两本规则驱动的股票书和一个 SPY 核心仓在 Alpaca 模拟账户上实时交易,虚拟资金 $100,000。每天收盘后打分,次日开盘下单,没有人工干预,也没有 LLM 做方向判断。这是回测之后的前瞻记录。Two rule-based stock books and an SPY core position trade live on an Alpaca paper account with $100,000 of simulated money. They are scored after each close and ordered for the next open, with no discretion and no LLM making directional calls. This is the out-of-sample record that follows the backtests.
数据截至Data as of 2026-10-02 · 模拟资金,非真实账户simulated money, not a real account
三本书净值Book NAV
每本书从首笔成交前一个收盘起 = 100;SPY、QQQ 自 2026-09-21 收盘起 = 100,用复权价(含分红)。书的线是模拟器口径加上持仓应得的分红(模拟账户不发分红);合计按日收益连乘;竞价口径 = 每笔成交按当天开盘竞价价重记、扣监管费。Each book = 100 at the close before its first fill; SPY and QQQ = 100 at the 2026-09-21 close, on adjusted prices (dividends included). Book lines are the simulator NAV plus the dividends the positions were entitled to (the paper account pays none); the total chains daily returns; the auction basis re-prices every fill at that day's opening cross, net of regulatory fees.
每天给全市场打分,四个经典因子等权合成,排名进前 30 买入、跌出前 60 卖出。怎么打分见下面「长线书怎么选股」。Scores the whole market daily on four classic factors, equal weight; buys the top 30 and sells below 60. The full method is under “How the long book picks stocks” below.
公司高管或董事在公开市场用自己的钱买入本公司股票(SEC Form 4)后,次日开盘买入,持有 5 个交易日卖出。入场用当天有效的限价单(前一天收盘价 + 3%),不用市价单:模拟器的市价单按开盘后的卖一成交,可能远高于开盘价;高开超过 3% 的不追,下一个交易日再试一次。When an officer or director buys their own company’s stock in the open market (SEC Form 4), the book buys at the next open and sells five sessions later. Entries are DAY limit orders at the previous close + 3%, not market orders: the simulator fills a market order at the first ask after the open, which can be far above the open; a name that opens more than 3% higher is not chased and is tried once more the next session.
被动持有 SPY。大盘股选股的九个规则都没有找到边际(S37),所以这部分资金只拿指数,不按规则卖出。Holds SPY passively. Nine large-cap stock-picking rules found no edge (S37), so this money simply holds the index and is not sold by any rule.
完全是规则,没有人工判断,也没有机器学习拟合。四个因子都是学术文献里存在了几十年的经典定义,参数在上线前写死。It is entirely rule-based: no discretion and no machine-learned weights. All four factors use textbook definitions that have been in the academic literature for decades, and every parameter was fixed before launch.
当天在交易的美股,过去 20 个交易日平均成交额至少 500 万美元。成交太少的股票进出成本太高,不进池。US stocks listed that day with at least $5M average daily dollar volume over the last 20 sessions. Thinner names cost too much to trade and are left out.
只用打分当天之前已经向 SEC 申报的财报(时点正确,不偷看未来),而且最近一份申报必须在 200 天以内。再排除三类会让比率失真的公司:市值低于 1 亿美元、总资产低于 1,000 万美元、股东权益不到总资产的 5%(权益接近零或为负时,ROE 和账面市值比都没有意义)。Only filings already public with the SEC on the scoring day are used (point in time, no look-ahead), and the latest must be under 200 days old. Three kinds of company are removed because their ratios break: market cap under $100M, total assets under $10M, and equity under 5% of assets (near-zero or negative equity makes ROE and book-to-market meaningless).
