What Most Smart Order Routers Get Wrong About Microstructure

Order routing decisions are only as good as the assumptions they’re based on. Most Smart Order Routers (SORs) are built around surface-level heuristics: best price, latency proximity, fee structures. But trading microstructure doesn’t behave like a static model. It reacts, adapts, and sometimes deceives. Misunderstanding these dynamics leads to execution that looks efficient in logs—but leaves edge on the table.

Fill Probability Is Not Just a Price Game

Most SORs prioritize venues based on top-of-book prices. If two exchanges show the same ask, the router sends the order to the venue with lower fees or lower ping. But quote equality doesn’t imply equal fill probability. On one venue, that top-level ask might be a fresh order with real intent. On another, it could be an iceberg on its last refill cycle, about to vanish. Microstructure tells you which is which—but most routers don’t listen.

Execution probability is shaped by queue position, refresh behavior, and cancellation speed. A quote that looks stable might be bait. A thinner book might fill faster if it’s driven by real market makers rather than latency arbitrage bots.

Ignoring Queue Depth and Volatility

Static best-price logic fails under volatility. During fast markets, spreads oscillate and books rearrange in milliseconds. SORs that don’t account for depth—or fail to model time-to-fill—will route to a venue that looks optimal for one snapshot but underperforms in execution. A quote 0.01 better doesn’t help if it vanishes before you land.

Depth modeling isn’t just about counting visible levels. It’s about tracking refill behavior, passive/aggressive flips, and cancellation patterns. These signals separate durable liquidity from surface noise.

Cancel-Replace Loops and Ghost Liquidity

A major failure point: handling of cancel/replace loops. Some SORs treat the book as if it were passive and stationary. But on certain exchanges, quotes rotate faster than the market moves. Traders continuously cancel and re-post to stay at the front of the queue. The book shows volume, but most of it’s reactive.

A router that sends child orders into this environment without anticipating that churn gets stuck. Orders rest behind fast-moving liquidity, never executing, while price moves away. This leads to hidden slippage—fill prices that trail the theoretical opportunity.

Trade Classification Blind Spots

Many routers don’t classify trades in real time. They rely on market data feeds that simply show prints—volume at a price. Without trade classification, they miss whether a trade was passive, aggressive, or midpoint. This skews their fill expectation models and feedback loops.

If a venue fills mostly on aggressive takers, a passive quote there has lower odds of execution. Without knowing this, the router may overweight that venue and underperform over time. Execution modeling demands real-time classification logic—something too few systems have implemented.

Symbology and Normalization Risks

Another common error: treating all instruments as structurally equivalent. Even when routing to the same underlying asset (say, BTC/USDT), slight variations in lot size, tick size, fee tier granularity, and minimum notional size can affect route efficiency. These details are often abstracted away under normalized routing engines.

But microstructure lives in these details. A router that doesn't account for minimum quote increments or hidden quote penalties introduces artificial constraints. What looks like a 1:1 mapping across venues is structurally biased.

Poor Adaptation to Real-Time Market State

Markets mutate during the day. Latency paths fluctuate, internalizers become aggressive, fee tier thresholds shift. Yet many SORs operate with fixed routing preferences that ignore these intraday signals. The system might route to Venue A all morning, even after Venue B overtakes it in fill speed and quote stability.

Routers must adapt not just at the trade level but at the regime level. Detecting microstructure regime shifts—like the rise of liquidity internalizers in US equities after 10:00 AM—is core to accurate routing. Most systems aren’t built for this kind of sensitivity.

The Danger of Over-Aggregation

Routing engines often aggregate quotes across venues to simulate a unified book. This is useful in theory, but dangerous in practice. Aggregation blurs venue-specific traits. It hides the fact that a quote at the same level on Exchange A is more resilient than one on Exchange B. Execution strategy should account for venue-specific latency, cancel rate, and queue mechanics. Aggregation flattens all that nuance.

Traders using execution logs to benchmark performance often miss these differences too. The router gets credit for sending to "best quote", even if the best quote wasn’t real.