No survivorship shortcut
Use the historical universe that actually existed on the as-of date.
REPRODUCIBLE FACTOR MODEL可复现因子模型
Free Quantitative Stock Screener免费量化选股器
Turn a stock shortlist into an auditable pipeline with fixed inputs, factor rules, and evidence.
用固定输入、因子规则和证据链，把候选股票名单变成可审计的流程。
TL;DR先看筛选框架
Goal
Convert a defined, investable universe into a smaller research list using rules that can be rerun with the same data and date.
Useful factors
Combine economically distinct value, quality, growth, momentum, risk, and liquidity signals instead of stacking several versions of one idea.
Output
Show raw values, normalized scores, exclusions, sector exposure, and the reason each candidate passed—not only a mysterious composite rank.
Verification
Check point-in-time definitions, filing availability, outliers, missing values, rebalancing turnover, and transaction-cost sensitivity.
目标
用能够在相同日期和数据上重复运行的规则，把明确且可投资的证券池缩小为研究候选名单。
常用因子
组合经济含义彼此不同的估值、质量、成长、动量、风险和流动性信号，避免叠加同一种观点的多个变体。
结果含义
展示原始值、标准化分数、排除原因、行业敞口和每只股票入选理由，而不是只给一个无法解释的综合排名。
核验重点
检查时点口径、申报可用日、异常值、缺失值、调仓换手和交易成本敏感性。
A screener is a reproducible pipeline筛选器应是一条可复现的流水线
Universe
Fix market, security type, liquidity, and as-of date.
Raw factors
Collect value, quality, growth, momentum, risk, and liquidity inputs.
Normalize
Winsorize, rank, or z-score with a declared peer group.
Combine
Apply weights and missing-value policy without silent substitutions.
Verify
Return raw inputs, scores, exclusions, and source timestamps.
证券池
固定市场、证券类型、流动性条件和截至日期。
原始因子
收集估值、质量、成长、动量、风险和流动性输入。
标准化
明确同类组后，再做缩尾、排名或 Z-score。
合成得分
按公开权重和缺失值规则计算，不暗中补值。
核验
返回原始输入、得分、剔除原因和来源时间戳。
Make the score auditable让评分可以审计
A ranking is defensible only when the universe, as-of date, factor definitions, normalization, weights, exclusions, and source revisions are stored with the result.
只有把证券池、截至日期、因子定义、标准化方式、权重、剔除项和数据修订一并保存，排名才经得起复核。
Factor scorecard因子评分表
| Factor | Example inputs | Common trap |
|---|---|---|
| Value | Earnings yield, book-to-price, FCF yield | Mixing negative and positive denominators |
| Quality | ROIC, margins, accruals, leverage | Using incomparable fiscal periods |
| Growth | Revenue and EPS growth | Ignoring base effects and acquisitions |
| Momentum | 6–12 month return, revisions | Look-ahead and stale prices |
| Risk | Volatility, beta, drawdown | Changing windows between names |
| Liquidity | Dollar volume, spread, turnover | Using average volume without price context |
| 因子 | 示例输入 | 常见陷阱 |
|---|---|---|
| 估值 | 盈利收益率、账面市值比、自由现金流收益率 | 把正负分母直接混排 |
| 质量 | ROIC、利润率、应计项目、杠杆 | 财务期间不可比 |
| 成长 | 收入和 EPS 增长 | 忽略低基数与并购影响 |
| 动量 | 6–12 个月收益、预期修正 | 前视偏差和陈旧价格 |
| 风险 | 波动率、Beta、回撤 | 不同股票使用不同窗口 |
| 流动性 | 成交金额、价差、换手率 | 只看成交量而忽略价格 |
Free quantitative stock screener factors that matter免费量化选股器应覆盖的核心因子
A multi-factor stock screener is useful when each rule has a clear purpose. Value asks what you pay, quality examines business strength, growth measures change, momentum captures price behavior, while risk and liquidity test whether a candidate is practical to research. MSCI’s factor investing overview identifies value, quality, momentum, size, yield, and low volatility as established factor categories.