| 族Family | 指标Measure | 怎么算Formula | 方向Direction | 为什么Why |
|---|---|---|---|---|
| 价值Value | 账面市值比Book-to-market | 股东权益 ÷ 市值equity ÷ market cap | 越高越好higher is better | 买得便宜cheap relative to net assets |
| 盈利收益率Earnings yield | 近 12 个月净利润 ÷ 市值trailing-12-month net income ÷ market cap | 越高越好higher is better | 每一块钱市值对应的利润profit per dollar of price | |
| 质量Quality | 毛利 ÷ 总资产Gross profit ÷ assets | 近 12 个月毛利 ÷ 总资产TTM gross profit ÷ total assets | 越高越好higher is better | 资产赚钱的效率(Novy-Marx)how productive the assets are (Novy-Marx) |
| ROE | 近 12 个月净利润 ÷ 股东权益TTM net income ÷ equity | 越高越好higher is better | 股东的钱的回报return on shareholders’ money | |
| 应计项Accruals | (净利润 − 经营现金流)÷ 总资产(net income − operating cash flow) ÷ assets | 越低越好lower is better | 利润里现金越多、会计估计越少越可信earnings backed by cash are more reliable | |
| 资产增长Asset growth | 总资产一年增长率one-year growth in total assets | 越低越好lower is better | 扩张太快的公司之后往往跑输fast-expanding firms tend to lag later | |
| 动量Momentum | 12-1 月动量12-1 month momentum | 过去 12 个月涨幅,跳过最近 1 个月return over the past 12 months, skipping the latest month | 越高越好higher is better | 强者恒强;跳过最近一个月是为了避开短期反转winners keep winning; the last month is skipped to avoid short-term reversal |
| 低波动Low volatility | 252 日波动率252-day volatility | 过去一年日收益的标准差standard deviation of daily returns over a year | 越低越好lower is better | 低波动股票的风险调整后收益更高calmer stocks earn more per unit of risk |
每个指标当天在全池里先截掉最高和最低 1%,再减均值、除以标准差,变成 z 分。一个族的分数是族内指标的平均。Each measure is clipped at the day's 1st and 99th percentiles, then turned into a z-score across the universe. A family's score is the mean of its measures.
综合分 = 四个族分数的简单平均,不按历史收益拟合权重。至少要有三个族有分数才参与排名。Composite = the plain average of the four family scores. No weights are fitted to past returns. A stock needs scores in at least three families to be ranked.
每天收盘后按综合分排名。排名进前 30 且有空槽位就买入,每个槽位 = 书净值 ÷ 30;已持有的股票只要还在前 60 就继续拿,掉出前 60 才卖。前 30 和前 60 之间的缓冲区是为了减少来回换手。订单在次日开盘成交;空着的槽位留现金,不分给其他股票。没有止损(回测里止损降低了收益)。After each close the market is ranked by composite. A name in the top 30 is bought if a slot is free; each slot is book equity ÷ 30. A held name is kept while it stays in the top 60 and sold when it drops out; the gap between 30 and 60 cuts churn. Orders fill at the next open. Empty slots stay in cash. There is no stop-loss (stops lowered returns in the backtest).
前 30 名每只股票的四个族分数(z 分)和综合分。图里每一段是一个族对综合分的贡献(族分数 ÷ 4),悬停看数值。The top 30 with their four family z-scores and the composite. Each bar segment is one family’s contribution to the composite (family score ÷ 4); hover for values.
| 排名Rank | 代码Ticker | 价值Value | 质量Quality | 动量Momentum | 低波动Low vol | 综合Composite |
|---|---|---|---|---|---|---|
| 1 | EDRY 持有held | -0.02 | +0.15 | +4.33 | -0.43 | +1.01 |
