The time it takes to send a message to an exchange and receive an acknowledgement has become increasingly important. Co-location of servers at an exchange’s site is one way of reducing the latency.

The proliferation of algorithmic trading models is having a profound affect on the securities industry. Traders use algorithms to optimize execution by electronically breaking up orders, selecting which venue to send them to and launching the orders at just the right time. This practice was first popularized in the US, but before long it spread to Europe, and now it is also appearing in parts of Asia.

In this environment, roundtrip latency – the time it takes to send a message to the exchange and receive an acknowledgement – has become increasingly important. Latency is caused by a range of factors, including hardware and software performance, network hops, bandwidth and distance. The faster the response time, the sooner traders can react to this information by initiating new orders or changing existing ones.

Given the potential for competitive advantage, algorithmic traders measure and monitor these numbers and use them to pressure exchanges and electronic communications networks (ECNs) to introduce latency-optimized solutions to improve performance. And nowadays, brokers in the US and Europe publish latency tables to gain marketing advantage.

This focus and arm-twisting has driven significant improvements in roundtrip latency in the past few years. “Seven years ago, the brokerage community’s key benchmark was a per second response time with a subsequent move to hundreds of milliseconds,” says Hirander Misra, FVP - Head of Strategic Product Development & Director of Instinet Chi-X Ltd. “Now it is just a small number of milliseconds, with a move towards microsecond benchmarking.”

»For a US-based client trading in London, eradicating the “pond hop” can save at least 100 milliseconds round trip.«

Market participants have devised a few ways to reduce latency. Closer typically means faster, so some clients have co-located their servers in their broker’s or network provider’s data center, which houses the gateway connections to the exchanges. When the client and the exchange are both in the UK, this eradicates the four millisecond roundtrip latency of the local connection in London. For a US-based client trading in London, eradicating the “pond hop” can save at least 100 milli-seconds roundtrip.

But co-location does not solve the whole problem. Since market data providers add a latency delay of about 250 milliseconds, traders may be sending orders to the exchange faster than the trades are reflected in the market data. To solve this issue, some clients have started taking direct data feeds from the exchanges and interfacing them with their applications. And now information vendors provide ultra-low latency full-tick feeds, direct from exchanges. For example, Reuters offers a service benchmarked to an average latency of less than one millisecond.

Even with co-location, many exchange systems are still slower than ECNs because messages sent from the client’s server are validated in an API layer before going to the core matching system. Each additional step or process adds latency.

The faster the response time, the sooner traders can react to information.

The question is whether latency can be reduced further. Adopting the FIX Adapted for Streaming Market Data (FAST) Protocol can help exchanges handle large volumes of data and reduce latency. At the simplest level, FAST is FIX that has been adapted using compression algorithms. It reduces the size of the messages up to 80 percent. Smaller messages can travel faster, so they get to the counterparties quicker. They also require less bandwidth, which yields a cost savings. Still, commodity technology has its limitations, says Peter Lundgren, Product Line Manager at OMX, and pushing the envelope too far would compromise flexibility, robustness and stability. Latency normally can be reduced further at the expense of properties such as transaction throughput and resilience. But in his opinion, reaching significantly below millisecond latency also requires specialized technology solutions such as more exotic network equipment or even bypassing layers in the communications stack on the network cards.

Before going down this route, exchanges should define their niche and decide what type of operator they want to be. Then they need to combine technology and services that provide the best compromise. Like broker and network providers, exchanges can offer co-location services. Exchanges can also host part of clients’ trading algorithms in their data center, although traders may be reluctant to do this because they want to protect their intellectual property.

Exchanges need to evaluate their decision in light of increased competition in the post-Regulation NMS and MiFID environment. Those that do not take a proactive approach will be left behind.

By Sherree Decovny Illustration Måns Adolfsson

MarketView 2007:1