Routing, Payment for Order Flow and Best Execution

Where your order went, and who got paid

A marketable retail order is usually filled by a **wholesaler** — a market maker that buys retail flow, agrees to fill it at or better than the NBBO, and pays your broker for the privilege. That payment is **payment for order flow (PFOF)**, and it is legal, disclosed, and enormous: a large share of U.S. equity volume is executed this way. The wholesaler can afford the payment because it sees the flow and can manage the risk — it takes the other side, earns the spread on average, and hedges while it holds inventory. Your broker, meanwhile, has a duty to seek the best available terms for you and a revenue stream that depends on where the order goes. That is a **conflict of interest** in the plainest sense, and it deserves to be named rather than softened. What follows is the part arguments skip. Because wholesalers compete for that flow, the competition shows up in the fills: price improvement, tight effective spreads and high fill rates. A broker that takes payment can deliver a better outcome for a small marketable order than a broker that routes everything to exchanges and pays the spread plus a taker fee. That is not an argument for PFOF — it is an argument for reading the numbers, which is what the disclosures are for. Not every jurisdiction agrees: the UK and the EU have moved to restrict or ban payment for order flow, on the view that the conflict is not worth managing by disclosure.

Best execution as a duty, and the paperwork that measures it

**Best execution** is not "the best price" — it is the best overall terms reasonably available, considering price, speed, likelihood of execution, order size and settlement. A broker's routing decisions are supposed to serve that standard, and it is judged against the market at the time: a wholesaler filling you a cent inside the protected quote has met it; a broker crossing two spreads to reach an exchange has not. Two disclosures let a customer check. **Rule 606** reports where orders were routed and what was paid for them — the map of the conflict. **Rule 605** reports market quality: effective spread, price improvement, fill rates, and how often trades were at or better than the NBBO. In 2024 the SEC expanded 605 reporting to broker-dealers with 100,000 or more customer accounts, so far more of the market now has to publish the numbers. The identity that makes the numbers readable is simple. **Effective spread = quoted spread − 2 × price improvement**, expressed per share and then in basis points of the price. A 2¢ quoted spread with 0.4¢ of improvement is a 1.2¢ effective spread; on a $180 stock that is 0.67 bps. Compare that between two brokers and you have priced your own execution — no ideology required. Pricing two brokers on a $180 stock, 100,000 shares a year — Broker A — wholesale routing: 2.0¢ quoted − 0.8¢ = 1.2¢ effective → 0.67 bps · Broker B — exchange routing: 2.0¢ quoted − 0.2¢ = 1.8¢ effective → 1.00 bps · Difference: 0.33 bps ≈ $600 of spread + $300 of taker fees ← A taker fee is the exchange charging you for removing liquidity, and it is not part of the effective spread — it is an extra line. When you compare brokers, compare both: spread first, fees second, and remember the spread is usually the bigger of the two.

The two-tier market the disclosures describe

Read 606 and 605 together and a structure appears that no chart shows. A retail marketable order generally does not compete on an exchange at all. It goes to a wholesaler that has agreed to fill it at or better than the public quote, and it is filled there. The exchange prints in a public data feed are, increasingly, the residue of institutional flow — a different population of buyers and sellers from the ones in your order. Three consequences follow, and all three are checkable. First, the public quote is a thinner sample of true interest than its name suggests: a wholesaler sitting on a large internal flow can fill you at a price no exchange participant ever saw. Second, the competition for that flow is genuine and shows up in your fill, which is why “no commission plus a cent of improvement” can beat “no commission plus the whole spread”. Third, the flow is concentrated in a handful of firms, and concentration is a form of fragility. January 2021 is the case for the third point. The broker restrictions on meme-stock buying were driven by clearinghouse collateral rather than by the wholesalers — but the episode showed how short and load-bearing the chain between your tap and your fill really is, and how quickly an arrangement made quietly in calm markets becomes the thing everyone is arguing about (M21). None of this argues that internalisation is bad. It argues that the market you trade in and the market you can see are two different places — which is precisely why the disclosures exist, and why reading them is a skill rather than a chore.