| 2 | TXG 持有held | -0.55 | +0.53 | +4.33 | -0.42 | +0.97 |
| 3 | VSNT 持有held | +2.28 | +0.27 | — | +0.33 | +0.96 |
| 4 | ADAM 持有held | +2.74 | -0.11 | +0.55 | +0.61 | +0.95 |
| 5 | STX 持有held | -0.54 | +0.91 | +3.89 | -0.49 | +0.94 |
| 6 | NESR 持有held | +0.02 | +0.03 | +3.73 | -0.04 | +0.94 |
| 7 | MU 持有held | -0.37 | +0.28 | +4.33 | -0.66 | +0.90 |
| 8 | WDC 持有held | -0.34 | +0.16 | +4.33 | -0.63 | +0.88 |
| 9 | ATEX 持有held | -0.25 | -0.49 | +4.33 | -0.17 | +0.86 |
| 10 | MRP 持有held | +1.49 | +1.53 | -0.31 | +0.71 | +0.86 |
| 11 | EVC | -0.46 | +0.85 | +4.06 | -1.14 | +0.83 |
| 12 | TEN 持有held | +1.28 | +0.06 | +1.47 | +0.50 | +0.83 |
| 13 | MFA 持有held | +2.73 | -0.09 | -0.22 | +0.86 | +0.82 |
| 14 | TWST 持有held | -0.61 | +0.06 | +4.33 | -0.51 | +0.82 |
| 15 | VSXY 持有held | -0.25 | +0.61 | +3.25 | -0.40 | +0.80 |
| 16 | NBR 持有held | +0.62 | +0.71 | +1.85 | +0.02 | +0.80 |
| 17 | FET | -0.28 | +0.14 | +3.21 | +0.03 | +0.78 |
| 18 | VISN 持有held | +2.33 | +0.13 | +0.43 | +0.18 | +0.77 |
| 19 | ERAS 持有held | -0.74 | +0.85 | +4.33 | -1.37 | +0.77 |
| 20 | PTEN | +0.17 | +0.45 | +2.36 | +0.08 | +0.76 |
| 21 | SNDK 持有held | -0.38 | +0.61 | +4.33 | -1.53 | +0.76 |
| 22 | GSL | +1.53 | +0.15 | +0.60 | +0.71 | +0.75 |
| 23 | RVMD 持有held | -0.69 | -0.52 | +4.33 | -0.15 | +0.74 |
| 24 | TRTX | +2.50 | -0.10 | -0.36 | +0.91 | +0.74 |
| 25 | CDNA | -0.35 | -0.10 | +3.70 | -0.33 | +0.73 |
| 26 | XIFR 持有held | +2.41 | +0.04 | -0.06 | +0.50 | +0.72 |
| 27 | RLAY 持有held | -0.61 | -0.24 | +4.00 | -0.26 | +0.72 |
| 28 | GNW 持有held | +2.25 | -0.10 | -0.11 | +0.84 | +0.72 |
| 29 | ECPG | +0.59 | -0.02 | +1.76 | +0.52 | +0.71 |
| 30 | PSEC 持有held | +2.56 | +0.15 | -0.39 | +0.53 | +0.71 |
四个族等权,但实际持仓往往分成两群。一群是动量极强的股票:动量 z 分能到 4 以上,而其他族很少超过 2,所以一只股票只要涨得足够多,就能靠动量一项排进前 30(今天前 30 里有 16 只动量分 ≥ 3)。另一群是深度价值股,常见的是商业发展公司(BDC)和抵押贷款 REIT,账面市值比和盈利收益率都很高(今天有 9 只价值分 ≥ 1.5)。基本面过滤的作用是把动量尾部里没有盈利、没有资产的公司挡在外面。回测检验过:把这两群拆开、改成排名标准化、或只做纯动量,效果都更差(S29、S30),所以规则保持原样。The four families are equal-weighted, but the holdings usually split into two groups. One is extreme momentum: momentum z-scores can exceed 4 while the other families rarely pass 2, so a stock that has risen enough can reach the top 30 on momentum alone (16 of today’s top 30 have a momentum score of 3 or more). The other is deep value, often business development companies and mortgage REITs with high book-to-market and earnings yield (9 today have a value score of 1.5 or more). The fundamentals filter keeps unprofitable, asset-light names out of the momentum tail. Rank-based scaling, splitting the two groups, and a pure-momentum book were all tested and did worse (S29, S30), so the rule stays as it is.
