Tick Data

Unlike aggregated OHLC bars, tick data preserves every atomic market event as a discrete record. Each entry represents a single trade or quote update, not a summary over a time window. This captures the exact sequence of price formation.

• Trade Tick: Records price, volume, and exchange timestamp of a completed transaction.
• Quote Tick: Records the best bid price, best ask price, and associated sizes at a specific moment.
• Microsecond Precision: Modern exchanges timestamp events with nanosecond granularity, enabling precise sequencing.

Tick data arrives asynchronously and at variable frequencies. During high-volatility events, thousands of ticks may fire per second; during quiet periods, seconds may pass without an update. This non-uniform sampling breaks standard time-series assumptions.

• Duration Modeling: The time between ticks carries information—long pauses often precede breakouts.
• Resampling Required: Most models require transforming irregular ticks into regular bars (time, volume, or dollar bars) before ingestion.
• Point-in-Time Integrity: The exact arrival sequence is critical for avoiding look-ahead bias in backtesting.

At the tick level, prices do not follow a smooth random walk. They oscillate rapidly between the bid and ask quotes, creating a sawtooth pattern known as the bid-ask bounce. This is not true price discovery but mechanical noise.

• Trade Direction: Trades at the ask are buyer-initiated; trades at the bid are seller-initiated.
• Roll Model: A classic microstructure model that estimates the effective spread from the serial covariance of price changes.
• Noise Filtering: High-frequency strategies must explicitly model this bounce to avoid trading on illusory reversals.

A single liquid instrument like SPY can generate terabytes of tick data annually. Capturing the full US equity tape across all exchanges requires petabyte-scale storage infrastructure and specialized compression techniques.

• Compression: Algorithms like delta encoding and run-length encoding reduce storage footprints by 90%+.
• Columnar Formats: Parquet and ORC files are standard for analytical queries on historical tick archives.
• Ticker Plants: Specialized hardware/software systems designed to consume, normalize, and distribute tick data in real-time.

In modern fragmented markets, a single security trades simultaneously across dozens of exchanges, dark pools, and alternative trading systems. A consolidated tick stream must merge and sequence these disparate feeds.

• National Best Bid and Offer (NBBO): The regulatory benchmark constructed by combining quotes from all lit exchanges.
• SIP vs. Direct Feeds: The Securities Information Processor feed is slower than proprietary direct exchange feeds, creating latency arbitrage opportunities.
• Timestamp Reconciliation: Each venue has its own clock; precise sequencing requires GPS-synchronized timestamps and clock drift correction.

Raw tick data includes metadata flags that qualify the nature of a trade or quote. Ignoring these codes can lead to catastrophic backtesting errors, such as including off-exchange block trades as actionable signals.

• Trade Conditions: Codes indicating late reports, odd-lot trades, or exchange acquisitions that should be filtered.
• Quote Conditions: Flags for locked/crossed markets, fast trading halts, or non-firm quotes.
• Correction Ticks: Exchanges issue cancellation and correction records for erroneous trades; these must be applied to maintain an accurate tape.