Multi-Venue Simulation and Routing

A stock like a large-cap US equity doesn't trade on one exchange — it trades simultaneously across a dozen or more lit exchanges and dark pools, each with its own book, fees, and latency. A simulator that models a single "consolidated" book — pretending all that liquidity sits in one place with one queue — skips the entire routing problem a real execution system has to solve: which venue to send to first, how to split an order across venues, and what it costs in fees and information leakage.

Why a single-venue model is misleading

Collapsing all venues into one book creates two distortions. First, it overstates available liquidity at the best price, since it silently sums size that in reality sits in separate queues at separate venues with separate access rules — an order can't "take" size on Exchange B by resting at Exchange A. Second, it ignores that different venues charge different fees (maker rebates versus taker fees vary by venue and can differ by several tenths of a cent per share) and have different latencies to reach, both real costs a routing strategy has to weigh. A single-venue backtest can make a strategy that chases the cheapest venue look identical to one that blindly sends everything to the first venue found, even though their live economics diverge substantially.

What a multi-venue simulator needs to track

At minimum: a separate order book per venue, per-venue fee schedules, per-venue latency to receipt and fill confirmation, and a routing policy — the logic deciding, for a given order, how to split it across venues and in what sequence. That routing policy is itself worth backtesting, since real smart order routers differ in whether they prioritize speed, price, or fee minimization, and the "right" priority depends on the strategy's urgency.

Worked example: routing a 2,000-share order across three venues

An order for 2,000 shares needs to buy at or below $50.05. The consolidated top-of-book shows 2,000 shares available at $50.05, but it's actually split: Venue A has 500 shares at $50.05 (taker fee $0.0030/share), Venue B has 800 shares at $50.05 (taker fee $0.0015/share), Venue C has 700 shares at $50.05 but with 8ms more latency than A or B.

A naive single-venue simulator fills all 2,000 shares instantly at $50.05 with one fee assumption. A multi-venue simulator routes in fee-and-latency order: A and B fill immediately (1,300 shares, blended fee $0.0021/share), and the order to C arrives 8ms later — by which point 200 of C's 700 shares have already been taken by other participants, leaving only 500 available, so the order fills 1,800 of 2,000 shares in this pass and the remaining 200 becomes a residual to route again. That gap between "2,000 shares instantly" and "1,800 shares this pass, with real fee variation" is exactly what routing simulation is meant to expose.

What this means in practice

Multi-venue simulation matters most for liquid, widely-listed names and for any strategy sized large enough that a single venue's displayed liquidity can't fill it alone. For small, single-listed instruments the distinction matters less. But for the common case — US equities, major FX pairs, listed futures with multiple matching venues — treating the market as one consolidated book overstates achievable size and understates real execution cost, in a way that compounds with strategy size.