LESSON 01
Low-Frequency Quant System Overview低频量化系统总览
Core idea核心思想A complete low-frequency quant loop = factor research → strategy construction → backtest validation → risk control → iteration. Every stage needs falsifiable criteria — a test without a falsification condition is no test at all.低频量化的完整闭环 = 因子研究 → 策略构建 → 回测验证 → 风控 → 迭代。每个环节都要有可证伪的标准——没有证伪条件的测试，等于没测。
- Write the "death lines" before testing. Nine falsification criteria (F1–F9: excess return, drawdown, turnover, persistent underperformance, factor incrementality, overfitting, underwater period, cost coverage, data-snooping) are fixed before any backtest runs — see the Factor Dashboard.先写"死刑线"，再开测。九条证伪标准（F1–F9：超额收益、回撤、换手、持续跑输、因子增量、过拟合、水下期、成本覆盖、数据窥探）全部在回测前写死，详见因子表盘。
- No look-ahead, ever. Strategies may only touch data that existed at the signal date. Financial statements are lagged 63 trading days by default (point-in-time discipline).绝对禁止上帝视角。策略只能接触信号时点已存在的数据，财务数据默认滞后 63 个交易日（PIT 纪律）。
- Failed tests are assets. Every eliminated strategy is archived with its cause of death — the graveyard teaches as much as the winners.失败的测试也是资产。每个被淘汰的策略都带着死因归档——墓园和赢家一样有教学价值。
LESSON 03
Multi-Factor Models: Fama–French Foundations多因子模型基础：Fama–French
Core idea核心思想A stock's return ≈ market beta + (factor exposures × factor premia) + residual. Decomposing returns into factor exposures tells you whether you earned "style money" or true alpha.一只股票的收益率 ≈ 市场β +（因子暴露 × 因子溢价）+ 残差。把收益拆成因子暴露，才能分清赚的是"风格的钱"还是"真 alpha"。
- Under the five-factor model, the value factor (HML) is largely absorbed by profitability (RMW) and investment (CMA) — value alone is a weak signal.五因子模型下，价值因子（HML）基本被盈利（RMW）和投资（CMA）吸收——纯价值是个弱信号。
- In A-shares: quality > value. The small-cap premium exists but is decaying — don't anchor on textbook US results.A股：质量 > 价值；小市值溢价存在但在衰减——不要锚定美股教科书结论。
- Factors are a risk-budgeting language, not just a stock-picking tool.因子首先是一套风险预算语言，而不只是选股工具。
LESSON 04
The Value Factor价值因子
Core idea核心思想Value pays for "resilient earnings mispriced by the market." In A-shares, pure value is a weak factor — it only earns excess return as enhanced value, filtered by quality, cash flow and dividends.价值因子的本质是为"盈利韧性被错杀"付费。A股的纯价值是弱因子——只有叠加质量/现金流/分红过滤、变成增强型低估值后，才有超额收益。
- Backtest verdict: pure value underperformed the equal-weight benchmark and was eliminated (strategy S001, triggered F1/F5).回测裁决：纯价值跑不赢等权基准，已淘汰（策略 S001，触发 F1/F5）。
- Enhanced value = OCF > net income, dividend yield > 3%, ROE volatility < 15% — cheap plus cash-flow health.增强型低估值 = 经营现金流 > 净利润、股息率 > 3%、ROE 波动 < 15%——便宜 + 现金流健康。
- Dividend yield ≈ a proxy for sustainable cash flow — defensive value in falling-rate environments.股息率 ≈ 可持续现金流的代理变量——利率下行环境里有防御价值。
LESSON 05
The Quality Factor质量因子
Core idea核心思想Buy companies with strong, stable profitability and low leverage. Quality is among the most persistent factors after publication — its driver is profitability itself, not the ROE number.买"赚钱能力强且稳定、负债低"的公司。质量是发表后衰减最少的硬因子之一——它的驱动是盈利能力本身，而不是 ROE 这个数字。
