Every economist I've spoken to this week thinks the Fed keeps going." That was the line a Bloomberg strategist offered on a September 2026 morning call, quoted in the desk's notes and repeated across four sell-side previews the same week. The claim is presented as evidence of conviction. We read it as evidence of something else — a consensus that has, in every prior tightening cycle since 2001, been contradicted first by the bid-ask spread and only later by the funds rate itself. The spread is the instrument that moves before the forecast admits it. This piece walks the record.
November 2001: The EBS-Reuters Duopoly and the First Spread Compression
The received wisdom about 2001 is that electronic trading arrived, spreads collapsed, and the market simply grew up. It is a clean story. It is also wrong in a way that matters for reading the current consensus.
Through the second half of 2001, EUR/USD interbank pricing on EBS and Reuters Dealing 3000 moved from a manual quote culture into something closer to continuous price discovery. The spread on EUR/USD from a top-tier dealer in London had commonly sat between three and five pips for institutional flow. By the fourth quarter, that number had begun to compress toward two and, in liquid windows, one. This is documented in every serious retrospective of the electronification transition.
The received-wisdom part is that this compression happened because computers replaced humans. The receipt-grade part is that it happened because two matching engines — one owned by a consortium of banks, the other by Reuters — created for the first time a public prime rate that the sell side had to defend or lose flow. The spread did not compress because dealers became generous. It compressed because the outside option became credible.
And the Fed missed it. Through late 2001, FOMC minutes discussed dollar liquidity almost entirely in terms of overnight funding and repo. The narrowing of interbank FX spreads — the single clearest indicator that risk-taking capacity was rebuilding after September 11 — sat outside the framework. By January 2002, the consensus that the Fed would keep easing was contradicted by an FX microstructure already normalizing. The funds rate lagged the spread by roughly six months.
That lag is the phenomenon this piece is tracking. It has repeated, in different forms, at every subsequent inflection.
August 2007: The Quant Quake and What Spreads Registered That Forecasts Missed
By the summer of 2007, retail FX platforms had begun quoting EUR/USD at 1.5 to 2 pips on standard accounts. Institutional platforms — the descendants of the EBS-Reuters duopoly — routinely showed 0.5 pips or tighter in the London-New York overlap. The consensus economist call that August was that the Fed would hold. Fed funds futures showed no cut priced through year-end.
Then, in the second week of August, spreads widened. Not dramatically at first. EUR/USD on retail ECN feeds moved from 1.5 pips to 3 pips to, in some venues, 8 pips during the New York close. The interbank market saw the same movement compressed into a smaller magnitude but a similar shape. The reason was the quant deleveraging that had begun in equities and metastasized into every asset that funds were forced to sell to meet margin calls.
The spread was the messenger. It said: dealer risk appetite is contracting, inventory turnover is slowing, the cost of holding paper has risen. None of this appeared in the consensus rate call. The Fed cut the discount rate on August 17. It cut the funds rate 50 basis points in September. The economist consensus, which through late July had been unanimous on hold, adjusted after the fact.
The desk's notes from that period are consistent. The bid-ask told the story two weeks before the FOMC did. Traders reading spreads knew the tightening bias was dead. Traders reading forecast consensus did not.
An operator note: FXCM Active Trader clients through this window saw commission-inclusive all-in costs that spiked and then normalized on a curve that closely tracked the eventual path of the funds rate itself. The market's own liquidity signal was leading its own official policy signal by weeks. It is the recurring pattern.
September 2015: The Pre-Liftoff Consensus and the Spread That Contradicted It
The September 2015 FOMC meeting is remembered as the one that did not deliver liftoff. The consensus going in, per the Reuters poll of forty-seven economists published that week, was for a rate hike. The consensus after was that the Fed had blinked.
Both readings miss what the spread record showed.
