Onpode
Cover art for Why bid-ask spreads and order flow matter more than you think

Why bid-ask spreads and order flow matter more than you think

November 14, 2026 · 14 min

Eliza Ward & Brian Reed

Eliza, hey — I want to start with a confession: I thought I understood what a bid-ask spread was. Past tense. Past tense. Because I was looking at some research this week — Stephan Grimm and Thomas Guhr — and their finding is that the spread widens more often from order deletions, from cancellations, than…

Market microstructure is the academic and practical study of how financial transactions are mechanically executed — encompassing the rules, technologies, and participant behaviors that govern how orders meet, how liquidity is supplied, and how information is incorporated into prices.

0:0014:06
Get the next episode on Finance →

Follow it free — new episodes land in your feed.

Or make your own — any topic, in minutes

More Onpode episodes on Finance →

About this episode

Market microstructure is the academic and practical study of how financial transactions are mechanically executed — encompassing the rules, technologies, and participant behaviors that govern how orders meet, how liquidity is supplied, and how information is incorporated into prices.

Grounded in 12 sources
How spread changes affect the order book: Comparing the price responses of order deletions and placements to trades ↗ · arxiv.org
Limit Order Book Dynamics in Matching Markets: Microstructure, Spread, and Execution Slippage ↗ · arxiv.org
Empirical Study of Market Impact Conditional on Order-Flow Imbalance ↗ · arxiv.org
Understanding the Impacts of Dark Pools on Price Discovery ↗ · arxiv.org
Information thermodynamics of financial markets: the Glosten-Milgrom model ↗ · arxiv.org
Microstructure and Empirical ↗ · assets.cambridge.org
The High-Frequency Trading Arms Race: Frequent Batch Auctions as a Market Design Response ↗ · doi.org
Limit order books ↗ · doi.org
Essays on Liquidity Risk and Modern Market Microstructure ↗ · doi.org
G losten– M ilgrom Models ↗ · doi.org
Introduction to the Market Microstructure Symposium ↗ · doi.org
Bid, ask and transaction prices in a specialist market with heterogeneously informed traders ☆ ↗ · sciencedirect.com
Read transcript

Brian Reed: Eliza, hey — I want to start with a confession: I thought I understood what a bid-ask spread was.

Eliza Ward: Past tense.

Brian Reed: Past tense. Because I was looking at some research this week — Stephan Grimm and Thomas Guhr — and their finding is that the spread widens more often from order deletions, from cancellations, than from actual trades executing. And I genuinely did not have a category for that.

Eliza Ward: What specifically broke your mental model?

Brian Reed: The part where nothing happens and the cost still moves. No trade prints. Someone just pulls a bid or an offer, and the spread — the gap between what you can buy at and what you can sell at — that widens. And almost no execution cost framework is measuring that.

Eliza Ward: Right. And the reason it's invisible is partly historical — market microstructure only became a formal field in the 1970s, and the whole analytical tradition built up around trades as the event worth studying. The Central Limit Order Book, the CLOB, runs on price-time priority and matches at the moment of execution. So 'execution' became the measurement unit. Cancellations aren't execution. They just... reshape the environment silently.

Brian Reed: And the CLOB is basically everywhere — right, more than half of global markets?

Eliza Ward: More than half, yeah. It's the dominant structure. Which means if the CLOB architecture is what makes deletions consequential — because of how the queue works, how depth forms and disappears — then this isn't a niche problem.

Brian Reed: It's everywhere the CLOB is.

Eliza Ward: Market microstructure isn't a passive backdrop — that's the thing. It's an active friction-generating mechanism. It taxes every trader independent of price level, independent of volatility. The deletion finding is the clearest example of a cost that the structure produces and the models miss.

Brian Reed: So if we take that seriously — that the mechanism itself generates the friction — what does that change about how we think about who's actually paying, and why?

Eliza Ward: Who's paying is the right question, and the answer starts one layer back — with what the spread actually is before we get to what moves it. The spread is a tax. Not a metaphor. The literal, immediate cost of needing to buy or sell right now instead of waiting.

Brian Reed: So the spread isn't greed — it's just... math?

Eliza Ward: That's exactly the frame. Think about a currency exchange booth at an airport — they buy your euros cheap and sell them back expensive. Not because they're evil. Because the person across the counter might know something they don't. Maybe the euro just moved on Reuters and you've got your phone. They have no idea. So the gap in the rate is protection, not profit.