- Pure ROE is weakly negative in A-shares — a value trap. High ROE alone often marks peak-cycle earnings, not quality.纯 ROE 在A股是弱负向——估值陷阱。单看 ROE 高，往往买到的是周期顶点的盈利，而非质量。
- Composite quality (cash-flow-based earnings + earnings stability + low leverage) is the version worth testing — it became the risk-filter inside the passing S005 strategy.合成质量（现金流口径盈利 + 盈利稳定性 + 低杠杆）才是值得测的版本——它成了通过的 S005 策略里的排雷器。
- Lesson: a good factor ≠ a good strategy — quality added nothing inside the dividend universe (S008), where high-dividend stocks were already mature and stable.教训：好因子 ≠ 好策略——在红利股票池里再叠质量筛选是负增量（S008），因为高股息公司本身已经偏成熟稳定。
LESSON 06
The Growth Factor & GARP成长因子与 GARP
Core idea核心思想Buy fast growers whose price hasn't priced in the growth. GARP (growth at a reasonable price) is the balancing act between growth and value — PEG is its core tool.买"增长快、但价格还没透支增长"的公司。GARP（合理价格买成长）是成长与价值的平衡术，PEG 是它的核心工具。
- Operating cash-flow growth is the strongest growth signal (IR 0.9793, p=1.04%) — cash growth = growth plus fraud filter.经营现金流增速是最强的成长信号（IR 0.9793，p=1.04%）——现金增长 = 增长 + 排雷。
- Never test pure growth alone — the growth trap. Only test GARP composites (growth + quality + PEG constraint).纯成长不测——成长陷阱。只测 GARP 复合（成长 + 质量 + PEG 约束）。
- S007 verdict: the growth factor contributed +1.33pp of standalone increment (and is nearly uncorrelated with quality, |ρ|=0.108), yet the strategy itself failed F1/F4/F7/F8 — the factor survives, the strategy doesn't.S007 裁决：成长因子贡献了 +1.33pp 的独立增量（与质量分几乎不相关，|ρ|=0.108），但策略本体触发 F1/F4/F7/F8 被淘汰——因子留下，策略走人。
LESSON 07
Momentum & Reversal动量与反转
Core idea核心思想A-shares' special recipe: strong reversal, weak momentum. Retail investors' chase-and-panic overreaction is the root of the reversal anomaly — US momentum literature does not transplant.A股的特殊配方是反转强、动量弱——散户追涨杀跌的过度反应是反转异象的根。美股动量文献不可硬移植。
- 1-month reversal is real in our data (RankIC 0.0475, t=3.07, significant); 12-1M momentum is weakly inverted (RankIC -0.0177). Buying past winners ≈ buying a negative factor.1个月反转在我们的数据上是实的（RankIC 0.0475，t=3.07，显著）；12-1M 动量是弱反向（RankIC -0.0177）。买过去赢家 ≈ 买了一个负向因子。
- S006: a significant factor killed by 449% annual turnover (triggered F3). For reversal strategies, cost is the life-or-death line — next test (S010) tries quarterly rebalancing.S006：显著的因子死于年换手 449%（触发 F3）。对反转策略来说，成本是生死线——下一测（S010）尝试季度调仓降频。
- W-cut research: large orders are the micro-source of reversal (IR 2.51 vs 1.20) — on hold until tick-level data is available.W式切割研究：大单成交是反转的微观来源（IR 2.51 vs 1.20）——缺逐笔数据，暂缓。
How these notes are made这些笔记是怎么来的
Each lesson pairs external sources (sell-side research, academic papers, open-source quant projects) with our own backtests on 10 years of A-share data (2016–2026, 4,605 stocks). A lesson only "counts" when its claims survive our data — paper conclusions that contradict our tests are flagged, not cited.每课都把外部来源（券商研报、学术论文、开源量化项目）和我们自己的回测（10年A股数据，2016–2026，4605只股票）配对。只有经得起我们数据检验的结论才算数——与我们实证打架的论文结论会被标出，而不是被引用。
Raw factor test numbers live on the Factor Dashboard. Nothing here is investment advice.原始因子测试数据在因子表盘。这里没有任何内容构成投资建议。