Latency, tick-to-trade

The term tick-to-trade latency is often used interchangeably with wire-to-wire latency. However, as the term tick implies, it's more precise to interpret tick-to-trade as a more specific form of wire-to-wire latency where the inbound event is a market data event and the outbound event is an order event.

Another term that is often used interchangeably is tick-to-order latency.

Tick-to-trade latency is usually measured as the time from last bit of inbound data message to last bit of outbound order message. When the venue or protocols are not specified, it's usually assumed that tick-to-trade latency was measured using a typical binary protocol like SBE or ITCH for the market data and SBE, FIX, or OUCH for the order entry.

Often, tick-to-trade latency is estimated from a simulated environment where market data is replayed from packet captures and fed to the NIC of a trading system. Otherwise, it can be obtained from actual order logs.

Tick-to-trade latency is usually measured without any strategy or model logic. Hence, it can be misleading since some low latency trading strategies compute a very large number of features and alpha signals in the critical path. On the other hand, techniques like speculative triggering and precomputation of signals take these off the critical path and may mean that a strategy with complex signals or order placement logic will run close to tick-to-trade latency of the bare trading platform.

Several off-the-shelf hardware solutions are commonly used to measure tick-to-trade latency:

• Commercial capture appliances like Corvil (now part of Pico) offer built-in functionality for measuring tick-to-trade latency.
• Other capture appliances like FMADIO and Endace do not inclujde tick-to-trade analytics out of the b ox, but can still be used to timestamp inbound market data and outbound order messages for latency measurement.
• Network cards with hardware timestamping, like Napatech, UberNIC and ExaNIC, can provide precise timestamps for latency measurement.
• Network switches with timestamping features like the Arista 7130 and 7135 series.

In production trading environments, passive or regenerative network taps are typically deployed to replicate inbound market data and outbound order traffic for latency measurement. These taps allow the monitoring system to pull packets off the wire and measure tick-to-trade latency precisely without interfering significantly with the trading system's critical path.

On software-based trading systems, the tick-to-trade latency is usually dominated by the round-trip hop over PCIe bus between NIC and CPU. On systems with PCIe 3.1 and a NIC capable of TCP checksum offload, like a Xilinx Solarflare X2522 NIC, along with a kernel bypass networking stack like OpenOnload, the lowest achievable tick-to-trade latency is usually just under 2 microseconds.

On hardware-based trading systems using FPGAs or ASICs, sub-microsecond latencies are very typical. As of 2024, most trading firms that compete on this front are capable of tick-to-trade latencies in the single to double-digit nanoseconds.