多因子选股工具的每条规则都应有明确目的：估值衡量价格与基本面的关系，质量观察企业经营稳健性，成长衡量收入和利润的变化，动量反映价格行为，风险与流动性则帮助排除不适合继续研究的标的。MSCI 的因子投资概览列出了估值、质量、动量、规模、收益率与低波动等常见因子类别。
Raw factor values are rarely comparable across the whole market. Banks, software companies, utilities, and early-stage firms have different accounting structures and distributions. A credible screen states whether it ranks within sector or industry, how it winsorizes extreme values, whether it converts values to percentiles or z-scores, and how it combines scores when one input is missing.
原始因子值通常不能直接横跨全市场比较。银行、软件公司、公用事业和早期企业的会计结构与数值分布并不相同。可靠的筛选器应说明是否在行业内排名、如何缩尾处理极端值、使用百分位还是标准分，以及某个输入缺失时如何合成总分。
Value and quality stock screener
Compare P/E, price-to-book, free-cash-flow yield, profitability, leverage, and cash-flow consistency. Definitions and sector norms matter.
Growth and momentum stock screener
Use revenue or earnings growth with price momentum over a stated lookback. Avoid treating recent performance as a forecast.
Risk and liquidity filters
Market capitalization, trading volume, volatility, drawdown, and leverage can prevent an attractive score from hiding practical constraints.
Rankings versus hard filters
Hard filters exclude stocks at a threshold; percentile ranks preserve more candidates and make cross-factor trade-offs easier to inspect.
Standardization and sector neutrality
Winsorize or cap extreme observations before ranking, then compare percentile or z-score behavior. Sector-relative ranks can reduce structural industry bets, but the chosen grouping and minimum group size must be explicit.
Factor overlap, turnover, and cost
Measure correlation between signals and how often candidates change. A slightly stronger paper score may be less useful if it creates high turnover, wide-spread trades, or repeated exposure to the same underlying factor.
估值与质量因子筛选
可比较市盈率、市净率、自由现金流收益率、盈利能力、杠杆和现金流稳定性，同时注意口径与行业差异。
成长与动量因子筛选
把收入或利润增速与明确周期的价格动量结合使用，不把近期表现当作未来收益预测。
风险与流动性过滤
市值、成交量、波动率、回撤与杠杆可避免高分掩盖实际交易和研究限制。
排名与硬性条件
硬性条件按阈值排除股票；百分位排名保留更多候选，便于观察不同因子之间的取舍。
标准化与行业中性
排名前应缩尾或限制极端观测，再比较百分位或标准分表现。行业内排名可以减少结构性行业押注，但必须明确分组方式和最小样本数。
因子重叠、换手与成本
衡量信号之间的相关性以及候选更换频率。纸面分数稍高的组合，如果带来高换手、大点差交易，或反复暴露于同一底层因子，实际价值可能更低。
Bias controls偏差控制
No future fundamentals
Apply filing availability dates, not only fiscal-period labels.
No silent imputation
Expose null policy and the count of missing inputs.
No unexplained turnover
Report why names entered or left the shortlist.
避免幸存者偏差
使用截至当时真实存在的历史证券池。
避免未来财务数据
按披露可用日期处理，而不是只看财报期间。
禁止静默填充
公开空值规则和缺失输入数量。
解释名单变化
说明股票进入或离开候选池的原因。
How to compare free factor stock screeners如何比较免费的多因子选股工具
| Check检查项 | What to verify核验内容 | Why it matters为什么重要 | Warning sign风险信号 |
|---|---|---|---|
| Market coverage市场覆盖 | Exchanges, security types, active and delisted stocks.交易所、证券类型、在市与退市股票范围。 | The starting universe changes every result.初始股票池会改变全部结果。 | Coverage is not disclosed.未说明覆盖范围。 |
| Factor definitions因子定义 | Formula, reporting period, units, and missing-value handling.公式、报告期、单位和缺失值处理。 | Similar labels may use different calculations.相似名称可能采用不同算法。 | Opaque scores without methodology.只有分数，没有方法。 |
| Data freshness数据时效 | Price delay, filing update time, and refresh schedule.价格延迟、财报更新时间和刷新频率。 | Stale inputs can alter a shortlist.过期输入会改变候选名单。 | No timestamp or source.没有时间戳或来源。 |
| Free-tier limits免费版限制 | Filters, exports, saved screens, alerts, and result caps.筛选项、导出、保存条件、提醒和结果上限。 | “Free” may describe only a trial or partial workflow.“免费”可能只是试用或部分功能。 | Paywall appears after rules are entered.输入规则后才出现付费墙。 |
| Normalization标准化 | Outlier cap, percentile or z-score, sector grouping.极值限制、百分位或标准分、行业分组。 | Raw units and sector structures are not comparable.原始单位与行业结构不可直接比较。 | A single outlier dominates the score.单个极端值主导总分。 |
| Point-in-time data时点数据 | Filing availability, restatements, old universes, delistings.申报可用日、重述、历史证券池和退市标的。 | Controls look-ahead and survivorship bias.控制前视偏差与存续偏差。 | Only today's values are available.只能获得今天的数值。 |
| Portfolio realism组合现实性 | Turnover, capacity, spread, rebalance frequency, concentration.换手、容量、点差、调仓频率与集中度。 | A screen must survive implementation costs.筛选结果需要经受实际执行成本。 | Only backtested return is shown.只展示回测收益。 |
How to use a free quantitative stock screener如何使用免费量化选股器建立候选股票池
1. Define the stock universe
Choose market, exchange, security type, minimum size, and minimum liquidity before selecting factors. This prevents accidental comparisons across unlike securities.
