Low Latency

Low latency, in electronic trading, describes systems and networks built to minimise the delay between an event and the response to it, for example between a price change on an exchange and the arrival of a trader's order reacting to that change. Latency is the delay itself, and low latency is the engineering goal of making it as small and as consistent as possible. The relevant units are milliseconds (thousandths of a second), microseconds (millionths) and nanoseconds (billionths).

Speed matters most to firms whose strategies depend on reacting first: high-frequency trading firms, electronic market makers and arbitrageurs. For them an order that arrives second often earns nothing. Brokers and exchanges also invest in low latency to serve these clients and to process growing message volumes.

• Distance. Signals cannot travel faster than light, so the physical length of the path sets a floor. Network latency between cities is dominated by this.
• Transmission medium. Light in optical fiber travels at roughly two-thirds of its speed in a vacuum. Radio and laser links through air are close to the full speed.
• Network equipment. Each switch, router and firewall adds delay, as does converting or repeating a signal.
• Exchange systems. The exchange's gateways and matching engine take time to accept, check and process an order.
• The trader's own systems. Decoding market data, running the strategy's logic, checking risk limits and building the order all take time. The interval from receiving a data packet to sending an order is called tick-to-trade latency.

The time a signal needs to cover a distance is:

• $t$ is the one-way delay.
• $d$ is the length of the path.
• $v$ is the speed of the signal in the medium.

The straight-line distance between the futures data centre near Chicago and the equity data centres in New Jersey is roughly 1,200 kilometres. In fiber, at about 200,000 kilometres per second, the one-way delay over that distance is $1{,}200 / 200{,}000 = 0.006$ seconds, or 6 milliseconds, and real cable routes are longer than the straight line. Through air, at about 299,700 kilometres per second, it is $1{,}200 / 299{,}700 = 0.004$ seconds, or 4 milliseconds. The 2 milliseconds saved explain why firms built a microwave network along the route.

Firms attack each source of delay in turn. Colocation places the trading firm's servers in the same data centre as the exchange, connected by a cross connect of equal length for every client, which reduces distance to metres. Proximity hosting in a nearby building is a cheaper alternative. Between data centres, firms lease the shortest fiber routes or use microwave and laser network links.

Inside the machine, software is written to avoid anything unpredictable. Common measures include bypassing the operating system's network stack, dedicating processor cores to single tasks, keeping data in memory and avoiding pauses for memory management. The fastest firms move logic from software into hardware, using a field-programmable gate array to parse data and send orders in well under a microsecond. Access is taken directly from the exchange's own data feeds and through direct market access, not through slower consolidated feeds or intermediary systems. Consistency matters as much as the average: variation in delay, called jitter, causes a firm to lose races unpredictably.

For a market maker, stale quotes are a cost. When news arrives, quotes that are not updated in time are picked off by faster traders, so speed is a defence as well as a weapon. For arbitrage between related instruments, such as an index future and the shares in the index, the first firm to act captures the price difference and later arrivals find it gone. The competition is relative: being fast matters only in comparison with rivals, which has led to repeated rounds of investment for ever smaller gains.

Many participants do not need extreme speed. A pension fund working a large order over a day, or a strategy holding positions for weeks, is affected far more by market impact and spread than by microseconds.

The pursuit of speed has drawn criticism as an arms race that consumes resources without improving prices for end investors. Exchanges and regulators have responded in several ways. Colocation services must generally be offered on fair and non-discriminatory terms. In the European Union, MiFID II requires trading venues and their members to synchronise their clocks to a common standard so that the order of events can be reconstructed. Some venues have introduced deliberate delays, known as speed bumps. The US exchange IEX applies a delay of 350 microseconds to incoming orders. Proposals to replace continuous trading with a frequent batch auction aim to remove the reward for marginal speed altogether.