Limit Order Book (LOB)

The primary matching algorithm governing most electronic LOBs. Orders are first ranked by price (best price gets priority) and then by time (earliest order at that price gets priority).

• Bid side: Higher prices have priority
• Ask side: Lower prices have priority
• A buy order at $100.05 placed at 10:00:00.000 executes before a buy order at $100.05 placed at 10:00:00.001
• This deterministic rule ensures fairness and discourages latency arbitrage at the queue level

Market depth refers to the cumulative volume of limit orders resting at each price level beyond the best bid and offer. It visualizes the liquidity profile of the book.

• Level 1 data: Best bid, best ask, last traded price
• Level 2 data: Full order book depth, typically 5-10 price levels on each side
• Depth chart: A visual representation where the x-axis is price and the y-axis is cumulative volume
• A deep book with large resting orders at multiple levels signals high resiliency—the ability to absorb large market orders without significant price impact

A LOB manages the full lifecycle of diverse order types, each with distinct execution rules and cancellation policies.

• Limit Order: A commitment to buy/sell at a specified price or better. Rests in the book until filled or cancelled
• Market Order: An instruction to execute immediately at the best available price. Consumes liquidity; never rests in the book
• Iceberg Order: Displays only a small visible portion of the total quantity, replenishing automatically to conceal large trading intentions
• Fill-or-Kill (FOK): Must execute in full immediately or be cancelled entirely
• Immediate-or-Cancel (IOC): Executes any portion possible immediately and cancels the remainder

The LOB is a two-sided marketplace where makers provide liquidity and takers consume it, with fee structures incentivizing the former.

• Maker: Places a limit order that rests in the book, adding liquidity. Typically receives a rebate or pays a lower fee
• Taker: Submits a market or marketable limit order that matches against a resting order, removing liquidity. Pays a higher fee
• Maker-taker fee model: Common on exchanges like Coinbase and Binance, where makers pay 0.00%–0.02% and takers pay 0.04%–0.10%
• This economic incentive ensures tight spreads by rewarding those who quote competitively

A LOB is not a static snapshot but a high-frequency event stream. Each change is broadcast as a discrete event, critical for algorithmic trading systems.

• Add: A new limit order is placed at a price level
• Modify: An existing order's quantity is changed (partial fill or manual adjustment)
• Cancel: An order is removed from the book before execution
• Execution/Trade: A match occurs between a taker and one or more makers, reducing resting quantity
• Exchanges like NASDAQ emit these events via ITCH and OUCH protocols at microsecond granularity, requiring event-driven architectures to process

The shape and dynamics of the LOB encode predictive signals about short-term price movements. Quantitative researchers extract features from the book's state.

• Order Flow Imbalance (OFI): The net difference between aggressive buy and sell volume at the best bid/ask over a time window. High positive OFI predicts upward price pressure
• Bid-Ask Spread: A wide spread signals low liquidity and high adverse selection risk; a narrow spread signals a competitive, liquid market
• Queue position: The ordinal position of an order within a price level. Orders near the front of the queue have a higher fill probability
• Volume-synchronized probability of informed trading (VPIN): Uses LOB volume imbalances to estimate the presence of informed traders

The fundamental matching algorithm governing most electronic limit order books. Orders are first ranked by price—the highest bid and lowest ask have the highest priority. At the same price level, orders are ranked by time of arrival, rewarding the earliest order. This deterministic rule ensures fairness and incentivizes competitive pricing and early liquidity provision. Violations of price-time priority, such as those in some periodic auction models, create distinct market microstructure effects.

Two distinct data feed representations of LOB depth. Market by Order (MBO) provides full order-level granularity, showing every individual resting order, its unique ID, and its exact size. This is essential for order flow analysis. Market by Price (MBP) aggregates all orders at each price level into a single total volume. MBP is more compact and common for visualization, but MBO reveals the true queue dynamics and iceberg order detection.

The LOB accepts a variety of instruction types beyond simple limit orders. Key types include:

• Market Order: Executes immediately against the best available opposite-side orders, consuming liquidity.
• Limit Order: A passive order to buy or sell at a specified price or better, providing liquidity.
• Iceberg Order: Displays only a small visible portion of the total quantity, concealing the full order size.
• Fill-or-Kill (FOK): Must be filled entirely and immediately or the entire order is cancelled.

A real-time metric quantifying the asymmetry between supply and demand at the top of the book. Calculated as (Bid Volume - Ask Volume) / (Bid Volume + Ask Volume) for the best N levels. A high positive imbalance indicates strong buying pressure and is a powerful predictor of short-term upward price movement. High-frequency market-making algorithms use imbalance signals to dynamically adjust quotes and manage inventory risk.

The probability of a resting limit order being filled depends critically on its queue position at a given price level. Orders ahead in the queue are filled first. Traders use sophisticated models to estimate their queue position based on order book snapshots, trade prints, and cancellations. The value of a limit order decays as it loses queue priority, a concept central to optimal execution and market-making strategies.

Raw LOB data requires extensive preprocessing for machine learning. Steps include:

• Tick-level alignment: Synchronizing updates from multiple venues.
• Outlier filtering: Removing erroneous quotes far from the mid-price.
• Level standardization: Normalizing to a fixed number of price levels (e.g., 10 levels) for model input.
• Temporal resampling: Converting irregular event-time updates to fixed-interval snapshots for architectures like Temporal Convolutional Networks (TCNs).