Reading a 605 report without being fooled

A 605 report is a table of market-quality statistics, and like any table it can be read correctly or flatteringly. Four definitions decide whether two reports are comparable at all, and each has a way of being presented in the best possible light. **Price improvement is measured against the quote in force at the time of the order**, not against the NBBO you can see now, and the reference matters: a wholesaler comparing itself to its own internal quote will always look excellent. **The share-weighted versus order-weighted distinction** changes the story too — a report showing strong average improvement may have achieved it on a few large orders while leaving most small ones at the quote. **Effective spread includes only executed trades**, so a venue that declines to fill the hard orders shows a narrower spread than one that fills everything. And **fill rates are reported for the orders the venue chooses to accept**, so a high fill rate can simply describe a filter. The practical approach is to compare like with like and to read the report against your own order flow. If you trade 100 shares at a time, the relevant row is the one for small marketable orders, not the headline blended average. If most of your orders rest rather than cross, the spread statistics describe someone else’s trading entirely. The report is evidence about a population, and the useful question is always whether you are in it. • Price improvement is measured against the quote in force at the time, not the NBBO today • Share-weighted averages can hide weak per-order performance • Effective spread only counts trades that happened — declines are invisible • Fill rates describe accepted orders, so a filter improves them • Find the row that matches your order size and your order type None of this makes a 605 report unreliable. It makes it a document with a population, a definition and a denominator — and a number without those three is not comparable to anything.

How best execution is actually policed

Best execution is a duty of care, and duties of care are hard to measure, so the regime substitutes a process requirement. A broker must review the execution quality it obtains — routinely, on the venues and order types it actually uses — and use what it learns to route better. That is why a 606 report exists at all: not to rank brokers publicly, but to force each one to look at its own numbers. What enforcement looks like in practice is instructive. The failures regulators bring are almost never “this fill was a cent worse than the NBBO”; they are failures of the *process* — quoting a single venue for years without review, claiming price improvement that was really the midpoint of a wide spread, or letting a payment arrangement stand in for a quality analysis. The lesson generalises: an execution claim is only as good as the comparison behind it, and a number without a benchmark is marketing. Read a broker’s disclosure the way a regulator would. Does it define its own terms? Does it compare like with like — marketable limit orders against marketable limit orders, rather than against every order it received? Does it report the share of volume receiving price improvement, and the size of that improvement in dollars per hundred shares? A disclosure that answers those questions is doing its job even when the fills are ordinary. Small differences compound, but only when they are measured at the same size and order type. A “better” statistic computed on a different order mix is not evidence of better execution.

How the routing decision is actually made

Routing is not a lookup table. A **smart order router** runs an optimisation for every order or slice: it takes the order’s side, size, limit and urgency, reads the current state of every venue’s book in that symbol, and chooses a destination or a sequence of them. What it maximises is expected filled price — which is not the same as the best quoted price, because a quote is worth only what your probability of filling against it is. That is why a router learns. A venue displaying a cent better may be a venue you have never filled on, while a venue at the touch fills reliably; the router tracks fill rates per venue per symbol and shifts its preference accordingly. The result is that two brokers given an identical order can send it to different places, and both can describe their choice as protecting your price. The constraints on the decision are where the conflicts live. The order protection rule forbids trading through a better protected quote, so the router cannot simply ignore a superior price. Best execution, meanwhile, is a duty that attaches to the *broker*, not to the software — which means the router’s objective function is set by the broker’s own policy. A broker that earns a rebate for adding liquidity at one venue, or receives payment for routing retail flow to a wholesaler, has a dollar reason that is not your price. Both arrangements are disclosed, and neither is secret; the point is that “the router decided” is not an explanation. There are two broad destinations and they suit different orders. Exchange routing puts the order into a lit book where it rests, takes, or sweeps several venues at once. **Internalisation** at a wholesaler has the order filled from that firm’s own inventory or its hedging, at or better than the best quote — which for a small marketable retail order usually produces a cent or two of improvement, and for a large order usually cannot, because the size has to be carried by lit venues that display it. For a retail order inside the displayed size, routing is a small lever: the reference is the best quote, the improvement is a cent or two, and the venue mix is mostly a footnote. It becomes a large lever in the tail — a big order, a thin symbol, a fast tape — which is precisely when the disclosures are least able to describe what happened. That asymmetry is the reason the reports reward reading the mix rather than the headline. • A router maximises expected filled price, weighting a quote by the odds of actually getting it. • Fill rates are learned per venue per symbol, so identical orders can be routed differently by different brokers. • Order protection restricts the router; best execution governs the broker, whose policy sets the objective. • Small orders are improved a cent or two by internalisation; large orders have to be carried by lit venues. When a broker claims its routing is “always best”, the claim is testable against its own disclosures. Compare its price improvement and fill statistics with a peer’s on the same kind of order, and the marketing resolves into a number.

Sources

• Rule 606 — order routing disclosuresSEC — Regulation NMS
• Rule 605 — market quality statistics, expanded in 2024 to firms with 100,000+ customer accountsSEC — Rule 605 amendments
• Payment for order flow and the retail execution debateSEC — Equity market structure literature review