Jump to content
March 30th 2009
Vol 24 No 26
OPEN PLATFORM

Keeping Up With the Dow Joneses

As machine-readable news algorithms gain more attention, firms and vendors alike must pay close attention to monitoring technologies to avoid the dual pitfalls of automated trading-latency and message loss. By Steve Wong, vice president of marketing at ClearSight Networks

In the securities markets, trading firms are constantly competing to obtain the most valuable information and to apply it in a way that optimizes the prices at which securities are bought and sold. Under these ideal conditions, firms can realize enormous trading profits.

Orders are often placed electronically, based on decisions made using a variety of data inputs such as news feeds, bid and ask prices, as well as trading volume history. In recent years, data inputs such as news feeds have become quite sophisticated. Much of the financial news available is distributed in a compact, machine-readable form (such as XML files) that can be quickly evaluated by software rather than a human reader-often in a millisecond or less. Examples are Thomson Reuters' NewsScope and Dow Jones' Elementized News Feed.

If a trading firm sets out to optimize the entire trading process, it becomes important to manage the input data faster and more appropriately than other firms. However, certain questions arise at the outset:

• How can I get the news information faster?

• How can I make decisions faster?

• How can I make better decisions?

• How can I get my orders to the market faster?

• How can I get my order confirmations faster?

If an organization can figure out the answers to these questions more quickly than its competitors, then it might be able to gain as much as a millisecond advantage in the time taken to process the input data and place a trade. This could mean the difference between placing a successful order and reaping the trading profits, or being pre-empted by a competitor.

According to one estimate published in Information Week, effective management of data can make a difference of $100 million per year for a major brokerage firm-or roughly $400,000 per trading day. Conversely, if one of the firm's competitors has an implementation that is one millisecond faster, that competitor could make an extra $100 million a year-most likely at the expense of the first broker.

A first step toward optimizing an automated trading system is to create a way of measuring possible obstructions that can interfere with the overall process of evaluating input data and placing orders. Two main culprits are latency and message loss.

The most effective way to keep track of these is to deploy a latency and message loss management system. This type of system typically works by deploying monitoring and recording probes at various points along the data path. The capabilities of this type of system vary from vendor to vendor. However, there are some key features to keep an eye out for, including:

• Solutions that can capture and store all network activity over relatively long periods of time, from days to months;

• Probes that can be accessed from a central server through an out-of-band network (critically important) so that data comparisons can be performed among the various recording probes located throughout the network;

• Probes that have an accurate time-sensing mechanism so that discrepancies between clocks on the probes are greatly reduced, if not eliminated.

Another way to track latency and message loss is for the system to notify users when an unacceptable amount of latency or packet loss occurs. The latency and message loss management system should allow users to set their own thresholds, and have a versatile alarm system, a way of logging all instances of unacceptable latency or packet loss, as well as having the capability to launch and execute scripts or programs. For example, if it becomes clear that data is arriving several milliseconds later than normal, the script could interrupt the trading program, to either stop placing orders, or to put more weight on transaction data-quotes and volume-rather than relying solely on the news feed, which may have reached a competitors' automatic trading system first. This would help to avert a potentially dangerous scenario in which a trading engine is making trading decisions based on old and most likely out-of-date information.

Finally, major brokerage firms need to deploy a system that can watch for unusual latency or packet loss events that occur only intermittently. Sometimes waiting for an alarm to be triggered and then attempting to capture data to see what is going on just doesn't work, since the ability to capture all of the packets of information may be the only way to detect these intermittent events. Since problems with latency and packet loss occur more often during periods of high traffic load, firms must be sure that their latency management system can keep up with the full network line speed without dropping any packets itself.

In summary, automated trading can produce substantial profits, but depends critically on the ability of an organization to assess the status of the dual enemies-latency and packet loss. For news feed providers, investing in equipment that can really tell them how quickly and reliably their news feeds are getting to subscribers is critically important. It is perhaps one clear way they can demonstrate value over competing news services.

Meanwhile, automated trading firms should invest in equipment that can tell exactly how old a piece of data is, and promptly communicate with their trading software to make the appropriate adjustments that can mean the difference between making a trading profit or not. A right decision based on old data becomes not only the wrong decision, but also a potentially disastrous one.