SuperMind 非科创 A 股：日振幅大于 1%与已完成周线 MA5/MA10 金叉

Hypothesis

Derived by the Research Lead from "(supermind量化-)振幅大于1、周线MA5金叉MA10、非科创_" by 同花顺 SuperMind (library) https://quant.10jqka.com.cn/view/article/11156: 保留正文的振幅>1%、周线MA5金叉MA10和非科创条件；研究补充振幅口径、固定测试篮子、下一开盘执行、退出和仓位，修正示例代码而不声称逐字移植。 根据同花顺 SuperMind《振幅大于1、周线MA5金叉MA10、非科创》正文提炼选股条件，在 CNEQ 日线数据上测试长期做多策略。归类为部分规则：来源提供选股信号，但没有完整退出和仓位规则。保留正文的振幅阈值、周线均线周期和科创板排除条件；不逐字执行相互矛盾的指标公式，也不将 Python 示例中的 vol>1 冒充振幅。研究选择将振幅定义为100×(当日最高价−当日最低价)/上一交易日收盘价，以完整自然交易周的最后交易日检查金叉。新增反向交叉退出、日收盘保护退出、最长持有期和权益相对仓位。固定八只非科创 A 股作为机制测试篮子，不声称复现历史全市场选股。日线决策按平台支持的下一交易日开盘成交。来源的收益或效果没有得到验证；二维码图片未读取，不作为规则依据。

提交内容明确针对中国 A 股，不能为满足配额换成加密货币或美股。CNEQ 仅占实验的0.3%，八标的测试也降低单标的集中；做多和同市场交易虽属于超额桶，但来自源信号的执行意图，不人为增加做空或跨市场腿。corpus_concentration为空，无法证明这些个股在最近90天的实际占比。与所列失败案例不同，本实验测试A股的已完成周线交叉筛选，不测试加密货币短周期指标、网格或配对回归；这是来源复现目的，不宣称均线机制本身创新。固定篮子带有存续偏差，结论仅适用于该篮子。

Reject this Research Lead operationalization on the fixed eight-stock CNEQ basket for failure of the enforced calendar-year persistence requirement: only 5 of 9 exposed years are positive, versus 60% required. This rejects its measured temporal persistence, not the SuperMind source screen or weekly crosses generally. Average net trade return 0.863% clears the supplied 0.20% economics floor. Not enough trades to validate and inconclusive significance are not the abandonment reasons. The permitted holding-horizon search has completed, and no demonstrated implementation defect or evidence-backed rule change addresses the hard failure. No further developer iteration is justified on this evidence.

Implementation

Long-only fixed eight-name non-STAR A-share basket. Buy completed-week MA5/MA10 bullish crosses when the final session daily amplitude exceeds 1%. Exit on bearish weekly cross, 9% daily-close loss or 40 elapsed sessions after the entry session. Decisions fill at the next session open.

Verification Results

The unoptimized smoke window reports -1.212% average return per closed trade across 78 trades. Assess net expectancy on full history; this is not a correctness failure. Conclusions remain limited to the fixed survivor basket and the platform's unmodeled A-share price-limit execution.

Backtest Review

Sharpe

0.15

Total return

2.58%

Max drawdown

5.94%

Trades

115

Win rate

41.7%

Profit factor

1.13

115 closed trades over 2018–2026 are reasonable for this weekly-cross strategy. Average net trade return 0.551% exceeds the CNEQ review floor of 0.15%; PF 1.133 and total return 2.581% are positive. QA’s negative smoke-window expectancy is not borne out on full history.

Trades are exclusively long, use 100-share lot multiples and have no same-session exits. Of 116 entries, 103 occurred on Mondays. Trade timing and code are consistent with completed-week MA5/MA10 crosses, the stated amplitude filter and next-open execution.

Evidence of an active edge is weak: Sharpe 0.147, five negative calendar years, alpha -0.0017 and information ratio -0.443. Optimization should assess the two registered exit tunables without assuming robustness.

The stored basket benchmark curve ends approximately +118.5% above initial capital, with approximately 44.24% drawdown, versus strategy return 2.581% and drawdown 5.944%. Different exposure limits raw-return comparison. Conclusions remain limited to the fixed survivor basket and unmodeled A-share price-limit execution.

Analysis

Final parameters improve full-history Sharpe from 0.147 to 0.191 and return from 2.581% to 3.843%, with trade count increasing from 115 to 116.

Average net trade return 0.863% exceeds the supplied 0.20% economics floor. Drawdown 6.697% is below the registered 15% limit; accounting is valid.

Baseline trade details confirm closed longs and no same-day round trips. No implementation or accounting defect is established.

Only 5 of 9 exposed calendar years are positive (55.6% versus the enforced 60% requirement). The supplied gates classify this as hard, not advisory; it cannot be waived for paper admission.

2019 contributes 68.5% of profit. Profit factor 1.169 and Calmar 0.0676 fail their soft floors.

Pooled outer OOS Sharpe is approximately 0.076, with mixed window performance. The protected holdout has 18 trades and Sharpe -0.584. Both registered statistical assessments are inconclusive, not scientific rejections.

Holding-horizon selections range from 30 to 50 sessions and are labelled unstable. Sensitivity finds no cliffs, but does not establish a validated improvement; final-parameter window replays are partly in-sample.

Full-history results meet the predicted trade count, average trade return and drawdown, but miss the predicted 8% return. Pooled OOS Sharpe also falls below the predicted 0.6 on its separate scope. These predictions are evidence, not additional gates.

Baseline alpha -0.0017 and information ratio -0.4425 provide weak benchmark-relative evidence. Optimized benchmark-relative metrics were not supplied.

soft

hard

Outcome Summary

Clearing trade economics and drawdown limits did not establish sufficient calendar-year persistence on this fixed basket.

The analyst abandoned it because only 5 of 9 exposed calendar years were positive (55.6%), below the enforced 60% hard requirement. The permitted holding-horizon search had completed, and no demonstrated implementation defect or evidence-backed rule change justified another developer iteration; risk review was not reached.

It tested long-only entries in a fixed eight-stock non-STAR A-share basket when daily amplitude exceeded 1% and completed-week MA5 crossed above MA10, with next-open execution and protective exits.

The baseline returned 2.581% with Sharpe 0.147 and 115 closed trades; final parameters reported in the analyst analysis raised return to 3.843% and Sharpe to 0.191 across 116 trades. Final average net trade return was 0.863% and drawdown 6.697%, while pooled outer OOS Sharpe was approximately 0.076 and holdout Sharpe was -0.584 across 18 trades.

Analysis

(supermind量化-)振幅大于1、周线MA5金叉MA10、非科创_