Tick and Level-2 Order Book Data Sources

Level-2 data is an advanced data source in quantitative trading, containing order book depth, tick-by-tick trades, and more. This article compares Tick/L2 sources for China, Hong Kong, US, Japan, futures, FX, and crypto markets.


1. What Level-2 Order Book Data Contains

TypeExample Fields
Bid/Ask DepthBid1-Bid10 prices and quantities, Ask1-Ask10
Order Details (some platforms)Queue size and order count at each price level
Tick-by-tick Trades + Aggressor SideTrade price, direction (active buy or sell)
Order Book Movement IndicatorsInternal/external volume, bid-ask ratio, volume ratio, bid-ask spread, etc.

2. Level-2 Data Acquisition Channels

2.1 Exchange Official Authorization (Most Authoritative)

ExchangeHow to ObtainNotes
SSE / SZSE (China)Through Level-2 market data authorized service providersReal-time data requires paid authorization, tick/trade-by-trade, bid-ask queues
CFFEX (China Financial Futures Exchange)For futures and options Level-2 dataEssential for HFT options/arbitrage
HKEX (Hong Kong Exchange)Real-time quotes, 5+ levels of bid/askRequires connection to "HKEX Orion Market Data Platform"
US Stocks (NASDAQ, NYSE)NASDAQ TotalView / NYSE OpenBookProvides full-depth Order Book (subscription required)
Crypto Exchanges (e.g., Binance)Free API for order book snapshot and websocket real-time streamGenerally supports top 20 to full book

Note: In mainland China, Level-2 real-time data requires authorization through exchange-certified third-party data service providers, which is relatively expensive. Delayed data can be obtained through some channels.

2.2 Major Level-2 Data Service Providers

PlatformData TypeAvailabilityNotes
WindA-share Level-2 (depth + matching)PaidStructured interface and terminal visualization
Tonghuashun iFinDLevel-2 depth + tick-by-tickEnterpriseSome content requires separate paid authorization
JoinQuantJQData data service only⚠️ Live trading discontinuedYichuang-JoinQuant shut down all services at end of 2023; JoinQuant pivoted to running its own fund — do not rely on it as a live-trading channel
MyQuant (掘金)Real-time quotes + paper/live tradingPaidmyquant.cn, available to individuals
RiceQuant (米筐)Research/backtest/data APIsPaidInstitutional tier supports L2
Juling, HundsunProfessional private fund useEmbedded in OMSFinancial institution partnerships, expensive
Xueqiu, NetEase Finance, etc.Display only bid5/ask5No APIViewable but no full-depth data download

2.3 International Tick / Level-2 Sources

Before comparing vendors, separate the data granularity:

GranularityMeaningBest For
L1 / NBBO / SIPBest bid/ask, trades, consolidated quote feedsLive monitoring, ordinary execution, second/minute strategies
L2 / MBPMarket-by-price depth, such as 10 levels or full depth by priceExecution simulation, order book imbalance, short-horizon signals
L3 / MBOMarket-by-order events, order IDs, add/modify/cancel messagesQueue-position simulation, market making research, HFT research
MarketMain ChannelsBest FitNotes
US EquitiesNASDAQ TotalView, NYSE OpenBook, NYSE ArcaBook, IBKR market data subscriptions, Databento, Massive (formerly Polygon.io), AlpacaIndividual research through institutional tradingTotalView / OpenBook are exchange depth products. Databento is stronger for historical Tick/L2/L3 APIs. Massive and Alpaca are developer-friendly for L1, trades, quotes, and aggregates rather than full exchange depth.
US OptionsOPRA, Cboe DataShop, Databento OPRAOptions backtesting, volatility research, market making researchOPRA is the core consolidated options quote/trade source. Full-chain options data is large enough that storage and bandwidth need their own budget.
US FuturesCME / CBOT / NYMEX / COMEX direct feeds, Databento, IBKR, CQG, RithmicIntraday futures, CTA, execution researchFutures L2 usually carries exchange fees. Replay support, timestamp quality, and matching-engine proximity matter more than headline API price.
Hong Kong EquitiesHKEX OMD-C Standard / Premium / FullTick, LSEG, Bloomberg, TickerPlant, WindHong Kong live trading and microstructure researchPremium provides streaming market-by-price data. FullTick provides market-by-order data for local order book construction.
Japan EquitiesTSE licensed real-time feeds, FLEX Historical, JPXI 10-Level Order Book Historical Data, J-Quants DataCube, authorized vendorsJapan-market research and live tradingThese are separate products. FLEX Historical contains raw daily FLEX messages for reconstructing books. The 10-level historical dataset, launched June 29, 2026, is pre-processed from FLEX Historical and delivered through Snowflake data sharing. DataCube is a pay-per-use marketplace for selected downloadable datasets; do not assume it is the delivery channel for the full FLEX archive.
Europe / Global EquitiesLSEG Tick History, Bloomberg B-PIPE, ICE, SIX, dxFeed, Barchart, QuoteMediaGlobal multi-market institutional researchCoverage is broad but sales-led. Confirm exchange agreements, redistribution rights, historical depth, and timestamp granularity.
FXDukascopy, Cboe FX, EBS, LSEG, TrueFXFX backtesting, macro/CTA researchDukascopy is useful for free historical tick downloads. Institutional FX order books still require professional feeds.
CryptoBinance, OKX, Coinbase Advanced Trade API, ccxt, Tardis.devThe easiest market for individuals to learn Tick/L2 processingExchange WebSockets can reconstruct real-time local order books. Tardis.dev is better for historical L2/L3, cross-exchange replay, and research datasets.