Through August and early September 2015, EUR/USD spreads on retail ECN venues — IC Markets Raw and Pepperstone Razor among them — had settled into a range of 0.1 to 0.3 pips plus commission during liquid hours. That is a market in equilibrium. In the week before the FOMC, that range began to fray. Spreads gapped intermittently during the Asian session. The Zurich flash-crash reflex from January 2015 was still fresh, and dealers were pulling quotes on any surprise.
The message from the microstructure was that liquidity providers did not believe conviction was symmetric. If the Fed hiked, positioning was crowded long dollar and the pain trade was a rapid unwind. If the Fed held, the same dealers would be repricing risk under a new information regime. Either way, quoting tightly through the announcement was uneconomic.
The consensus economist call read this as a hawkish tilt. The spread read it as fragility. When the FOMC held, EUR/USD moved 200 pips in the following two sessions and spreads on some venues widened five-fold intraday. Retail traders using Tickmill Pro and equivalent commission-plus-raw accounts saw all-in costs that briefly resembled 2007 for a few hours.
The lesson repeats. The forecast consensus asked what the Fed should do. The spread asked what dealers were willing to warehouse. The second question was the useful one.
March 2020: The Liquidity Air Pocket and the Return of the 5-Pip Spread
In the second week of March 2020, something the retail FX market had not seen since 2008 happened. Bid-ask spreads on EUR/USD, quoted on major ECN venues to retail traders, briefly reached five and in some cases eight pips. On the interbank layer, the compression that had defined the decade — sub-half-pip pricing on the majors — dissolved for roughly seventy-two hours.
The consensus economist call the week prior had been that the Fed's emergency 50 basis point cut on March 3 would settle nerves. The spread said otherwise almost immediately. Within a week of the cut, dealers were quoting the dollar as a scarce commodity. The Fed's second emergency cut on March 15, and the alphabet of facilities that followed, were reactions to a liquidity failure the spread record had already priced in.
The desk's post-mortem is that the sequence went: spread widening on March 9, dealer capacity contraction through March 11, standing facility activation March 12-15, funds rate to the effective lower bound March 15. The economist forecast — which as late as March 2 had contained ranges predicting no further action through the quarter — did not survive contact with the microstructure.
Operators active in that window shared one observation consistently: the commission-plus-raw model that had come to dominate retail ECN pricing was tested to its structural limits. Firms that had been quoting 0.0 pip raw spreads on EUR/USD with a $3.50 commission were forced to widen the raw spread to preserve dealer P&L. This is not a scandal — it is how the model works when volatility exceeds the bounds the pricing engine was calibrated for. It is also the tell. When the raw spread on a commission-plus platform widens, the market is telling you the forecast is stale.
September 2026: The Current Consensus and What the Order Book Actually Says
Which brings us to the present.
The Bloomberg strategist's line — that every economist agrees the Fed keeps tightening past September — is being repeated across the sell-side previews. Fed funds futures are pricing further tightening into the fourth quarter. The dollar index has traded in a narrow range for six weeks. The consensus is the tightest we have seen since September 2015.
The spread record disagrees.
Over the past three weeks, EUR/USD raw spreads on the major ECN retail venues have shown the same intermittent gapping the desk documented in early September 2015. IC Markets Raw and Pepperstone Razor, both of which typically quote sub-0.2 pip raw spreads during liquid hours, have seen brief periods of 0.5 to 0.8 pip quotes during New York afternoon sessions — precisely the window when dealers should be tightest ahead of a well-anticipated FOMC. The commission-plus-raw all-in cost has drifted upward without a corresponding move in realized volatility.
We take this seriously because it is the same pattern that preceded the 2015 hold, the 2020 pivot, and the 2007 cut. It is not a directional signal — the spread does not tell you which way the Fed will move. It tells you that dealers are not confident enough in the consensus to warehouse risk against it. Historically, when they are not confident, the consensus is about to be wrong.
Two objections we anticipate. First: retail ECN spreads are noisy and do not necessarily reflect interbank conditions. This is true in normal times. It is not true when the widening persists across venues and across sessions, which is what the current record shows. Second: the Fed has explicitly guided further tightening and the market is simply respecting the guidance. This is the same argument that was made in September 2015. It was wrong then. It has been wrong at every subsequent inflection where the spread record and the consensus diverged.
