A spread, narrowly defined, is the distance between the best bid and the best offer. That definition is also, for the institutional desks reading this, the least interesting thing about it.
By the time a fill prints on a desk's ticket, the headline spread has been deformed by at least six distinct cost components. Some of them are visible in the venue's published feed. Most of them are not. The desks that consistently outperform their peers do so not by negotiating tighter top-of-book — that is a commoditised exercise — but by understanding exactly which of those six components are taxing them, and at what rate.
This is the anatomy of an effective spread in 2026. It is presented in the order the model breaks down for a typical institutional FX or index-CFD ticket: from the moment the trader's intent forms, to the moment cash settles.
Component 1: Quoted top-of-book
The visible spread on a streaming feed is the cheapest, most-shopped, most-honest number on the desk. It is also the least predictive of what a real fill will cost. In the EUR/USD inter-bank market on a normal afternoon, the top-of-book spread compresses to fractions of a pip; in the same currency pair across a non-farm payrolls release, it widens to multiples. The information conveyed by the quoted top-of-book is therefore not a forecast of execution cost. It is a forecast of conditions.
Reading conditions from quoted spread is one of the few things a counterparty can do without paying for the privilege. A desk that watches the median-five-minute spread across its top six LPs through a London-New York handover can usually anticipate which side of the book will be cheaper for the next twenty minutes. That is not edge — but it is hygiene.
Component 2: Available depth at the working price
Top-of-book is a single tick. Real institutional flow consumes the next four or five ticks. The implied cost of moving down the stack to the volume the desk actually needs to clear is the second component, and the one that distinguishes a well-routed counterparty from a poorly-routed one.
On a healthy LP stream, the second tick is two to four times the volume of the top tick, and the third another two to four times the second. On a stressed stream, the second tick is half the first and the third is missing entirely. Either way, the spread the desk pays for a child order at three ticks' worth of notional is the volume-weighted average price across the consumed levels, minus the mid at the moment of decision. That number is rarely the same as the quoted spread, and the difference compounds across every clip.
Drovix's pricing engine continuously reconstructs depth across its connected LPs and exposes a synthetic stack that aggregates the visible volume per tick. The mechanics of that aggregation — and the reason it is harder than it looks — are the subject of the capacity curve of aggregated liquidity.
Component 3: Adverse selection
This is the cost that hides in plain sight. When a desk takes liquidity, it is by definition trading against an LP that has, in the moment, less information than the taker. The LP knows this — it cannot identify which counterparty is the well-informed one, so it prices a small adverse-selection premium into every quote. That premium is felt by every counterparty, regardless of whether the flow is in fact informed.
In quoted-spread terms, the adverse-selection premium adds somewhere between 0.1 and 0.4 pips to liquid majors in normal conditions, and substantially more during news events. It is not negotiable in the conventional sense; the LP cannot remove it without bleeding to the informed minority. But it is sensitive to one variable the counterparty controls: the observed toxicity of the desk's prior flow.
Desks that are consistently identified as uninformed — meaning their post-trade mark-out over the next thirty seconds is essentially flat — pay a lower adverse-selection premium across all venues. Desks that are identified as informed pay a higher one. The half-life of that identification, the rate at which an LP forgets a bad streak of fills, is the topic of the half-life of information. It governs how long good behaviour has to be sustained before the discount lands.
Component 4: Last look (when present)
On streams that allow last look, the LP retains the right to reject a fill within a small holding window — typically 5 to 50 milliseconds. Rejection is not a cost in itself; the desk simply does not get filled. But the residual market move during the holding window, combined with the desk's adjusted timing on the re-quote, frequently is.
The honest version of last look — symmetric, time-bounded, audited per-LP — is a tool that lets the LP defend against latency arbitrage and tighten its top-of-book by a measurable margin. The dishonest version — asymmetric, opaque, with rejections clustered on losing fills and acceptances on winning ones — is a hidden cost that does not appear anywhere on the ticket.
Detecting which version a given LP is running requires per-LP reject taxonomy, mark-out at the reject time, and a non-trivial statistical test on the reject distribution. The desks that do this well usually have a counterparty that does it for them. Asymmetric last look and where the bias hides is the full piece on this.
Component 5: Market impact during the clip
On any clip of meaningful size, the act of taking liquidity moves the price against the taker by an amount that is roughly proportional to the square root of the clip's notional relative to the venue's normal depth. Square-root impact is not a heuristic; it is what falls out of half a century of empirical microstructure work, and it holds across asset classes, exchange types and time periods within an order of magnitude.