Practical selection guide:

GoalStart WithAvoid at the Beginning
Learn order books and local book reconstructionBinance / Coinbase / OKX WebSocket, Dukascopy tick dataBloomberg, LSEG, direct exchange feeds
US equity strategy researchDatabento historical data, IBKR subscriptions, Massive / Alpaca L1 dataFull-market direct feeds
Hong Kong / Japan researchHKEX / JPX licensed data or authorized vendorsScraped web quotes or broker UI data
Execution simulationExchange L2/L3, Databento, HKEX FullTick, JPX real-time dataMinute bars or data without cancels/modifies
Display or redistributionExchange license plus explicit redistribution rightsPersonal broker subscriptions

Where can you get free Level 2 order book data?

For learning and local order book reconstruction, crypto exchanges such as Binance, Coinbase, and OKX provide free real-time depth over WebSocket, while Dukascopy offers useful historical FX tick exports. Free historical crypto depth is scarcer: OKX publishes L2 order book history, Binance's public archive has only aggregated futures depth, and Tardis.dev lets you download the first day of each month without an API key (details in 2.4). Full-depth US equity data is generally not truly free once exchange licensing and professional-use rules are included.

Which APIs provide historical Level 2 data?

For API-first historical depth, Databento covers US equities, futures, and options; Tardis.dev specializes in crypto L2/L3 replay; and official exchange or institutional feeds remain the main route for market-specific licensed archives. Always verify whether a dataset contains snapshots only or the add, modify, cancel, and trade events needed to reconstruct the book.

2.4 Crypto Level-2 order book data

Crypto is the easiest market in which an individual can read real-time L2 straight from the venue, for free. Two things are different from equities, though. Every exchange is its own book, with no consolidated feed across venues. And a free live stream gives you no history: whatever you did not record yourself, you have to download from an exchange archive or buy.

Live depth from the exchanges (official documentation, checked 2026-10-06):

ExchangeChannelDepthUpdatesAccess
Binance Spot<symbol>@depth<levels> (partial book)5, 10 or 20 levels1000 ms or 100 msPublic market data
Binance Spot<symbol>@depth (diff) + REST GET /api/v3/depth snapshotSnapshot up to 5,000 levels per side1000 ms or 100 msPublic market data
Coinbase Advanced Tradelevel2Snapshot, then incremental updatesEvery update; delivery of all updates is guaranteedPublic channel, no JWT needed
OKXbooks400 levelsSnapshot, then increments every 100 msPublic
OKXbooks-l2-tbt / books50-l2-tbt400 / 50 levelsTick-by-tick, every 10 msLogin and VIP4 or above
Bybit (spot and linear)orderbook1, 50, 200 or 1000 levels10, 20, 100, 200 ms respectivelyPublic stream
Krakenbook10, 25, 100, 500 or 1000 levelsSnapshot, then updates, with a CRC32 checksum of the top 10 levelsPublic; the level3 (individual orders) channel needs an API token

Historical crypto depth you can download (vendor and exchange pages, checked 2026-10-06):

SourceWhat it containsAccess
Binance public data (data.binance.vision)Spot: trades, aggregated trades and klines only, with no order book. USDⓈ-M and COIN-M futures add bookDepth (cumulative depth at ±0.2% to ±5% around mid, about every 30 seconds, judging from the files) and bookTicker (best bid/ask; the BTCUSDT USDⓈ-M files stop at 2024-03-30)Free download
OKX (historical data)L2 order book from March 2023. The market-data-history API lists 400-level books and 5000-level books (from 2025-11-01) for spot, futures, swaps and options, usually available at T+3; OKX notes that backfill is still in progressPublic download page and API
Tardis.devIncremental L2 updates, 25- and 5-level snapshots, trades, quotes, liquidations and more, recorded from the WebSocket feeds of 50+ exchangesFirst day of each month downloadable without an API key; published subscriptions start at $350/month (Academic tier, Perpetuals plan, quarterly or yearly billing)
Kaiko"Bids and asks": order book snapshots and updates as daily CSV filesCommercial vendor
Crypto Lake20-level depth snapshots per side across 10 exchangesCommercial vendor

