Lecture, Tutorial
Making Trades in Microseconds and Nanoseconds
- Trading systems operate across a spectrum of time horizons rather than a single universal latency requirement.
- Ultra-low latency regimes require packet turnaround under 100 nanoseconds, rendering standard CPUs ineffective due to hardware constraints.
- Decisions in this regime must be executed on FPGAs directly wired to the network to achieve speeds where a packet departs before the CPU fully consumes it.
- Computational complexity is severely limited at this speed; decision logic is necessarily simple.
- Mid-range regimes operate on timescales ranging from microseconds to milliseconds, allowing for more sophisticated decision models.
- Long-duration regimes accommodate decision turnaround times of an hour or a full day for strategic processes.
- Ultra-low latency regimes require packet turnaround under 100 nanoseconds, rendering standard CPUs ineffective due to hardware constraints.
- There exists a fundamental trade-off curve balancing the intelligence of the decision (model complexity) against the speed of turnaround.
- Optimized strategies utilize an ensemble approach, deploying different decision-making processes tailored to specific latency buckets.
- Competitive positioning requires matching the complexity of the logic to the available time horizon for each specific trade type.