High-Frequency Trading: Latency, Co-location, and Strategies

Module: Fintech AI | Difficulty: Advanced

HFT Strategies

Market Making PnL

Optimal Quote Placement

Latency Requirements

| Tier | Latency | Technology | |------|---------|------------| | Ultra-low | <1μs | FPGA, custom ASIC | | Low | 1-10μs | Co-location | | Medium | 10-100μs | Cross-connect |

import numpy as np

class MarketMaker:
    def __init__(self, inventory_limit=100, spread=0.01):
        self.inventory = 0; self.inv_limit = inventory_limit
        self.spread = spread
    def quote(self, mid_price, inventory_skew=0.1):
        skew = inventory_skew * self.inventory
        bid = mid_price - self.spread/2 - skew
        ask = mid_price + self.spread/2 - skew
        return {'bid': bid, 'ask': ask, 'bid_size': 100, 'ask_size': 100}
    def execute(self, side, price, size):
        if side == 'buy':
            self.inventory += size
        else:
            self.inventory -= size
    def pnl(self, current_price):
        return self.inventory * current_price

Research Insight: HFT profitability depends on latency advantage. When multiple HFTs compete, spreads tighten and profits diminish. The arms race for lower latency has diminishing returns — moving from 100μs to 10μs provides more advantage than 10μs to 1μs.