What to watch for when you build on these:

  • Diff streams need a sequence check. Binance's documented procedure: buffer the stream, fetch a snapshot (and fetch again if it is older than the first buffered event), drop buffered events already covered by its lastUpdateId, and throw the book away and start over when an event's first update ID skips ahead. Its snapshot stops at 5,000 levels per side, so deeper levels stay unknown until they change.
  • Use the checks the exchange gives you. Kraken's checksum lets you prove your local top 10 matches theirs on every message.
  • "Top of book" is a separate, cheaper stream. If you only need best bid/ask, Binance bookTicker, OKX bbo-tbt or a 1-level Bybit book is enough.
  • Archives differ in what they keep. An aggregated depth band (bookDepth) or a fixed-level snapshot cannot replay queue changes; incremental L2 updates can. The acceptance test below applies to crypto archives too.

3. Acceptance test before buying a dataset

Do not buy from a coverage logo alone. Ask for one symbol-day containing the open, a halt or auction if available, and the close, then run this gate:

CheckPass conditionWhy it matters
SchemaSupplier states whether records are trades, MBP, or MBO“Tick” can mean trades only; it may contain no resting orders
Sequence integrityNo unexplained sequence gaps; recovery snapshots are documentedA single missed incremental update can corrupt the local book
Event coverageAdds, modifies/replaces, cancels, trades, resets and status messages are definedSnapshots alone cannot reproduce queue evolution
TimestampsExchange-event and receive timestamps are named, with timezone and precisionLatency studies fail when unlike clocks are compared
SymbologyPoint-in-time mappings, contract rolls and corporate actions are availableCurrent symbols create false histories
RightsResearch, derived-data, display and redistribution rights are written explicitlyTechnical access does not grant redistribution permission
ReplayRebuilt top-of-book reconciles with supplied snapshots/checkpointsValidates the parser and the delivered bytes together

Databento's schema documentation is a concrete example of why the distinction matters: MBO contains order-ID events across the book, MBP-10 contains price-aggregated updates at ten levels, and trades contain executions only. Nasdaq's market-data policies separately govern use and distribution. Treat schema sufficiency and licensing as two independent acceptance gates.

Once a sample is in hand, the next step is to rebuild the book from it and check the result against the vendor's own aggregation. Historical Level 2 Data in Practice: Rebuilding an Order Book from MBO Messages in Python does exactly that with a Databento MBO sample, a dependency-free replay script, and an event-by-event comparison against MBP-10.

4. Level-2 Data Costs (Approximate)

Platform / ChannelCost Model (Reference)Notes
Wind Level-2 (A-shares)CNY 20,000-50,000+/monthIncludes API access and usage license; usually institutional
MyQuant Level-2 (Live trading)CNY 3,000+/monthData stream available before/after market open
Retail Level-2 via Eastmoney/THS (A-shares)~CNY 298-998/year10-level depth + tick-by-tick for screen watching and research; no programmatic API or redistribution rights
JoinQuant Level-2DiscontinuedYichuang-JoinQuant shut down at end of 2023; JoinQuant pivoted to a private fund
HKEX OMD / vendor L2Exchange license fee plus vendor feeStandard, Premium, and FullTick contain different data; FullTick has the largest volume
US exchange depth feedsExchange-by-exchange subscriptions; professional and non-professional rates differBrokers such as IBKR list NASDAQ TotalView-OpenView, NYSE OpenBook, NYSE ArcaBook, and related products separately
DatabentoHistorical data is metered; live data is included in paid plansGood API-first path for US equities, futures, and options Tick/L2/L3; start with small samples to estimate cost
Massive (formerly Polygon.io) / AlpacaSubscription plans or brokerage-account data accessUseful for developer onboarding and L1/trades/quotes; do not treat them as the primary complete US equity L2 source
Tardis.devSubscription/API/download modelCrypto historical Tick, L2/L3, funding, liquidations, and cross-exchange research data
DukascopyFree historical exportUseful for FX tick learning and backtesting; not a substitute for institutional ECN order book data

Pricing changes with exchange, user status (professional / non-professional), redistribution rights, and commercial display use. Budget for exchange licenses, number of end users, storage, replay, compliance audit, and redistribution rights, not just the API subscription fee.

Research updates

Get the next field note by email

At most two emails a month. Choose Agent systems, Quant systems, or both—only new engineering notes, data updates, and substantial essays.

