Stock Data API
for Backtesting用于回测的
股票数据 API
Choose historical stock data that preserves what was actually knowable on each trading date—not merely a clean chart of today’s survivors.
选择能还原每个交易日真实可用信息的历史股票数据，
而不只是由当前幸存股票拼成的一张漂亮走势图。
TL;DR核心摘要
Bars need exchange calendars, corporate actions, stable identifiers, session rules, and an explicit adjustment method.
Membership, fundamentals, and classifications must reflect what was available on the simulated decision date.
Include delisted names, realistic signal-to-trade timing, liquidity filters, fees, spreads, and borrow constraints.
Store raw responses, request parameters, schema version, adjustment mode, and dataset retrieval time.
K 线还要配合交易日历、公司行动、稳定标识、交易时段规则和明确的复权方法。
成分股、基本面和行业分类必须对应模拟决策日当时真正可用的信息。
保留退市股票，并纳入信号到成交的时间差、流动性、手续费、价差及融券限制。
保存原始响应、请求参数、数据结构版本、复权模式和数据获取时间。
What a backtesting data API must provide回测数据 API 必须提供什么
A fast API is not automatically a good backtesting API. Research quality depends on what the dataset knew at each historical moment, how prices were adjusted, and whether dead or renamed securities remain available.
速度快不等于适合回测。研究质量取决于数据在每个历史时点包含的信息、价格如何复权，以及退市或更名证券是否仍可获取。
Confirm interval coverage, exchange timezone, session boundaries, holiday and early-close calendars, and whether a timestamp marks bar open or close. A daily close cannot be used for an order filled at that same close unless the signal existed earlier.
Ask whether open, high, low, close, and volume are adjusted for splits and cash distributions. Keep raw bars and a separate corporate-action table whenever possible; this makes the transformation inspectable and prevents dividend cash flows from being counted twice.
A ticker is not a permanent identifier. Mergers, share-class changes, exchange moves, reused symbols, and spin-offs require a durable security ID plus effective dates. Otherwise unrelated histories can be joined into one fictional instrument.
Current index members or currently listed symbols create survivorship bias. A credible dataset preserves historical membership, IPO dates, inactive securities, delisting dates, and final returns or cash consideration when available.
A fiscal period end is not the date investors learned the number. For factor research, require filing, publication, or accepted timestamps and preserve later restatements as separate versions rather than silently rewriting history.
确认时间粒度、交易所时区、交易时段边界、节假日与提前收市安排，还要弄清时间戳代表 K 线开始还是结束。如果信号直到收盘后才生成，就不能假设订单已按当天收盘价成交。
确认开高低收和成交量如何处理拆股与现金分红。条件允许时，同时保留原始 K 线和独立的公司行动表，便于审查转换过程，也能避免把分红收益重复计算。
股票代码并不是永久标识。并购、股类调整、转板、代码复用和分拆都需要稳定的证券 ID 与生效日期，否则两只不同证券的历史可能被错误拼接。
直接使用当前指数成分或当前上市股票会产生幸存者偏差。可靠数据应保留历史成分、上市日期、失效证券、退市日期，以及可获得的最终收益或现金对价。
财报期末并不等于市场得知数据的日期。因子研究应使用申报、发布或接收时间，并将后续重述保存为不同版本，而不是静默覆盖历史。
How to compare stock data APIs for backtesting如何比较用于回测的股票数据 API
| Criterion标准 | What to verify检查内容 | Why it matters重要性 | Common risk常见风险 |
|---|---|---|---|
| Coverage覆盖范围 | Markets, instruments, history depth, and intervals.市场、品种、历史深度与时间粒度。 | Your actual strategy universe.策略实际使用的标的池。 | A long date range may still exclude delisted names.日期很长也可能不含退市证券。 |
| Integrity数据完整性 | Adjustments, corrections, identifiers, and calendars.复权、修订、标识符与交易日历。 | Avoiding false returns and broken joins.避免虚假收益和关联错误。 | Marketing labels rarely explain methodology.宣传标签通常不解释处理方法。 |
| Access访问能力 | Rate limits, bulk export, pagination, and latency.限流、批量导出、分页与延迟。 | Large universe and parameter sweeps.大规模标的池和参数扫描。 | Per-request APIs can make research slow or costly.按请求接口可能拖慢研究并增加成本。 |
| Reproducibility可复现性 | Snapshots, versions, changelogs, and licensing.快照、版本、变更日志与许可。 | Audits, collaboration, and production parity.审计、协作与生产一致性。 | Live corrections can change old results.在线修订会改变旧回测结果。 |
A reproducible API-to-backtest workflow可复现的 API 回测流程
Define the universe, date range, bar interval, timezone, sessions, adjustment mode, identifiers, required events, and the exact information cutoff used by the strategy.