Brian Reed: The booth can't tell if you're just traveling or if you just checked the news.

Eliza Ward: Right — and that's exactly the problem Glosten and Milgrom formalized in 1985. The model says market makers can't see who they're trading against. So they set the bid-ask spread wide enough to cover the expected losses from informed traders, using the gains from everyone else. The technical term for that risk — selling into rising prices, buying into falling ones — is adverse selection.

Brian Reed: And 'everyone else' is the noise traders.

Eliza Ward: Noise traders, yeah — people buying for reasons that have nothing to do with private information. Pension fund rebalancing, an ETF tracking an index, someone just needs cash. They're not informed. And they're actually... load-bearing. Without enough uninformed flow, the market maker can't offset what they lose to the smart money, so the spread has to widen, or they just pull quotes entirely.

Brian Reed: Hold on — so liquidity actually depends on enough uninformed traders showing up?

Eliza Ward: Structurally, yes. The Glosten-Milgrom model makes that explicit — it's not incidental. Noise traders subsidize price discovery. Without them the math breaks.

Brian Reed: Okay and so — the part I don't get — where does the Central Limit Order Book actually fit in this? Like, the CLOB is the mechanism, but how does price-time priority change the spread-setting calculus?

Eliza Ward: So price-time priority means if two market makers quote the same price, whoever got there first gets filled first. Which creates this competitive pressure to post tighter quotes — better price to jump the queue, faster to hold your spot. That compression is actually good in normal conditions. It's what makes spreads narrow. But it also means the whole structure is wired around speed and position, and that has consequences when — actually, let me make this concrete.

Brian Reed: Please.

Eliza Ward: Say a museum's endowment fund needs to sell a block of municipal bonds on a Friday afternoon — no urgency, no information edge, pure liquidity need. They're textbook noise traders. The market maker on the other side has posted a resting limit order, earned their queue position. They're seeing this order come in, they don't know if it's the endowment or someone who just got off a call with the CFO. So the spread they quoted already baked in that uncertainty. The endowment pays it regardless. That's the adverse selection premium landing on someone who has zero information.

Brian Reed: And that endowment — she's already paid the adverse selection premium before we even get to the deletion problem. Which is the part that actually breaks the model.

Eliza Ward: Break it how?

Brian Reed: So take a pension fund operations manager — it's 2 p.m. on a Friday, she's executing a routine bond index rebalance. Nothing urgent. She pulls up the order book, sees what looks like reasonable depth on both sides — resting limit orders sitting there. She starts her order. And by the time it lands, the book has thinned. Not because a trade happened. Because limit orders were cancelled in the seconds before her order arrived.

Eliza Ward: And her model didn't see that.

Brian Reed: Her model was looking at the last trade print. Which told her nothing about the live state of the book at the moment of execution. And that's the Grimm and Guhr finding made concrete — deletions widen spreads more frequently than actual trades do, and the standard execution cost framework is measuring trades. So she's flying on instruments that are reading stale air.

Eliza Ward: Wait — and the depth figure she saw was real when she saw it.

Brian Reed: Exactly — the order book depth was genuine at that moment. Resting limit orders, volume sitting at and around the best bid and ask. It just... wasn't there anymore when she needed it. And the slippage she gets isn't because she's informed, isn't because volatility spiked — it's because the book thinned from cancellations and her realized price ran above her model's estimate. That gap between intended and actual execution price — that's the market impact she's now explaining to her investment committee.

Eliza Ward: And order flow imbalance makes it worse, right? Because the cancellations themselves distort the signal about directional pressure.

Brian Reed: Yeah — I mean, the net buy-versus-sell pressure in the book is a significant predictor of short-term price impact. That's well established. But if cancellations are pulling bids or offers, the imbalance signal is... it's corrupted. You think you're reading supply and demand. You're partly reading who decided to leave.

Eliza Ward: And nobody's cost model has a line item for that.

Brian Reed: Most public execution cost frameworks — they're built around trade-driven slippage. Cancellations are just... absent from the accounting. So you get this systematic undercount of actual friction. The pension manager's post-trade report looks like a rounding error. But it's structural.

Eliza Ward: And actually — that gets worse when you scatter the same asset across multiple venues. The fragmentation piece, the dark pool sorting problem — that's where this hidden deletion effect compounds in ways the lit-venue spread alone really can't capture.

