The Primary SMA Lagged by More Than Half

The result in one sentence

The paper-highlighted daily SMA12 did make money in this evaluation. After the frozen primary cost, it returned +284.30%. But identically costed buy-and-hold returned +729.21%, and the daily excess-return evidence was negative.

SMA12 therefore did not meet the frozen excess_return_v2 criterion. The retained SMA24 member also did not meet it. The completed family closed DID_NOT_MEET_CRITERIA.

The useful lesson is not that a positive return proves a strategy worked. The question was whether the published primary rule established excess return over the benchmark named in advance.

What the original paper reported

Marina Resta, Paolo Pagnottoni and Maria Elena De Giuli examined technical analysis in Technical Analysis on the Bitcoin Market: Trading Opportunities or Investors’ Pitfall?, published in Risks on 6 May 2020.

The paper used Bitcoin HLOC data available through Kaggle, covering 1 January 2012 through 20 August 2019. It evaluated moving-average trading signals in separate three-month windows, reporting annualized return, Sharpe, downside and spread measures. It describes the daily 12-time-unit simple moving average, SMA12, as its best daily SMA result; SMA24 was retained here as a published descriptive robustness member.

The source’s result is a historical claim under its own data and signed-return construction. This study does not reconstruct that calculation.

This was a later historical validation, not a literal reproduction or forward test

The frozen question was:

Do the published daily SMA12 and SMA24 parameterizations, with SMA12 as the paper’s primary daily SMA result, meet the frozen Strategy Lab primary-member excess-return criterion versus identically costed buy-and-hold on Binance Spot BTCUSDT over [2020-06-01, 2026-09-01)?

The interval begins after the paper’s publication and after its stated market sample. It is therefore a post-publication, out-of-sample validation. It was still evaluated from historical data that existed before the research specification was frozen. It was neither a live trial nor a genuinely forward-running observation.

The material substitutions were declared before evaluation:

One source ambiguity also remains visible: its displayed moving-average equation and prose describe the lookback window differently. This validation follows the unambiguous SMA12 and SMA24 labels and the source prose for the horizons. It does not claim to resolve that equation ambiguity.

What was frozen before evaluation

Both members used daily BTCUSDT bars, a common evaluation interval, LONG/FLAT exposure, next-daily-open execution, fractional all-in accounting with terminal liquidation, and an identically costed buy-and-hold benchmark. The primary cost was 15 basis points per executed leg; 20 and 25 bps were descriptive sensitivity checks.

The frozen primary_plus_robustness decision rule made sma_12 decisive. The criterion was a positive mean daily excess log return and a positive one-sided 95% HAC lower bound relative to identically costed buy-and-hold. sma_24 was kept in its original order and reported in full, but was never a substitute for SMA12.

The shared dataset contained 2,308 normalized daily rows, from 8 May 2020 through the 1 September 2026 terminal-support candle. The earliest rows supply the longer 24-day warm-up; each member starts independently FLAT at the common first evaluation open.

Both results were positive in isolation and negative against the benchmark

At the frozen 15 bps-per-leg primary cost, buy-and-hold returned +729.21% cumulatively and annualized at +40.24%. Its log-return Sharpe was 0.593 and maximum drawdown was −76.63%.

The charts below show the annualized-return comparison and the frozen excess-return evidence. The deterministic factual report retains the full primary-cost table, including log Sharpe and closed trades. The mean-excess and lower-bound chart values convert the exact daily-log-return decimals in the publication handoff to basis points per day for readability; they do not re-estimate the study.

SMA12 had a materially shallower maximum drawdown than the benchmark, −60.43% versus −76.63%. That is a descriptive risk characteristic, not a replacement for the frozen relative-return criterion. SMA12’s mean daily excess log return was −3.3686 bps/day, and its one-sided 95% HAC lower bound was −9.9447 bps/day. The decisive member did not meet the test.

SMA24 also had a positive standalone return but lower return, lower Sharpe and a negative mean-excess estimate and lower bound. It supplied no conflicting evidence about the primary-member result.

What the family outcome means

The supported conclusion is:

the frozen Resta, Pagnottoni and De Giuli daily SMA12 primary member did not establish the declared excess-return claim over identically costed buy-and-hold on Binance BTCUSDT during the 2020–2026 historical validation, under Strategy Lab’s stated data, execution, accounting, cost, benchmark and statistical conventions. SMA24 also did not meet that criterion.

DID_NOT_MEET_CRITERIA does not establish that moving-average strategies are universally unprofitable, that the source paper was wrong, or that every market, position model, cost model or evaluation period would give the same outcome. The historical simulation also does not establish future profitability or live execution feasibility.

Evidence and limits

The normalized dataset is bound by SHA-256:

f3c2508b4548100c17d8dc63dcc3afb7fbf2cc5d693a0ba9c21eb884c5038f64

The frozen family identity is:

ff5152547e2c7a31376fef446bd733289cf40854e5ed196e15eaa2f24d7a7502

The selected public evidence preserves the frozen research declaration, authoritative publication handoff and byte-identical factual report. It deliberately excludes raw provider archives, binaries and private runner material. The files permit inspection of selected evidence; they do not provide a complete independent rerun package.

The members share one BTCUSDT series and overlapping lookbacks, so they are strongly dependent. The frozen primary-plus-robustness rule is a declared decision rule, not a calibrated family-level confidence test. These limits define the scope of the conclusion rather than weakening the need to report the unfavorable result.

Continue reading

• How the methodology treats frozen criteria and historical evidence
• Why a positive annualized return can still fail a study