Broker Selection and Execution Quality: How to Tell If Your Broker Is Actually Doing the Job

You pay your broker a commission. They route your order. You get a fill. How do you know if the fill was any good?

Most buy-side traders don’t ask this question often enough. They pick a broker based on a relationship, send flow for years, and never seriously evaluate whether the execution quality justifies the commission. Then one day a PM asks why the desk’s costs are 5 bps higher than a comparable shop’s, and someone has to explain why nobody ever looked.

Harris (2003, Ch. 24) calls the broker-client relationship one of the least-examined corners of market microstructure, and the reason splits into three sentences: the buy-side pays the broker to handle a complex execution problem; the broker’s incentives don’t always align with the buy-side’s; and the information asymmetry between them is large. This whole article is about the third sentence. Scorecards, parallel tests, FIX messages, TCA reconciliation — every one of them is a tool the buy-side uses to shrink that asymmetry. By the end you should be able to walk up to your brokers with noticeably sharper questions.

The Broker’s Job, Properly Understood

Before you can evaluate a broker, you need a clear list of what you’re paying for. A broker isn’t just forwarding orders to a venue; they’re doing five things at once. They decide where each child order goes — lit exchange, dark pool, ATS, internalizer — which is the smart order router logic. They run VWAP, POV, and IS algos to work parent orders over time. They keep quiet: a broker that leaks your order’s size and side to other clients is costing you money directly. They source block liquidity, especially in less-liquid names, which is where block desks and high-touch traders earn their keep. And they report: post-trade TCA, fill quality, venue analysis.

The buy-side pays for all five through commission and spread. The trouble is that very little of the five is directly observable from your side of the wall — which is precisely the shape of the information asymmetry, and precisely why the measurement apparatus below exists.

The Scorecard

Serious buy-side desks maintain a broker scorecard, and what it measures is exactly the observable projection of those five jobs.

The heaviest component is execution quality. The standard measure is implementation shortfall versus arrival price, decomposed by the broker’s fills: at equal urgency, if Broker A’s fills average 2 bps worse than arrival and Broker B’s average 5 bps worse, A wins this category. The catch is the data source — brokers report their own TCA, computed on their own data. You have to reconcile against your own fills, ideally with an independent third-party TCA provider in the loop. When your number and the broker’s number disagree, somebody is wrong, and finding out who is itself informative.

Fill rate comes second, and on its own it lies. A broker with access to deep liquidity will fill more of your orders than one without — but a broker that fills every order at the offer is also “filling,” at terrible cost. Fill rate only means something read next to execution quality.

Information leakage is the hardest to measure, but it’s detectable: if the market keeps moving just before your fills — prices front-running your order’s arrival — your broker is leaking. Third-party TCA tools or in-house signaling analysis can catch the pattern. Block liquidity access matters most in less-liquid names: can the broker actually find blocks? Score it on the fraction of orders filled as blocks, average block size, and block discount versus the displayed market. Algo quality is measured with parallel tests, covered below. And then there’s service — how fast the sales trader responds, whether they call you before or after something goes wrong. Hard to quantify, but real.

Assembled into a table, a typical scorecard looks like this (scored 1-5):

Numbers like these get refreshed quarterly. Note the intended use: the point isn’t to crown a “best” broker, it’s to identify where each one excels and allocate flow accordingly. Most desks split it — 30% to Broker A for blocks, 30% to Broker B for VWAP in liquid names, 20% to Broker C for small-cap liquidity, 20% to a fourth for special situations. That’s “broker diversification,” with the same logic as portfolio diversification — and it happens to preserve your bargaining power and your option to replace anyone at any time.

Commission Doesn’t Buy Execution

The biggest open secret in execution: commission stopped covering the real cost of execution a long time ago. Institutional commissions are typically quoted per share — 1-3 cents in US equities is common; on a $50 stock, 1 cent is 2 bps. Meanwhile the actual cost of execution (spread + impact + opportunity) usually runs 5-15 bps in liquid names and more in illiquid ones. Commission is a fraction of the real bill.

That creates the incentive problem: a broker paid by commission has no direct incentive to minimize your execution cost — only the indirect one of wanting your future flow. The indirect incentive exists, but it’s weaker than you’d like, which is why none of the measurement tools in this article are optional.

One structural mitigation is the commission sharing agreement (CSA). A CSA splits the per-share commission into an execution pot and a research pot — the execution portion goes to the broker who actually worked the order, while the research portion can be directed to third-party research providers. Once the two are unbundled, execution flow can be allocated purely on execution quality, without the hostage logic of “their research is good, so they get the orders too.” The incentives at least decouple. A separate thread is payment for order flow (PFOF): it plays essentially no role in the economics of institutional flow, but for US retail flow it’s a major revenue stream — internalizers like Citadel Securities and Virtu pay retail brokers for their order flow, under SEC disclosure rules and amid ongoing regulatory controversy. Retail and institutional economics are different animals; don’t read a broker’s financials as if they were one.