Brian Reed: So the Friday-afternoon pension manager isn't just fighting the deletion problem on one book — she might be watching depth that's spread thin across venues she can't even fully see.

Eliza Ward: Right — and that's the illusion. Venue fragmentation means the depth she's looking at is real, it's just... scattered. Resting orders split across a half-dozen exchanges, maybe some sitting in dark pools. The headline number looks fine. The actual fillable liquidity at the moment her order hits? That's a different question entirely.

Brian Reed: So the total volume is there, it's just not — accessible simultaneously.

Eliza Ward: You can't reach all of it at once. The orders are resting across venues that aren't connected in real time. So when her order arrives, she's executing against whatever's local, and the rest of the depth — it might as well not exist for the next few milliseconds.

Brian Reed: That's the illusion of depth.

Eliza Ward: And dark pools make this more complicated, not less. Because dark pools aren't random — Linlin Ye's model shows there's a sorting equilibrium. Strong-signal informed traders stay on lit exchanges. Moderate-signal traders move to dark pools specifically to reduce their adverse selection exposure. Weak-signal traders don't trade at all.

Brian Reed: Hang on — so dark pools are pulling a specific type of informed flow away from the lit market?

Eliza Ward: Selectively, yeah. Which sounds like it might help the lit exchange — fewer informed traders to worry about — but actually, wait, that's not quite right either. Because price discovery depends on that moderate-signal information making it into prices. If it routes to a dark pool instead, the lit venue just... misses it.

Brian Reed: The price on the lit exchange gets worse at reflecting what's actually known.

Eliza Ward: Price discovery degrades. That's the casualty. Individual traders using dark pools may get better execution prices — the mechanism genuinely works for them. But the aggregate, the consensus fair value that everyone else is relying on? It's being built on less information.

Brian Reed: So the fix for one trader's adverse selection problem is everyone else's price discovery problem.

Eliza Ward: That's the tension, yeah. And it's why Budish, Cramton, and Shim — their QJE paper — goes after the continuous matching rule itself rather than any individual venue. The argument is that price-time priority in continuous time creates a race to be first at infinitesimal speed advantages. HFTs investing in microsecond infrastructure not to discover prices, but just to be first in queue. Socially wasteful, by their framing.

Brian Reed: And their proposed fix is frequent batch auctions — orders accumulate over, what, a hundred milliseconds, then clear at a uniform price. Speed stops mattering.

Eliza Ward: Right. The speed advantage disappears because everyone inside that window gets the same price. The arms race becomes pointless. In their model, market quality holds — actually, I mean, the theoretical case is tight. But.

Brian Reed: Has anyone actually tried it?

Eliza Ward: Not at scale. That's the gap. Martin Gould and colleagues — their landmark LOB survey — essentially flags this: after decades of research, the relationship between microstructure design and price discovery efficiency is still an open empirical problem. The batch auction proposal is rigorous. It just hasn't been tested where it would actually matter.

Brian Reed: And that's what strikes me — not the batch auction question specifically, but what it implies. Because if the fix has to be a different machine, a different architecture, then everything traders do to manage costs right now — avoiding certain times of day, waiting for volatility to settle — that's all... it's aimed at the wrong target. The friction isn't circumstantial. It's baked in.

Eliza Ward: Yeah. That's actually the cleanest way I've heard it put. The spread isn't wide because the market is nervous today. It's wide because the mechanism — price-time priority, continuous matching, the deletion dynamics that Grimm and Guhr measured — the mechanism is designed that way. Wider spreads, thinner books, fragmented venues, adverse selection — those aren't weather. They're structure.

Brian Reed: So timing the trade better doesn't fix it.

Eliza Ward: It might shave something at the margin. But the floor — the irreducible friction — is still there. The Central Limit Order Book produces it in calm markets too, not just stressed ones. That's the uncomfortable part for anyone who built a whole cost-control strategy around reading volatility signals.

Brian Reed: I think that's where I actually landed this week. Not that the system is broken in some dramatic way — just that the costs most managers are tracking are, let me think — they're a subset. The deletion-driven spread widening, the fragmentation hiding real depth, the adverse selection penalty landing on people with no information edge. Those aren't anomalies. They're the system working as designed.

Eliza Ward: Working as designed. That's a decent place to stop, actually. Grimm and Guhr didn't find a bug. Glosten-Milgrom in '85 didn't describe a failure. They described the machine. The machine just costs more than the invoice says.