Send me

By subscribing, you agree to receive the selected updates. Every email includes an unsubscribe link. Privacy

5. FPGA and DMA (Market Access)

5.1 What is FPGA?

FPGA (Field Programmable Gate Array) is a programmable hardware chip that can directly execute specific tasks, such as parsing market data, making decisions, and sending orders, without OS or CPU involvement.

Role in Quantitative/High-Frequency Trading:

TaskFPGA Capabilities
Market Data ProcessingNanosecond-level parsing of Level-2 data, Order Book construction
Signal DecisionExecute strategy decisions directly in hardware
Order MatchingSend trading instructions, connect directly to trading gateway
Latency ReductionReduce and stabilize selected processing stages; end-to-end latency still includes feed, network, gateway, venue, and measurement overhead

FPGA logic can reduce latency and jitter in narrowly defined stages, but no hardware choice guarantees a particular end-to-end latency or profitability. Measure the complete path under production-like load.

5.2 What is Direct Market Access (DMA)?

DMA refers to technology where trading systems bypass traditional broker front-end systems and send trading requests directly to the exchange matching system (or trading gateway).

TypeDescription
DMA (Direct Market Access)Electronic access through a broker's infrastructure to a venue; broker and venue risk controls still apply
SA (Sponsored Access)Client access under a sponsoring broker's membership and regulatory controls
Low-latency sponsored accessArchitecture optimized for latency while retaining required pre-trade controls, monitoring, and broker responsibility

Some latency-sensitive strategies use DMA or sponsored access, but profitability depends on edge, queue position, costs, capacity, controls, and competition—not the access label alone.

5.3 DMA + FPGA Combination

In HFT scenarios,

DMA provides "channel access," FPGA provides "speed and decision-making power"

Institutions deploy at Equinix, @Tokyo, and other exchange-adjacent locations:

  • FPGA cards (for order book + strategy)
  • DMA lines (shortest fiber/microwave routes)
  • Risk control services (self-managed or minimized intermediaries)

CIIS Sample L2 Data https://www.ciis.com.hk/hongkong/sc/historicaldata1/sampledata/index.shtml

Shanghai Stock Exchange L2 https://english.sse.com.cn/markets/dataservice/products/

Shenzhen Stock Exchange https://www.szse.cn/market/stock/indicator/index.html

SZSE L2 Authorization http://www.szsi.cn/cpfw/fwsq/hq/yw-2.htm

Hong Kong Stock Exchange https://www.hkex.com.hk/Services/Market-Data-Services/Real-Time-Data-Services/Overview/Real_time-Datafeeds?sc_lang=en

HKEX OMD-C https://www.hkex.com.hk/Services/Market-Data-Services/Infrastructure/HKEX-Orion-Market-Data-Platform-Securities-Market-OMD-C?sc_lang=en

NASDAQ TotalView https://www.nasdaq.com/solutions/data/equities/nasdaq-totalview

NYSE Real-Time Market Data https://www.nyse.com/market-data/real-time

IBKR Market Data Pricing https://www.interactivebrokers.com/en/pricing/market-data-pricing.php

Databento Equities https://databento.com/equities

Databento Historical Metered Pricing https://databento.com/docs/api-reference-historical/basics/metered-pricing

Massive Rebrand from Polygon.io https://massive.com/blog/polygon-is-now-massive

Alpaca Market Data https://alpaca.markets/data

Coinbase Advanced Trade WebSocket https://docs.cdp.coinbase.com/coinbase-app/advanced-trade-apis/websocket/websocket-channels

Binance WebSocket Depth Stream https://developers.binance.com/docs/binance-spot-api-docs/web-socket-streams

Tardis.dev https://tardis.dev/

Dukascopy Historical Data Export https://www.dukascopy.com/swiss/english/marketwatch/historical/

JPX / TSE Real Time Market Data https://www.jpx.co.jp/english/markets/paid-info-equities/realtime/index.html

Cite this chapter
Zhang, Wayland (2026). Tick & Level-2 Order Book Data: Databento, HKEX, IBKR (2026). In AI Quantitative Trading: From Zero to One. https://waylandz.com/quant-book-en/Tick-and-L2-Order-Book-Data-Sources/
@incollection{zhang2026quant_Tick-and-L2-Order-Book-Data-Sources,
  author = {Zhang, Wayland},
  title = {Tick & Level-2 Order Book Data: Databento, HKEX, IBKR (2026)},
  booktitle = {AI Quantitative Trading: From Zero to One},
  year = {2026},
  url = {https://waylandz.com/quant-book-en/Tick-and-L2-Order-Book-Data-Sources/}
}