| 书Book | 代码Ticker | 入场日Entered | 入场价Entry | 最新Last | 收益Return | 权重Weight | 计划出场Planned exit |
|---|---|---|---|---|---|---|---|
| SPY 核心仓SPY core | SPY | 2026-09-25 | 768.54 | 769.64 | +0.1% | 92.2% | 排名跌出前 60until rank > 60 |
| 内部人短线Insider short-term | HHH | 2026-09-28 | 68.49 | 71.04 | +3.7% | 5.2% | 2026-10-05 |
| 内部人短线Insider short-term | INR | 2026-09-28 | 13.33 | 12.91 | -3.1% | 5.1% | 2026-10-05 |
| 内部人短线Insider short-term | CX | 2026-09-28 | 9.64 | 9.53 | -1.2% | 5.1% | 2026-10-05 |
| 内部人短线Insider short-term | INBX | 2026-09-28 | 105.33 | 96.99 | -7.9% | 5.1% | 2026-10-05 |
| 内部人短线Insider short-term | KBDC | 2026-09-28 | 13.34 | 12.67 | -5.0% | 5.0% | 2026-10-05 |
| 内部人短线Insider short-term | AVR | 2026-10-02 | 7.84 | 7.71 | -1.7% | 4.9% | 2026-10-09 |
| 内部人短线Insider short-term | COUR | 2026-10-01 | 5.04 | 5.02 | -0.4% | 4.9% | 2026-10-08 |
| 内部人短线Insider short-term | BPRE | 2026-10-01 | 12.62 | 12.69 | +0.6% | 4.9% | 2026-10-08 |
| 内部人短线Insider short-term | FUL | 2026-10-01 | 49.64 | 50.20 | +1.1% | 4.9% | 2026-10-08 |
| 内部人短线Insider short-term | DMRA | 2026-09-29 | 20.93 | 20.65 | -1.3% | 4.9% | 2026-10-06 |
| 内部人短线Insider short-term | LLYVK | 2026-09-29 | 101.81 | 97.87 | -3.9% | 4.8% | 2026-10-06 |
| 内部人短线Insider short-term | PRTA | 2026-10-02 | 8.58 | 8.36 | -2.5% | 4.8% | 2026-10-09 |
| 内部人短线Insider short-term | DFDV | 2026-09-28 | 5.79 | 5.25 | -9.3% | 4.5% | 2026-10-05 |
| 内部人短线Insider short-term | GPI | 2026-09-29 | 246.50 | 249.84 | +1.4% | 4.4% | 2026-10-06 |
| 内部人短线Insider short-term | NCT | 2026-09-28 | 4.24 | 1.68 | -60.4% | 2.0% | 2026-10-05 |
| 长线综合因子Long composite | TXG | 2026-09-22 | 77.81 | 93.50 | +20.2% | 3.9% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | TWST | 2026-09-22 | 164.80 | 188.67 | +14.5% | 3.6% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | RVMD | 2026-09-22 | 192.10 | 206.64 | +7.6% | 3.6% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | RLAY | 2026-09-22 | 18.39 | 18.96 | +3.1% | 3.5% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | SYRE | 2026-09-22 | 93.81 | 93.19 | -0.7% | 3.4% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | TEN | 2026-09-30 | 47.91 | 50.05 | +4.5% | 3.4% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | VISN | 2026-09-30 | 5.96 | 6.21 | +4.2% | 3.4% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | HG | 2026-09-22 | 34.06 | 34.13 | +0.2% | 3.4% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | ATEX | 2026-09-22 | 84.00 | 84.33 | +0.4% | 3.4% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | NBR | 2026-09-30 | 78.56 | 79.64 | +1.4% | 3.3% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | VSTS | 2026-09-22 | 14.30 | 14.03 | -1.9% | 3.3% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | ERAS | 2026-09-22 | 15.21 | 14.74 | -3.1% | 3.3% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | ARI | 2026-09-22 | 6.47 | 6.20 | -4.2% | 3.3% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | XIFR | 2026-09-22 | 10.84 | 10.39 | -4.1% | 3.2% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | GNW | 2026-09-22 | 10.09 | 9.55 | -5.4% | 3.2% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | VSXY | 2026-09-30 | 90.70 | 91.73 | +1.1% | 3.2% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | DFTX | 2026-09-30 | 38.34 | 37.40 | -2.5% | 3.1% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | VSNT | 2026-09-30 | 32.42 | 30.94 | -4.6% | 3.1% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | PSEC | 2026-09-22 | 2.23 | 2.00 | -10.3% | 3.1% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | MFA | 