FIX Protocol: What Your Order Looks Like on the Wire

The other asymmetry-shrinking tool is lower-level: knowing what your order actually looks like in flight. Almost all institutional equity trading runs over the FIX (Financial Information eXchange) protocol, a tag-value messaging standard in use since 1992. A new order is a message like this:

8=FIX.4.2|9=178|35=D|49=BUYSIDE|56=BROKER|11=ORD12345|21=1|55=AAPL|54=1|60=20160115-10:30:00|38=10000|40=2|44=195.50|10=128

The key fields: 35=D is “new order single,” 55=AAPL is the symbol, 54=1 means buy, 38=10000 is 10,000 shares, 40=2 means limit order, 44=195.50 is the limit price. The broker receives it, processes it, and sends back an execution report:

8=FIX.4.2|9=156|35=8|49=BROKER|56=BUYSIDE|17=EXEC9876|11=ORD12345|150=1|39=1|55=AAPL|54=1|38=10000|14=5000|6=195.50|60=20160115-10:30:05|10=200

This says “ORD12345, 5,000 of 10,000 shares filled at an average of 195.50” (150=1/39=1 mark it as a partial fill). More reports arrive as the rest fills.

FIX defines dozens of order types and time-in-force values. The ones a trader actually meets:

• 40=1: Market order
• 40=2: Limit order
• 40=3: Stop order
• 40=4: Stop limit
• 40=P: Pegged (auto-adjusts to bid/ask/mid)

Time-in-force:

• 59=0: Day order
• 59=1: GTC (good till cancel)
• 59=2: OPG (opening auction only)
• 59=3: IOC (immediate or cancel)
• 59=4: FOK (fill or kill)
• 59=5: GTX (good till crossing)
• 59=6: GTD (good till date)

Venues differ in the details — some exchanges accept IOC and some don’t, some take FOK for blocks and some don’t. The broker’s systems handle the mapping, but the trader should know what’s being sent. You don’t need to master FIX to trade; you need to know that every order travels as a structured message, because that’s what makes a missed fill, a wrong price, or a delayed acknowledgment debuggable. Most EMS platforms log the raw FIX messages and let you inspect them.

The EMS and the OMS

The EMS (execution management system) is the buy-side trader’s workstation, connected to brokers, venues, and data sources. Common platforms:

• Bloomberg EMSX
• Charles River IMS
• Eze EMS (now part of SS&C)
• FlexTrade
• TradingScreen

Each has its lean: Bloomberg EMSX has the best data integration (it’s Bloomberg), Charles River is the dominant OMS/EMS combination for traditional asset managers, Eze is strong in hedge fund workflows, FlexTrade is common in high-touch equity trading. The EMS provides the blotter (working orders), order entry (generating those FIX messages), real-time market data, charts and analytics, controls for broker algos, and TCA integration. For an institutional trader it’s the main tool — you live in it 8-10 hours a day.

Its counterpart, the OMS (order management system), belongs to the PM and the back office: positions, compliance, trade booking. Common ones include Charles River, Eze, and Bloomberg AIM. The division of labor: the trader receives an order from the PM via the OMS, works it in the EMS, and the fills flow back to the OMS for booking. For execution quality, the EMS is the one that matters. The OMS just has to keep up.

Parallel Tests: How Algo Quality Gets Measured

The standard process for evaluating a broker’s algos is not complicated. Pick a basket of orders drawn from the same distribution you actually trade — same stocks, similar size and urgency. Send the same orders to multiple brokers’ algos at the same time. Measure the fills against arrival price, VWAP, and closing price. Then look at variance: for most shops, a consistent low-variance algo beats a high-mean, high-variance one. Most brokers will let you run a parallel test before you commit volume; the serious ones expect it.

def evaluate_algo(fills, arrival_price, side):
    """Compute basic algo quality metrics from fills."""
    if side == 'buy':
        shortfalls = [(p - arrival_price) / arrival_price for p in fills]
    else:
        shortfalls = [(arrival_price - p) / arrival_price for p in fills]
    return {
        'mean_shortfall_bps': 10000 * sum(shortfalls) / len(shortfalls),
        'std_shortfall_bps': 10000 * (sum((s - sum(shortfalls)/len(shortfalls))**2 
                                for s in shortfalls) / len(shortfalls)) ** 0.5,
        'fill_rate': len(fills) / len(fills)  # placeholder - a real fill rate needs the parent order size
    }

The 80/20 of algo evaluation is mean shortfall. The other 20% is variance and out-of-sample testing.