Include an IPO, a delisting, a split, a large dividend, a ticker change, an exchange holiday, and an illiquid name. Happy-path megacaps reveal very little about dataset integrity.
Map the provider schema into your internal model, but retain original field names, response files, provider timestamps, and transformation code. Never infer a zero price or zero volume from a missing observation.
Lag fundamentals and classifications until their actual publication time. Generate signals from information available before the order and fill them only in a later executable bar.
Model commissions, spread, slippage, volume participation, market impact, rejected orders, short availability, and corporate-action cash flows separately from the alpha model.
Store a snapshot identifier or content hash with code revision, parameters, calendar version, provider, retrieval time, and validation report. The same run should consume the same bytes months later.
先定义标的池、日期范围、K 线粒度、时区、交易时段、复权模式、证券标识、必需事件，以及策略每一步的信息截止时间。
样本应包含 IPO、退市、拆股、大额分红、代码变更、休市日和低流动性股票。只测大型活跃股票，很难暴露数据质量问题。
可以把供应商字段映射到内部模型，但要保留原始字段名、响应文件、供应商时间戳和转换代码。缺失观测不能擅自解释为零价格或零成交量。
基本面和分类数据要等到实际发布时间后才能使用。信号只能读取下单前已知的信息，成交则放到后续真正可执行的 K 线中。
将佣金、买卖价差、滑点、成交量占比、市场冲击、拒单、融券可得性和公司行动现金流，与策略信号分开建模。
保存数据快照 ID 或内容哈希，同时记录代码版本、参数、日历版本、供应商、获取时间和验证报告。数月后复跑时，应读取完全相同的数据字节。
Use QVeris to inspect market data capabilities使用 QVeris 检查市场数据能力
QVeris helps developers and agents inspect callable market-data capabilities before provider-specific code spreads through a research stack. Start in the QVeris Playground to inspect schemas and sample responses, then use the QVeris documentation when turning the validated request into a repeatable pipeline.
QVeris 可帮助开发者和 Agent 在供应商专用代码进入研究栈之前，先检查可调用的市场数据能力。可以先在 QVeris Playground 中核对字段结构和样例响应，再参考 QVeris 文档，把验证过的请求接入可重复运行的数据管道。
- Search for historical price, corporate-action, fundamentals, and point-in-time data capabilities.
- Inspect schemas and required parameters before writing provider-specific integration code.
- Keep your own validation, caching, licensing checks, and research audit trail.
- 搜索历史价格、公司行动、基本面和时点数据能力。
- 编写供应商专用代码前，先检查数据结构和必需参数。
- 仍需自行完成验证、缓存、许可检查和研究审计记录。
FAQ常见问题
Most strategies need OHLCV bars, exchange calendars, corporate actions, stable identifiers, and a historically accurate security universe.
It depends on the strategy. Whatever you choose, apply one documented method consistently and avoid mixing adjusted and raw fields.
Removing failed companies leaves a universe of winners and usually inflates historical performance. Delisted names also need an explicit terminal-return policy.
You can, but mutable upstream history harms reproducibility. A versioned local snapshot is usually safer for research; refresh it through a controlled ingestion job.
多数策略需要 OHLCV K 线、交易日历、公司行动、稳定标识和历史准确的证券池。
取决于策略。无论选择哪种方法，都应保持一致并避免混用复权和原始字段。
剔除失败公司后，标的池只剩“赢家”，通常会夸大历史表现。对退市标的还应明确最终收益如何计算。
可以，但供应商对历史数据的修订会破坏复现。研究阶段更适合使用带版本的本地快照，再通过受控任务更新。
References and next steps参考资料与下一步
Alpha Vantage market-data API documentation
Financial Modeling Prep API documentation
Inspect market-data calls in QVeris Playground
Build a repeatable workflow with QVeris Docs
Alpha Vantage 市场数据 API 文档
Financial Modeling Prep API 文档
在 QVeris Playground 中检查市场数据调用
参考 QVeris 文档搭建可重复的数据流程