2026-09-22 | 8.54 | 7.65 | -10.4% | 3.0% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | NESR | 2026-09-30 | 26.97 | 24.67 | -8.5% | 3.0% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | MRP | 2026-09-30 | 26.76 | 24.61 | -8.0% | 3.0% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | SNDK | 2026-09-30 | 1750.00 | 1719.99 | -1.7% | 3.0% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | STX | 2026-09-22 | 868.47 | 848.99 | -2.2% | 2.9% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | ADAM | 2026-09-22 | 8.68 | 7.81 | -10.0% | 2.9% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | ORKA | 2026-09-22 | 94.99 | 84.52 | -11.0% | 2.9% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | WDC | 2026-09-22 | 445.01 | 415.29 | -6.7% | 2.9% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | EDRY | 2026-09-22 | 69.60 | 55.88 | -19.7% | 2.7% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | CLYM | 2026-09-22 | 13.89 | 10.94 | -21.2% | 2.7% | 排名跌出前 60until rank > 60 |
| 长线综合因子Long composite | MU | 2026-09-22 | 1040.64 | 1074.89 | +3.3% | 1.9% | 排名跌出前 60until rank > 60 |
| 书Book | 代码Ticker | 入场日Entered | 入场价Entry | 出场日Exited | 出场价Exit | 收益Return |
|---|---|---|---|---|---|---|
| 内部人短线Insider short-term | GSHD | 2026-09-25 | 45.31 | 2026-10-02 | 46.11 | +1.8% |
| 内部人短线Insider short-term | OFIX | 2026-09-25 | 9.21 | 2026-10-02 | 9.12 | -1.0% |
| 长线综合因子Long composite | BZ | 2026-09-22 | 14.51 | 2026-09-30 | 14.25 | -1.8% |
| 长线综合因子Long composite | MCHB | 2026-09-22 | 15.39 | 2026-09-30 | 14.87 | -3.4% |
| 长线综合因子Long composite | PACS | 2026-09-22 | 42.89 | 2026-09-30 | 41.20 | -3.9% |
| 长线综合因子Long composite | GMRS | 2026-09-22 | 11.90 | 2026-09-30 | 9.75 | -18.1% |
| 长线综合因子Long composite | PERI | 2026-09-22 | 8.92 | 2026-09-30 | 8.59 | -3.7% |
| 长线综合因子Long composite | UHAL | 2026-09-22 | 64.81 | 2026-09-30 | 60.45 | -6.7% |
| 长线综合因子Long composite | TK | 2026-09-22 | 13.89 | 2026-09-30 | 14.25 | +2.6% |
| 长线综合因子Long composite | DEC | 2026-09-22 | 13.71 | 2026-09-30 | 13.41 | -2.2% |
| 长线综合因子Long composite | GSBD | 2026-09-22 | 9.96 | 2026-09-30 | 9.15 | -8.1% |
| 内部人短线Insider short-term | PRTA | 2026-09-23 | 8.98 | 2026-09-30 | 8.52 | -5.1% |
| 内部人短线Insider short-term | GSAT | 2026-09-23 | 82.58 | 2026-09-30 | 82.36 | -0.3% |
自 2026-09-22 起 72 笔成交,成交价对当天开盘竞价价平均 +nan%/边(正数 = 比竞价价差);开盘竞价单 2 笔 +0.26%;DAY 市价单 22 笔 +1.25%;DAY 限价单 48 笔 +nan%。只有开盘竞价单在竞价里成交,DAY 单由模拟器在开盘后逐单撮合,那部分偏差不代表真实账户的成本。72 fills since 2026-09-22, mean gap to the day's opening cross +nan% per side (positive = worse than the cross); opening-auction orders 2 at +0.26%; DAY market 22 at +1.25%; DAY limit 48 at +nan%. Only opening-auction orders fill in the cross; DAY orders are filled one by one by the simulator after the open, so that part of the gap is not a real account's cost.
回测 · 按年收益 vs SPY、QQQBacktest · calendar-year returns vs SPY and QQQ
2017-01-01 → 2026-08-31,次日开盘入场,每边扣半个报价价差,闲置资金放 SPY(模拟盘的闲置资金是现金)。长线 CAGR +19.4%,同期 SPY +15.3%、QQQ +21.2%;对 SPY 超额 +4.1%/年(主动收益 t 0.92)。内部人 CAGR +20.4%,对 SPY 超额 +5.1%/年(t 1.17)。都不显著,所以要用模拟盘攒记录。2017-01-01 → 2026-08-31, entry at the next open, half the quoted spread per side, idle cash in SPY (the paper books hold idle cash as cash). Long book CAGR +19.4% against SPY +15.3% and QQQ +21.2%; excess over SPY +4.1%/yr (active t 0.92). Insider line CAGR +20.4%, excess over SPY +5.1%/yr (t 1.17). Neither is significant, which is why the paper record exists.
评估点在模拟盘开始之前就写死了。到点之前,不根据好坏改规则。The evaluation points were fixed before the paper record started. Until then no rule changes on the strength of results.