FMZ排序多空框架：美股30日相对输家—赢家五日反转篮子

Hypothesis

Derived by the Research Lead from 善《实现有序排列的多空均衡权益策略》: 将文中的横截面排序、等金额多空和30日动量对随后五日收益呈负相关的发现，操作化为USEQ日线反转篮子。这是Research Lead的研究规则，不是作者完整策略的复现。原文是研究演示：未来收益只用于事后评价；momentum函数用全样本最后价格作分母，存在前视；两日分组代码、五日前向分析和月度交易叙述也不一致。本合约明确采用当日可知价格作分母、买输家卖赢家、持有五个交易日。选择原文股票池中平台配置已列出的12只股票作为固定研究篮子，不声称复现原32股历史宇宙。全部26幅附图已查看：S1呈现负相关、噪声较大的每日相关、月际变化及累计收益回撤；S2为配对交易教学图；S3/S4为数据和部署示意；S6为历史币价与跨所价格清洗图。它们不提供本策略可验证的业绩证据。补充的篮子数量、风险预算及保护规则均在参数计划登记。每日收盘作决策，订单在下一交易日开盘成交；保护止损由平台在日内匹配。

corpus_concentration为空，不能声称这些股票未被研究过。该方案减少已达78.7%的长仓策略占比并增加多标的研究；保留来源关于美股短期横截面反转的具体发现，因此不为了配额改成加密资产或期权。区别于已失败的三年形成/一年持有ETF逆向组合，以及拟合协整关系的双寡头配对：这里以30个交易日形成排名，以五个交易日兑现，无协整筛选或全样本相关性筛选。当前管线只提供部分名称，无法保证与未列出的132项完全不重复，应交由完整新颖性门检查。

negative_expectancy: the Research Lead's fixed twelve-stock, 30-session ranking and five-session long-loser/short-winner operationalization loses after modeled costs on 1,437 completed baskets. Average trade return is -0.0733% versus the +0.05% USEQ review floor, with PF below one and negative Sharpe. This rejects this operationalization, not the source author's complete strategy or equity reversal generally. No optimization sweep has been performed, so alternative basket breadth is untested.

Implementation

Fixed 12-stock USEQ basket ranks causal 30-session momentum, buys three losers and shorts three winners with equal-dollar allocations and 60% total gross exposure. Whole-share positions hold five sessions, with 10% per-leg intrabar stops and a 4% basket-equity loss exit at daily closes.

Implements the Research Lead operationalization, not a source port. Scores use (current close - close 30 sessions ago) / current close, with ascending instrument IDs breaking ties on both sides. Every leg must have all 31 matching calendar closes. A six-session calendar grid anchored to the first USEQ trading session of 1970 allows five invested sessions followed by one session to confirm closing fills before reopening. Decisions occur at closes and fills at following session opens. Initial net exposure after whole-share rounding must be within 2% of equity. Realized stop losses remain included in the basket-loss reference through mark-to-market account equity. Only names_per_side is tunable from 2 to 4; all registered fixed values are preserved. Book state uses scalar attributes.

Verification Results

Optional: with an open basket, keep holding through a missing-data day until the scheduled exit or the basket-loss exit. Gate only new entries on complete data. Otherwise, document that early liquidation is the intended reading.

When any of the 12 names lacks an exactly matching closed bar on a day, current_rows is emptied. If a basket is open, that triggers a full basket close (`not self.current_rows`). The contract says missing_data_policy=skip_whole_basket. That clearly covers skipping entry, but it does not say whether an OPEN basket should be liquidated on a single missing bar. A one-day Yahoo gap in one name therefore exits all legs early.

No change needed.

min_bars_required is overridden to 31. This is deliberate and matches warmup_bars=31. It is noted only because Layer 1 flagged the base_shadow.

The sandbox is one unoptimized draw with names_per_side=3: 1620 closed legs (270 rebalances x 6 legs, so the ledger books every leg), avg -0.087% of notional, PF 0.94, Sharpe -0.26. Losses concentrate in the high-volatility tercile (-13.5%). USEQ has zero commission; modeled borrow cost is $762. A 5-day, 12-name reversal needs a gross move per leg well above the ~0.05% spread+impact floor. The sandbox shows none: win rate is about 49%, and short legs win 47%. The analyst should first check whether the loss comes from the short-winner legs (momentum continuation) and from the 10% leg stops cutting reversals.

Backtest Review

Sharpe

-0.23

Total return

-63.64%

Max drawdown

68.21%

Trades

8622

Win rate

48.9%

Profit factor

0.96

Trade records show 1,437 completed baskets, each with three long and three short legs. Initial legs use approximately 10% of capital each; sampled next-open fills and seven-calendar-day holds are consistent with the five-session basket.

All twelve input audits pass; the evaluation covers 1992–2026.

Reliable negative net expectancy: 8,622 closed legs, average trade return -0.0733% versus the USEQ review floor of +0.05%, PF 0.955, Sharpe -0.229 and total return -63.64%.

QA's economic concern is borne out. Closed long legs earn approximately $68,453, while short legs lose $131,780. Shorting ranked winners is the principal losing side; stop-loss causality has not been independently established.

Every volatility tercile loses. The positive signal diagnostic measures the primary-instrument forecast and does not establish profitability of the executed cross-sectional basket.

Only names_per_side is tunable, from two to four. No measured viable region supports spending two hours searching basket breadth.

Benchmark-relative rejection is inappropriate because benchmark_meaningful is false.

Outcome Summary

A positive primary-instrument forecast diagnostic does not establish profitability of the executed cross-sectional basket, and alternative basket breadth remains untested.

The backtest-review verdict was abandon because net expectancy was negative and average trade return fell below the +0.05% USEQ review floor. Optimization, post-optimization analysis, and risk review were not reached.

It tested five-session equity reversal by buying three 30-session losers and shorting three winners with equal-dollar legs from a fixed twelve-stock USEQ basket.

The 1992–2026 backtest recorded 8,622 closed legs across 1,437 completed baskets, returning -63.64% with Sharpe -0.229 and maximum drawdown 68.21%. Average trade return was -0.0733% and profit factor was 0.955; short legs were the principal losing side.

Analysis

实现有序排列的多空均衡权益策略

Analysis

根据数据驱动技术进行配对交易

时间序列数据分析与Tick数据回测

Linux托管者安装和升级最佳方法

Free Download

量化分析数字货币市场