By John Knuff (Part 6 in an 8-part series)
Right time processing of alpha and risk
Managing space and time are crucial for successful trading strategies.
In M2M 2.0 this means optimizing when and where we process market data and trading decisions, but also how we allocate infrastructure capacity.
These processes are linked since latency is a function of geography, connectivity and capacity. Nowhere on the planet is potentially much more than a blink-of-the-eye away (0.15 seconds), but many connections are rather slower.
Moreover, what really counts is relative time: the difference between my cycle time to market versus yours.
In practical terms round-trip times for low and high frequency traders might differ by 50:1 or more on short-haul journeys, but sudden bottlenecks or technical glitches can quickly distort this.
Thus latency anomalies are key danger signals.
An automatic network failover to a backup circuit could delay traffic by many milliseconds even after “recovery.”
Yet if the algorithm cannot detect such shifts, fill rates will decline.
Alternatively if a trading engine suddenly generates too many orders per second, the burst itself could saturate an exchange gateway and trigger delays.
The only way to know the reliablity of market data is to keep a tight control on unexpected latency deviations.
High performance algorithmic trading really demands a continuous calibration and tuning process because market infrastructure and trading conditions are constantly evolving even intraday.
One global bank aims to manage time to within 20 microseconds. Such confidence can be invaluable for order routing or statarb decisions. Some algorithms try to track everything including latencies, inter-tick time delays, and event spreads between markets.
Since traders rely on financial networks or their brokers for market access, the right partner could well depend on a sufficiently accurate timekeeping system.
Pay-as-you-grow finance using cloud and other scalable SaaS technologies is another new Infrastructure rule for M2M 2.0.
Inadequate resources immediately impact on performance, yet in a distributed architecture prediction of demand peaks by node is doubly difficult, as liquidity shifts intraday and volume spikes depend on sentiment.
This is exacerbated by the shortening of investment life cycles, as we approach zero latency.
Thus, there is little incentive to over-engineer a solution, since the infrastructure may itself become obsolete long before it is needed to meet real demand.
Yet predictable response times are crucial just when markets are congested and volatile.
Wherever practical, traders are turning fixed asset purchase decisions into services streams with pay-as-you-grow finance.
This is just another example of being in the right place at the right time.
Coming up in part 7: Integrating the Back Office Into the Front Office
“The faster we get, the the more focus we put on safety and risk management.”