2. Add a small set of independent factors
Start with one or two rules from value, quality, growth, and momentum. Highly correlated filters can create false confidence without adding information.
3. Review the shortlist, not just the score
Compare factor values, sector concentration, missing data, outliers, and the reason each stock passed. Save the exact rules and date.
4. Verify with current primary data
Read current company filings and confirm recent figures before making any decision. Check the issuer, reporting period, units, restatements, and source timestamp.
5. Normalize and combine without hiding the inputs
Apply documented outlier handling, rank within the intended peer group, and combine factors with explicit weights. Preserve every raw value, transformed score, missing-value decision, and exclusion reason.
6. Test stability and implementation cost
Shift thresholds, lookbacks, and rebalance dates slightly. Compare candidate overlap, sector weights, turnover, spreads, and performance after conservative costs; a robust research screen should not collapse after a tiny parameter change.
7. Freeze a point-in-time research snapshot
Store the eligible universe, source dates, filing availability, factor formulas, code version, ranking, and rejected rows. Rebuilding the screen later should reproduce the same list without consulting data published afterward.
1. 定义股票池
先选择市场、交易所、证券类型、最低市值与最低流动性，再添加因子，避免比较性质不同的证券。
2. 添加少量相互独立的因子
从估值、质量、成长和动量中各选一至两个规则。高度相关的条件可能增加确定感，却没有增加有效信息。
3. 复查候选名单而不是只看分数
比较各因子数值、行业集中度、缺失数据、异常值以及每只股票入选的原因，并保存筛选规则和日期。
4. 用最新的一手数据核验
决策前阅读最新公司披露并确认数值，检查发行人、报告期、单位、重述情况与来源时间戳。
5. 标准化并合成，但不隐藏原始输入
按文档处理异常值，在目标同类组内排名，再用明确权重合成因子；同时保留原始值、转换后分数、缺失值决策和排除原因。
6. 检验稳定性与执行成本
小幅移动阈值、回看周期和调仓日期，比较候选重合度、行业权重、换手、点差，以及扣除保守成本后的结果。可靠的研究筛选不应因参数轻微变化就完全失效。
7. 冻结历史时点研究快照
保存合格证券池、来源日期、申报可用时间、因子公式、代码版本、排名和被排除记录。之后重建时，不应依赖当时尚未发布的数据。
Worked example: turn raw factors into an auditable score实算示例：把原始因子转成可核验分数
Suppose Stock A is ranked only against eligible companies in its industry. After clipping each factor at documented cross-sectional limits and orienting every score so “higher is better,” it has value z = 1.2, quality z = 0.6, low-volatility z = 0.4, and no valid momentum observation. With target weights of 30% value, 30% quality, 25% momentum, and 15% low volatility, do not turn missing momentum into zero. Reweight the three available factors: (0.30×1.2 + 0.30×0.6 + 0.15×0.4) ÷ 0.75 = 0.80. Return the 0.80 composite together with the missing-factor flag, original 75% weight coverage, peer group, raw values, clipping limits, and as-of time.
