L3

L3 market data is the most comprehensive form of market data available, delivering information on all active orders in the market. It includes not only the best bid and offer (L1) and the depth of orders at each price level (L2), but also the ability to see each individual order in the book.

L3 data typically includes all trades, fills, adds, cancels, modifies (or replaces), book clear events, and, depending on the venue and dataset, other special order events.

The term "Level 3" originates from the Nasdaq Level 3 service, which was a specialized service available to registered Nasdaq market makers that provided them the ability to enter, modify, and cancel quotes, as well as confirm trades.

Today, however, the term has been repurposed by most vendors and trading participants to refer to market data with full order book granularity. This draws from the idea that L1, L2, and L3 correspond to increasing granularity in the data. This shift in terminology reflects the growing trend of markets offering more granular forms of market data, like L2 and L3, to meet the critical needs of algorithmic trading.

The main distinguishing feature of L3 data is that it provides some way to identify every order and subsequent updates to each order. As such, every market data event is keyed by a unique order ID, sometimes also called an order reference number. Beyond this, the implementation of L3 data differs slightly from one trading venue to another.

Most L3 feeds, like Nasdaq TotalView, implicitly indicate queue position based on the timestamp that an order is added. However, some L3 feeds like CME's MDP 3.0 feed provide a separate tag 37707-MDOrderPriority number to position the order against other orders of the same price.

Most L3 feeds rely on the implicit assumption that an order retains its queue position when its size decreases, but loses its queue position when its size increases. However, a feed may make this explicit by including a flag in order modification messages that indicates whether queue position is retained or lost—IEX's DEEP+ feed is one example of this.

L3 feeds vary on the extent to which the taker side and passive side of a trade is published. Most L3 feeds, like Nasdaq TotalView, only publish trade executions from the perspective of the passive side of each match, and the client has to infer the size of the aggressing order from individual fills when an aggressor is matched against multiple orders. CME's MDP 3.0 feed provides both the size of the aggressing order and the individual fills in each match. Moreover, some L3 feeds, like Coinbase's, will indicate the price of the aggressing order, and thus also reflect whether the aggressor had submitted a market order or a marketable limit order.

Depending on the venue, other events may be included with L3 data, such book clear events, imbalance, cross trades, or indications of fill-or-kill activity.

Compared to L2 data, which only provides a limited number of price levels, a L3 feed has to show events at every price level. This means that it's impractical for the wire protocol of a L3 feed to disseminate book updates with snapshots of every level like some L2 feeds, which would require significant bandwidth.

Instead, a L3 data feed usually publishes every order update as a trade (execution or fill), add, cancel or modify (replace) event. The user of a L3 data feed is then required to construct a limit order book representation on client side to keep track of the state of the order book and compute useful values like the best bid and offer.

A primary benefit of L3 data is that it allows a participant to simulate fills of passive orders more precisely, allowing them to create a better backtesting engine. In contrast, a participant that only has access to L2 data will be required to estimate the queue position of each order based on a model.

A common misconception is that L3 data only benefits low-latency or high-frequency traders. A trading participant that's not latency-sensitive may still benefit from improved simulation or backtesting accuracy.

Another main benefit of L3 data is that it exposes an additional class of signals that cannot be constructed from L1 or L2 data. For example, a consumer of L3 data can determine the exact duration that each order has been resting on the order book, which allows them to construct a trading signal based on staleness of each order.

Despite L3 data feed providing more information and activity at more price levels than a L2 data feed, L3 feed protocols tend to be more optimized than L2 feed protocols, providing the user with an additional latency advantage. (Counterintuitively, L2 data may require more storage than L3 data, depending on the exact implementation.) Some trading venues may also deliberately offer more granular feeds on a faster path in their infrastructure, further extending the latency advantage.

In most cases, L3 data indicates the highest level of granularity and most comprehensive type of data available on a market. However, it's possible to obtain data that shows more than just every individual order.

Certain trading venues and platforms may offer premium data services that provide even more information than what's available on their standard L3 feed. For example, Eurex offers an optional service called the Eurex IOC Liquidity Indicator for Options, on top of its standard L3 Enhanced Order Book (EOBI) feed.

Many OTC markets don't operate in an anonymized manner, and may expose information that's more granular than a typical L3 feed. For example, on some spot FX platforms, dealers or liquidity providers may have privileged information about the counterparty of each order or trade, allowing them to profile the toxicity of their counterparty.

Moreover, some vendors, like Databento and Pico, sell packet captures (PCAPs) of L3 feeds, which provide an additional capture timestamp near the handoff or boundary switch, over any timestamps natively embedded in the data. The additional timestamp is extremely valuable to participants that practice algorithmic trading or high-frequency trading, as it yields additional information such as the transmission time to and from the matching engine and matching engine latency. Packet captures also preserve the original packet sequencing and message order, which could be used to construct additional trading signals based on specific microstructure.

The existence of L3 data and other privileged information, like those described above, is a significant driver of information asymmetry in financial markets. Often, exchange operators and trading venues will charge a significant premium for L3 data and other such types of data feeds, hence participants with more resources tend to be more informed than participants that cannot afford these feeds.

Not all trading venues provide L3 data. Many venues offer a L3 feed as a separate, premium service over a standard L1 feed. For example, Nasdaq offers L1 and L3 data through their Nasdaq Basic and TotalView feeds respectively; Cboe offers L1 and L3 data through their Top and Depth feeds respectively, and NYSE offers L1 and L3 data through their BBO/Trades and Integrated feeds respectively.

On some trading venues, the L1 and L3 feeds are complementary services and the L3 feed does not necessarily provide a superset of information on the L1 feed. Notably, the Nasdaq Basic feed with NLS Plus provides additional information like trades on the Nasdaq TRFs, BX and PSX, as well as delayed consoliated trading volume—all of which are not found on Nasdaq TotalView.

For US stocks and ETFs, users must source L3 data directly from proprietary exchange feeds as the CTA and UTP SIPs only provide L1 data. Likewise for US equity options, users must source L3 directly from proprietary exchange feeds as OPRA SIP only provides L1 data.

L3 data is more often accessed through direct feeds than vendor normalized feeds. Much fewer vendors provide L3 data than L1 data due to the additional complexity and bandwidth requirement; Databento is one of few vendors that offers normalized L3 data.

Yes, Databento provides L3 data through our Market by order (MBO) schema.