was frozen before evaluation

The family contained three symmetric members in this order:

Every member used daily BTCUSDT bars, LONG/FLAT exposure, next-daily-open execution, fractional all-in accounting with terminal liquidation, a 15 bps-per-leg primary cost, and the same identically costed buy-and-hold benchmark.

The family rule was k_of_n, with k = 2 and n = 3. No member was designated a post-result winner. Two of the three frozen members had to meet excess_return_v2 for the family threshold to be satisfied.

The shared admitted dataset contained 2,759 normalized daily rows from 14 June 2018 through the 1 January 2026 terminal-support candle. The longest 201-day warm-up explains why the capture begins before the evaluation interval.

The three frozen results

At the primary cost, buy-and-hold returned +2260.989703% cumulatively, annualized at 57.037319%, with a log-return Sharpe of 0.709154 and maximum drawdown of −76.629251%.

All three breakout members made money in absolute terms. None met the declared member criterion.

The excess-return columns above are presentation conversions of the exact daily-log-return decimals in the publication handoff into basis points per day. No strategy result was re-estimated for this article.

Why the 50-day member is the useful result

trb_50 is the row that makes this case worth publishing.

If we looked only at conventional descriptive statistics, it would be easy to present the 50-day breakout as a success. Roughly speaking, $1 became about $19 after the frozen primary costs. Its log-return Sharpe was higher than buy-and-hold. Its maximum drawdown was −62.63%, versus −76.63% for the benchmark.

But the frozen research question was not “can we find an attractive descriptive row?” It was whether the published family established excess return over an identically costed benchmark under the declared validation contract.

For trb_50, the publication handoff reports a mean daily excess log return of

-8.35266853620345e-05

and a one-sided 95% HAC lower bound of

-0.0007238471564021261.

The point estimate itself was already negative. The member therefore did not meet the declared historical benchmark test.

This distinction matters because Bitcoin’s buy-and-hold return over the same seven-year interval was exceptionally large. A strategy can compound strongly, reduce drawdown and even show a higher standalone Sharpe while still failing a frozen relative research claim.

The 150- and 200-day members did not create a rescue path

The two longer horizons were less ambiguous on absolute performance relative to the benchmark.

trb_150 annualized at 16.94% and trb_200 at 41.14%, versus 57.04% for buy-and-hold. Both had negative reported mean excess returns and negative one-sided HAC lower bounds.

The family rule required two members to meet the criterion. The terminal count was:

• MET_CRITERIA: 0
• DID_NOT_MEET_CRITERIA: 3
• INCONCLUSIVE: 0
• EVALUATION_FAILED: 0
• MISSING: 0

There was therefore no unresolved member that could have changed the 2-of-3 family decision.

The 20- and 25-bps diagnostic costs are also preserved in the handoff, but they are descriptive sensitivity checks. They were not allowed to replace the primary 15-bps classification after the result was known.

What DID_NOT_MEET_CRITERIA means

DID_NOT_MEET_CRITERIA is deliberately narrower than “this strategy is universally unprofitable.”

Every member in this evaluation produced a positive absolute cumulative return. The 50-day member in particular had economically interesting descriptive properties. None of that changes the family decision, because those properties were not substitutes for the frozen research claim.

The supported conclusion is:

the frozen Gerritsen 50/150/200-day breakout family did not establish the declared excess-return claim over identically costed buy-and-hold on Binance BTCUSDT during the 2019–2025 historical validation under Strategy Lab’s data, execution, accounting, cost, benchmark and statistical conventions.

It does not establish that:

• trading-range breakouts are universally unprofitable;
• Gerritsen et al.’s original historical calculations were wrong;
• a different market, execution model, accounting convention or objective would produce the same result; or
• these historical simulations establish future profitability or live execution feasibility.

Those would be different claims and would require different evidence.

Reproducibility and preserved identities

The normalized dataset is bound by SHA-256:

8994ffc23a62f3e824565779777921849150ba0c9b9cca326c672768cfa93b65

The frozen family identity is:

67bfba2f4a8871a7e87d39a7750f087db489cd17d1e4b2eb01240780a2223f57

The publication handoff also records the exact research plan, capture manifest, source attribution, pre-run provenance, numerical implementation, renderer, environment, executable and final result hashes. The original authoritative publication handoff is identified by:

bac1582bea07b452a1ee157a290e7180159b65fd295d81b128173f223de865b9

The dataset used 92 retained Binance Public Data archives and contains all three members in their original frozen order. The public evidence below preserves whitespace-compacted JSON copies of the result export and frozen research declaration, plus the byte-identical deterministic factual report, without republishing the raw market archives.

The result should still be read within its stated limits. These are historical hypothetical simulations, the family members share one BTCUSDT series and overlapping lookbacks, and the frozen 2-of-3 rule is a declared decision rule rather than a calibrated family-wise confidence test.

That is enough to support the conclusion actually made here—and no broader one.