Smart Order Routing

A single stock does not trade in one place. The same US equity trades on a dozen-plus lit exchanges and dozens of dark pools, each with its own book, its own fees, and its own queue. That scattering is Market Fragmentation, and it creates a genuine problem: to buy 5,000 shares at the best price, you may have to hit several venues at once. A smart order router (SOR) is the software that solves this, it takes one "parent" order and decides, in real time, how to slice and route it across venues to get the best all-in execution.

What the router is optimizing

The naive goal is "get the best price," but the real objective is more complete. For each venue the router weighs:

• Displayed price and size, where is the liquidity, and how much of it?
• Fees and rebates, a fill at the same price nets differently on a maker-taker venue than on an inverted one (Maker-Taker Fees). The router optimizes price plus fee, not price alone.
• Fill probability, a quote you can see may vanish before your order arrives (quote fading), so a nominally better price on a slow-to-reach venue may not actually be gettable.
• Latency, the round-trip time to each venue, because the book you're routing against is already stale by the time you act.

Putting these together, the router minimizes total expected cost across a split of the order, choosing how many shares to send to each venue so that the marginal all-in cost is equalized. In one line, it seeks the allocation that solves

where $q_v$ is the shares routed to venue $v$, $\text{price}_v$ and $\text{fee}_v$ are that venue's price and fee per share, and $Q$ is the total order, all constrained by how much size each venue actually shows and how likely it is to fill.

The rule that forces routing: best execution

Routers are not just a convenience, they are partly a legal requirement. In the US, Regulation NMS and its Order Protection Rule say a venue cannot execute a trade at a price worse than the best displayed quote on another venue (a "trade-through"). So if a better price exists somewhere else, the order must be routed there or that quote must be cleared first. Combined with brokers' general best-execution duty to their clients, this means a compliant router must survey the whole market, not just its home exchange.

Worked example

You want to buy 1,000 shares. The router sees:

• Lit A: 400 shares offered at $30.00, taker fee $0.0030/share.
• Lit B: 600 shares offered at $30.01, taker fee $0.0030/share.
• Dark pool: no displayed quote, but the lit midpoint is $30.005, so a dark fill would cross there with no taker fee.

A price-only router would sweep 400 at $30.00 from A, then 600 at $30.01 from B, average $30.006, plus $0.0030 fee on all 1,000 → all-in $30.009. A smart router does better: it first pings the dark pool for a midpoint fill. Suppose 300 shares cross dark at $30.005 with no fee. It takes the remaining 700 from the lit books, 400 at $30.00 and 300 at $30.01, for a blended lit price of $30.0043 plus $0.0030 fee. The all-in average across all 1,000 shares now works out around $30.006, meaningfully cheaper than the price-only sweep, because the dark slice saved both the extra tick and the taker fee on 300 shares.

Where it goes wrong

• Quote fading. By the time your order reaches a venue, the quote you routed against may be gone, fast traders cancel on the signal of an incoming sweep. Routing against stale prices means missing the fill you counted on.
• Information leakage. Spraying child orders across many venues signals a large parent order. Predatory traders reconstruct your intention from the pattern and trade ahead. Good routers randomize timing and venue choice to blur the footprint.
• Rebate conflict. A router tuned to maximize the broker's rebate rather than the client's fill quality is the classic best-execution abuse (Maker-Taker Fees). The right benchmark is realized cost, not headline price.
• Latency arms race. Because the freshest book wins, routing decays into a speed contest, a core Latency & High-Frequency Trading dynamic.

Smart order routing is the tactical layer beneath the bigger execution question: how fast to trade a large parent order over time, which is the domain of Optimal Execution and schedules like TWAP, VWAP & POV. The router decides where each slice goes; the execution algorithm decides when.