假设股票 A 只与同一行业内符合条件的公司比较。各因子先按已记录的截面边界缩尾，再统一方向，使“分数越高越好”。处理后，股票 A 的估值因子为 z = 1.2、质量因子为 z = 0.6、低波动因子为 z = 0.4，动量数据无效。目标权重分别为估值 30%、质量 30%、动量 25%、低波动 15%。此时不能把缺失的动量直接记为 0，而应对其余三个有效因子重新归一化：(0.30×1.2 + 0.30×0.6 + 0.15×0.4) ÷ 0.75 = 0.80。返回综合分 0.80 时，还应同时给出因子缺失标记、原始权重覆盖率 75%、同类组、原始值、缩尾边界和数据截至时间。
Translate rank quality into implementable results. If a rebalance replaces 35% of portfolio notional and the conservative one-way cost is 20 basis points per traded dollar, the immediate cost estimate is 0.35 × 20 bp = 7 bp per rebalance, before taxes or market impact. Report gross and net results, turnover, spread assumptions, and capacity together. Historical tests must also retain securities that later delisted, use fundamentals only after their real publication time, and apply splits, dividends, symbol changes, and other corporate actions without rewriting the past universe.
排名有效不等于策略可以低成本执行。 如果一次调仓替换了组合名义金额的 35%，每一元成交金额采用保守的单边成本 20 个基点，那么该次调仓的直接成本约为 0.35 × 20 bp = 7 bp，尚未包含税费和市场冲击。结果应同时披露毛收益、净收益、换手率、点差假设和容量。历史测试还必须保留后来退市的证券，只在基本面数据真实发布后使用，并正确处理拆股、分红、代码变更及其他公司行动，不能用今天仍存续的股票池改写过去。
Connect the model to inspectable data把模型连接到可检查的数据
Use QVeris to inspect candidate market and fundamental-data capabilities before they feed a ranking model. Continue with FMP provider details, then confirm the calling contract in the QVeris REST API reference.
可通过 QVeris 检查候选行情与基本面能力，再决定是否将其输入排名模型。 可继续打开FMP 服务商详情，并通过QVeris REST API 参考核对调用契约。
FAQ常见问题
What is a quantitative stock screener?
It applies measurable rules to a defined stock universe and returns a smaller list for further research.
Can I screen stocks by multiple factors for free?
Some tools offer free multi-factor filters, but market coverage, delayed data, exports, saved screens, and advanced factors may be limited.
Which factors should a beginner use?
Start with a small, explainable mix such as valuation, profitability, growth, momentum, leverage, and liquidity. Learn each definition before combining rules.
Does a stock screener recommend what to buy?
No. It narrows a universe. Verify filings, data freshness, risks, valuation context, and personal suitability independently.
Should factors be ranked across the whole market?
Not automatically. Accounting and valuation distributions differ by industry. Compare whole-market and sector-relative ranks, disclose the grouping method, and inspect whether neutrality removes information you intended to keep.
How should extreme and missing values be handled?
Use a documented rule: cap or winsorize plausible extremes, quarantine calculation errors, and distinguish not applicable from unavailable. Never convert every missing value to zero without testing the economic meaning.
Why can a strong factor screen fail in practice?
Look-ahead bias, survivorship bias, crowded exposures, unstable thresholds, high turnover, poor liquidity, and ignored costs can turn an attractive historical shortlist into an unusable workflow.
什么是量化选股器？
它对明确的股票池应用可计算规则，返回更小的候选名单，供用户继续研究。
免费工具可以进行多因子选股吗？
部分工具提供免费多因子筛选，但可能限制市场覆盖、数据时效、导出、条件保存或高级因子。
新手量化选股应使用哪些因子？
从少量可解释指标开始，例如估值、盈利能力、成长、动量、杠杆和流动性，组合前先理解口径。
选股器会推荐应该买什么股票吗？
不会。它只负责缩小范围，仍需独立核验公司披露、数据时效、风险、估值背景与个人适用性。
因子应该直接在全市场排名吗？
不一定。各行业的会计与估值分布不同，应比较全市场与行业内排名，说明分组方法，并检查中性化是否删除了原本希望保留的信息。
异常值和缺失值应该怎样处理？
应采用明确规则：限制合理极值、隔离计算错误，并区分不适用与暂无数据。不能不考虑经济含义就把所有缺失值补成零。
为什么历史表现很好的因子筛选可能无法落地？
前视偏差、存续偏差、拥挤暴露、阈值不稳定、高换手、流动性不足和忽略成本，都可能让漂亮的历史候选名单在真实执行中失效。