Where Else the Money Leaks

Beyond headline commission and execution cost, brokers have a few less visible extraction channels, each worth watching.

Information leakage is the ugliest: a broker that tells its prop desk about your large order before it executes has already handed the other side of your trade to its own people by the time your order reaches the market. Illegal in most jurisdictions, unevenly enforced. Internalization is far milder: the broker fills your order off-exchange against its own inventory or another client’s order, at the NBBO or better — but the flip side of “or better” is that you might have done better still on a lit exchange. Soft dollars are the old research-for-commission arrangement: you pay 3 cents a share, the broker keeps 1 cent and spends 2 paying for third-party research. The SEC requires soft dollar arrangements to be disclosed and to benefit the client, but the line between “research” and “broker benefit” has never been crisp. And PFOF, as above: retail flow gets paid for, institutional flow doesn’t, but the broker’s overall economics are shaped by both sides of its business.

A good buy-side desk monitors these channels. The really serious ones run signaling analysis: they watch the market in the moments before their orders arrive, looking for evidence that someone traded ahead.

The Bar for Firing Is Lower Than You Think

Brokers should be evaluated quarterly. The common triggers for cutting or ending flow: persistent underperformance on execution quality (IS worse than peers by 2+ bps over a quarter), detected information leakage, TCA reports that don’t show up on time, the departure of key sales traders (relationships do matter), or compliance trouble — a broker under investigation.

Most desks overweight the gravity of firing. Run the arithmetic: a broker that’s consistently 3 bps worse on $500M of annual flow costs you $150,000 a year in bad execution. Over three years, that’s pushing half a million. Brokers are replaceable, and that number is the reason.

The relationship variable has to be managed against that backdrop. Execution is relationship-driven — the buy-side trader calls the sales trader, and the better the relationship, the more attention you get when you need a block or hit a special situation. But relationships get exploited: the trader who always lunches with one broker’s sales team is probably over-allocating to them. The professional posture is to maintain several relationships, allocate by scorecard, and never let one relationship dominate. That’s what the scorecard is for — making sure the relationship can’t outvote the numbers.

DMA and HFT: Two Boundary Questions

Some buy-side firms skip the broker’s routing entirely with DMA (direct market access), wiring their EMS straight to the exchanges. That removes the broker’s markup on execution, at the cost of owning routing, compliance, and venue selection yourself. It makes sense for firms with heavy flow and serious infrastructure; for most desks, the broker’s routing is good enough that DMA isn’t worth it. The middle path is sponsored DMA — the broker sponsors your direct connection while you keep your own routing — which some firms reserve for the most liquid names where execution quality matters most.

The other boundary is high-frequency trading. Buy-side algos are mostly slow, working orders over minutes to hours; HFT firms trade in microseconds and coexist with buy-side flow by providing liquidity at the touch. The relationship is symbiotic — HFT needs buy-side flow to make markets against, and the buy-side needs HFT liquidity to fill at tight spreads — but the tension is real too: aggressive HFT can detect slow buy-side flow and trade ahead of it. Harris (2003, Ch. 22) works through this at length. The short version: the buy-side benefits from HFT liquidity and pays for it through information leakage, so the optimal play is to use that liquidity for fills while minimizing the signal your order sends.

Compliance Makes All of This a Duty

Finally, this evaluation apparatus isn’t just good hygiene — it has regulatory teeth. US brokers owe a duty of best execution (FINRA Rule 5310, operating alongside Reg NMS’s order-protection rules): reasonable diligence to obtain the most favorable terms for the client’s order, examined by the SEC and enforced by FINRA. The buy-side carries a parallel obligation. Most large buy-side firms have best-execution policies that require evaluating broker execution quality and documenting the evaluation — and if you can’t show you evaluated brokers and chose on quality, the regulator can come asking you too.

The practical upshot is a written version of everything in this article: keep the scorecards, document the evaluations, run the parallel tests, and maintain a best-execution policy on paper. Most buy-side desks have this. The ones that don’t have stacked compliance risk on top of their information asymmetry.

References

• Harris, L. (2003). Trading and Exchanges: Market Microstructure for Practitioners. Oxford University Press. Ch. 24. ISBN: 978-0-19-514470-3.
• FIX Trading Community. FIX Protocol Specification. Version 4.2 / 5.0.
• SEC Regulation NMS (2005). Rules 611 and 610.
• Almgren, R., & Chriss, N. (2000). “Optimal Execution of Portfolio Transactions.” Journal of Risk, 3(2), 5-39.