For each asset class and line of business two latency thresholds are identified:
- A low threshold, which must be reached to be considered a leading performer
- A high threshold, above which firms are exposed and clearly trailing the market
The three zones of latency posture are delineated by the two thresholds:
- Below the low threshold is the “opportunity zone”.
- In between the two thresholds lies the “safe zone”.
- Above the high threshold is the “danger zone”.
These thresholds may vary market by market and evolve across time. To keep up with the pace, a firm must divide the latency in its trading systems by two every three years. This may seem a gentle rate of progress but it actually means that the end-to-end processing time of electronic trades has been reduced by 94% over the past decade just to stay competitive; and all this with volumes that are increasing and trading decisions that are ever more sophisticated. Reducing latency is a continuous and iterative endeavour in which there are no shortcuts. The pursuit of lower latency is a cyclical process always beginning with measurement, followed by monitoring and improvements, finally returning to measurement to start another cycle.
Accounting for latency in the technology layers without the corresponding business context is of limited value. Any latency measurement and reduction strategy must cover both the business and technology levels. From a business perspective, the data collection should focus on trading outcomes, combined with volumes and order flow rates. Technical measurements should, in turn, concentrate on both end-to-end as well as step-by-step latencies.
Many approaches are available to reduce latency within a trading platform, each technique targeting a specific layer or component within the chain. Since each strategy carries its own challenges, compromises and price tag, it is paramount to have a prioritised and reasoned selection process to pick the most appropriate solution to a qualified business requirement. The application layer remains the place where most of the latency is generated. It is therefore the initial target for improving latency.
Infrastructure can give a competitive advantage; however ‘buy’ is now clearly winning over ‘build’. With current offerings, where the wire is shared among those who bought the service, there is an assurance that performance is similar across parties. Designed and built for purpose, infrastructure is advantageous but comes with high costs: design, deployment and maintenance. Ultra-low latency is achieved if the propagation delay is reduced to the minimum. Proximity hosting and co-location are on the rise as propagation can add double-digit milliseconds latency. Proximity hosting and colocation offer good trade-offs, explaining their growing popularity despite their relative high costs.