The desk's defence against impact is twofold. First: clip-size discipline — never ask the market for more than it is offering at any given level. Second: routing — splitting the parent order across venues so that no single LP absorbs more than the fraction of its instantaneous depth that triggers visible quote regression.
The combined effect of clip-size discipline and routing is the difference between paying 1.2 pips of effective spread on EUR/USD in size and paying 4.5. That is not a 30% improvement; it is a 275% improvement. Square-root impact is the most expensive component on any clip above five times the venue's top-tick depth, and the one that responds most to engineering.
Component 6: Settlement, funding, and the spread of money
The final component is the one most often overlooked because it does not appear on the trading platform at all. The cost of holding the position from execution to settlement, the cost of margin posted against it, and the spread the counterparty pays to fund collateral in the relevant currency, all roll up into the realised cost of the trade.
On a one-day FX position, the funding component is typically a fraction of a pip. On a leveraged equity-CFD position held for a quarter, the funding cost can dwarf the entry spread by an order of magnitude. The relevant question for an institutional desk is whether the funding rate it is being charged moves in sympathy with broader market stress — that is, whether the counterparty's margin and funding are pro-cyclical. A pro-cyclical counterparty raises your funding cost when you can least absorb it.
Putting it together: the realised spread
Across the six components, the realised effective spread on a typical institutional EUR/USD clip of $10M notional in normal conditions decomposes roughly as follows: visible top-of-book ~0.2 pips, depth degradation ~0.3, adverse-selection premium ~0.2, last-look residual (where applicable) ~0.1, square-root impact ~0.4, funding (per-day basis) ~0.05. Total: about 1.25 pips of realised cost on a quoted 0.2-pip spread.
The components are roughly additive in normal conditions and roughly multiplicative under stress. In a stress window the same clip can easily realise 4 to 6 pips of effective spread, almost entirely driven by depth degradation and impact compounding into each other.
What the desks that win actually do
They do not negotiate the quoted spread. That fight is fully priced. They negotiate the components above it: a counterparty with deeper aggregated depth and proven LP curation reduces components two and five. A counterparty with a documented last-look policy and per-LP reject reporting reduces component four. A counterparty that publishes per-ticket TCA with mark-out windows lets the desk continuously verify components three and four against reality.
This is the work Drovix is built around. Aggregated depth from a curated panel of bank and non-bank LPs, instrumented for per-LP fill and reject behaviour, with the resulting TCA exposed to the counterparty on the same surface the desk uses to trade. The headline spread is whatever the LP panel happens to be quoting at the moment. The realised spread — the only one that matters — is what the engineering on the other side of the connection delivers.
If a desk wants to know which of the six components is taxing it most, the answer is in the trade tape. Drovix runs that analysis on every fill, for every counterparty, every day.
Methodology and data
The six-component decomposition reflects a survey of 18 institutional FX desks (10 brokers, 6 hedge funds, 2 prop firms) executing in EUR/USD, USD/JPY, GBP/USD, and gold during the trailing twelve months ending 2026-04. Component magnitudes are weighted means of per-desk effective-spread reports reconciled against streaming top-of-book medians captured by Drovix's pricing engine. Adverse-selection cost is estimated from 30-second post-trade mark-out (mid-to-mid). Sample size: 4.2 million parent orders, 71 million child clips. All figures are normal-conditions averages excluding the 2026-03 volatility spike.
Limitations and scope
Sample skews toward European-hours flow and notionals in the $1M–$25M per-clip range. Crypto-CFD pairs are excluded because their pricing topology (no genuine wholesale tier-1) makes the six-component model inapplicable. Mark-out windows shorter than 30 seconds materially overstate adverse selection on slow-decay flow; longer windows understate it. The decomposition is descriptive, not predictive — desks should reproduce it on their own tape before sizing risk against any single component.
Further reading
→ Architecture of a Fair Spread — How fair pricing is constructed before it is consumed. See /blog/architecture-of-a-fair-spread.
→ Execution Cost Decomposition — The five-component TCA framework used on institutional desks. See /blog/execution-cost-decomposition.
→ Signed Flow Toxicity and Information Half-Life — Why some flow ages faster than others. See /blog/signed-flow-toxicity-and-information-half-life.
Drovix Research is the in-house institutional desk of Drovix MU Ltd, regulated by the Financial Services Commission of Mauritius. All notes are informational only and do not constitute investment advice, a solicitation, or a recommendation to transact in any financial instrument.
Analyst Desk
Drovix Research Desk
Institutional Research
Drovix Research Desk publishes institutional-grade analysis covering macro events, cross-asset correlations, and execution insights for professional market participants.