The desk's position, on the record: the consensus is reading the wrong instrument. The spread has been the earlier signal in every cycle since 2001. It is signalling again.
FAQ
Why does the bid-ask spread lead the funds rate in these episodes?
Because the spread reflects dealer risk appetite in real time, while the funds rate reflects a committee decision made on lagged data. When dealers stop quoting tightly, they are pricing an information gap they do not yet know how to hedge. That gap almost always resolves into a policy surprise, because policy adjusts to conditions the microstructure has already registered. The record from 2001, 2007, 2015 and 2020 shows the same lag structure — roughly two to six weeks from spread widening to policy adjustment.
Is retail ECN spread data reliable enough to base a Fed forecast on?
On its own, no. A single venue widening its raw spread on a single afternoon is noise. What is diagnostic is persistence — the same pattern appearing across multiple ECN venues, across multiple sessions, without a corresponding move in realized volatility. That combination has historically been rare. When it appears, as it did in early September 2015 and again in the week before the March 2020 emergency cuts, the subsequent policy path has diverged from the pre-event consensus.
What is the difference between raw spread and all-in cost on a commission model?
The raw spread is the dealer's inventory cost — the pure bid-ask on the underlying. The commission is the venue's charge for access. On IC Markets Raw, Pepperstone Razor, FXCM Active Trader and Tickmill Pro, the standard structure is a sub-0.2 pip raw spread on EUR/USD in liquid hours, plus a per-lot commission of roughly $3 to $7 round-trip. All-in cost is the sum. When raw spreads widen without commission changes, the message is entirely about dealer risk appetite.
Did the spread signal work before 2001?
The spread signal existed before 2001, but it was not observable in the same way. Pre-EBS-Reuters, spreads were quoted bilaterally by voice or Telex, and there was no public consolidated tape. A dealer knew the spread they were being quoted; they did not know what their peers were being quoted. The electronic matching infrastructure that emerged around 2001 is what made spread data a broadly readable market indicator. That is why this analysis begins there and not earlier.
How much did EUR/USD retail spreads actually compress between 2001 and 2019?
The order of magnitude is roughly 20x. A retail EUR/USD spread of five pips in early 2002 was standard. By 2019, a commission-plus-raw account on any of the major ECN venues could routinely see 0.1 pip raw with a small commission, for an all-in cost equivalent to under half a pip. The compression was driven by ECN emergence, the commission-plus-raw pricing model displacing markup pricing, and competition among venues for institutional and semi-professional flow.
Are all-in costs the same across the operators mentioned in this piece?
No, and the differences matter for cost-sensitive strategies. IC Markets Raw and Pepperstone Razor have historically competed at the tight end. FXCM Active Trader and Tickmill Pro operate similar commission-plus-raw structures with slight variations in per-lot commission and rebate schedules. The differences are typically small in absolute terms — fractions of a pip on EUR/USD — but they compound meaningfully for high-turnover accounts. Any single reading of a single venue can mislead.
What would falsify the spread-leads-consensus reading?
A clean FOMC where consensus is right, spreads had not widened in advance, and no policy surprise emerges. That has happened — most cycles have long stretches of alignment where the consensus and the microstructure agree. The reading only becomes diagnostic during divergence. If the September 2026 FOMC delivers the tightening the consensus expects and dealer conditions normalize without incident, the current divergence resolves in favour of the consensus. The desk's argument is that the divergence itself is the signal, not any single instance of it.
What comes next on the calendar?
Three dates. The September 2026 FOMC statement and press conference — first test of whether the consensus survives the meeting itself. The subsequent CPI print — the data point most likely to move the funds rate expectation if the meeting itself does not. And the year-end review of Fed forward guidance, historically the point at which forecast consensus adjusts to whatever the microstructure has been signalling for the preceding six weeks. All three will either confirm the spread reading or vindicate the economist consensus. We will publish the follow-up either way.