Lila Soto: Hugo, I have been sitting with this image all week and I need to know if it holds. A farmer's market vendor, the one who always has apples, always willing to buy or sell — but they need a small cut on every transaction just to keep the lights on. One Tuesday morning, a thousand customers show up at once. What does that vendor do?
Hugo Vance: Well. They panic, raise their prices, run out of stock, and close the stall.
Lila Soto: That's GameStop. January 2021. A thousand percent spike — and the part nobody explained well was that this wasn't really about retail investors discovering value. It was a stress test. And the stall closed.
Hugo Vance: The coordinated retail buy orders overwhelmed market makers' inventory capacity. They widened spreads and retreated. Which cascaded into a liquidity crisis. And the supply-and-demand explanation — the official story — told you almost nothing about why that sequence happened in that order.
Lila Soto: Nothing. And that gap — between the tidy story and the actual mechanism — that's kind of where I want to start today. Because the real question is: what is actually doing the work of price discovery?
Hugo Vance: Not a supply-and-demand curve. A data structure. The Central Limit Order Book — which is what every serious exchange runs on. NYSE, NASDAQ, CME, Eurex, the major crypto venues. Every resting bid and every resting ask, consolidated, matched by price and then by time of arrival. That is the engine of price formation.
Lila Soto: So price discovery — the process of turning a thousand private intentions into one number on a screen — it's happening inside that structure.
Hugo Vance: Continuously. Every incoming order either consumes what's resting in the book or adds to it. The bid-ask spread — the gap between the best resting buy price and the best resting sell price — is where you see the cost of that process in real time.
Lila Soto: And that spread is doing two jobs at once, right? It's the fee you pay if you want to trade right now. And it's how the market maker — the vendor — pays rent.
Hugo Vance: Precisely. Simultaneously the cost of immediacy for whoever takes the trade, and the primary revenue source for whoever is providing it. When that spread widens — as it did dramatically during GameStop — you are watching the vendor raise apple prices because the crowd got too large and too unpredictable.
Lila Soto: Hm — and the unpredictability is the thing. They don't know who's in that crowd or what they know.
Hugo Vance: That is exactly the problem we are going to be inside all episode. Information asymmetry. The market maker's permanent condition.
Lila Soto: But that's — wait, I want to push on what 'resting' actually means, because I think that's where the clean picture breaks down.
Hugo Vance: Good. Because it does break down. A limit order is — you see, it's a promise. 'I will buy at this price if you bring it to me.' It sits in the Central Limit Order Book waiting. A market order is the opposite: a demand. 'Sell me now, whatever it costs.' And when a market order arrives, it eats the best resting limit order first.
Lila Soto: Eats it — meaning it's gone.
Hugo Vance: Gone. Consumed. And if the market order is large enough, it consumes the next price level. And the next. The price moves — not because sentiment shifted, but because the orders at that level were exhausted. That's price impact. Walking the book.
Lila Soto: So the price we see on a screen — it's not a consensus. It's more like... a scar left by the most recent collision.
Hugo Vance: That is actually a precise description. The last traded price is the residue of a market order eating through whatever happened to be resting there. And that 'whatever happened to be resting' — the depth, the available liquidity at each level — varies tick by tick.
Lila Soto: And a big order — a fund manager unloading, say, fifty million dollars in a mid-cap — they can't hide that, right? The book just... reveals them.
Hugo Vance: Well — yes, and here's what I find genuinely striking about the mechanics. That fund manager hits the book with a limit order two percent below the current bid. Twenty seconds of nothing. Then a market maker fills half the size — exactly half — at that price. She's been seen. The spread widens immediately after. Her remaining position now costs more to exit.
Lila Soto: Oh — that's order flow revealing intention before the price has actually moved.
Hugo Vance: Precisely. And the formalization of all this — the CLOB as the dominant mechanism, market microstructure as a formal academic discipline — that happened fast. The 1970s and 1980s. Open-outcry floor trading at NYSE and other exchanges, replaced by electronic matching engines within roughly a decade. And almost overnight, there were papers, models, the Glosten-Milgrom model in 1985 formalizing exactly this problem — how should a market maker price trades when some counterparties know more than they do.
Lila Soto: So the discipline was born trying to answer a question the floor traders had just — kind of muddled through by instinct for centuries.
Hugo Vance: By relationship, by reputation, by reading the room. The CLOB mechanized the problem but did not solve it. Price impact is still proportional to order size relative to depth. The book is still a live ledger of intention that resets every tick. We just made it invisible to most of the people trading in it.
Lila Soto: Invisible — yeah, that's the part I keep snagging on. Because someone has to keep that ledger populated. Like, who actually fills it? Who keeps posting bids and asks when everybody else is just waiting?
Hugo Vance: Market makers. That is their entire job — post a bid, post an ask, simultaneously, thousands of times a day. Earn the spread on each fill. Repeat.
Lila Soto: Which sounds almost mechanical. Until you realize what they're absorbing.
Hugo Vance: Two risks. The first is inventory risk — if order flow runs one direction, they accumulate a position they never wanted. Net long in a falling market. Net short in a rising one. That's painful but manageable. The second risk is — well, it's more existential. Adverse selection.
Lila Soto: Trading against someone who already knows the answer.
Hugo Vance: Precisely. And this is what Glosten and Milgrom formalized in 1985 — the first rigorous treatment of that specific problem. Their insight was that the market maker cannot distinguish an informed trader from a noise trader in real time. So the spread becomes — I mean, it's not just a fee. It's a hedge against ignorance. Every quote is priced to survive the possibility that the other side knows more.
Lila Soto: Oh — so the spread is literally the market maker saying 'I don't know who you are, and that costs you.'
Hugo Vance: And the number that stopped me cold when I first encountered it — bid-ask spreads correlate with per-trade volatility at an R² exceeding 0.9. Across electronic markets. That is not a mild association. That is almost mechanical. The spread is a real-time meter of fear.
Lila Soto: R-squared over 0.9 — that's closer to a law than a correlation.
Hugo Vance: It suggests adverse selection — not inventory costs alone — is the primary driver of spread width. Inventory risk you can hedge. Ignorance, you cannot.
Lila Soto: And the Avellaneda-Stoikov model — that's the quantitative answer to 'okay, given both these pressures, how do you actually optimize the quotes you post?'
Hugo Vance: Yes — it gives the market maker a framework for adjusting bid and ask dynamically, accounting for both inventory drift and adverse selection simultaneously. It's the operational answer to Glosten-Milgrom's theoretical problem. But — and here's what makes this whole picture genuinely uncomfortable — liquidity is not a structural constant. It is a permission granted by those with sufficient capital and risk tolerance. When adverse selection rises, market makers pull quotes. The book thins. And that connects to something we haven't touched yet — why the transparency of the CLOB may actually make the information hierarchy invisible rather than collapsing it. That's where the real fragility sits.
Lila Soto: Transparency that hides the thing it's supposedly showing. Yeah — that's the one I want to pull apart next.
Hugo Vance: Transparency that hides — yes, but I want to press on what's underneath that, because the comfortable version of the story is just 'market makers widen spreads when they're scared.' Scared of what, exactly? That's the layer we haven't named. Information asymmetry is the engine. Some participants genuinely know more about an asset's true value than others do. And the spread is — well, it's the market maker's confession that they can't tell who's who.
Lila Soto: The confession built into the price.
Hugo Vance: Grossman and Stiglitz formalized this in 1980. Their argument: if prices ever fully reflected all information, no rational actor would pay to acquire information. But then no one acquires it, prices stop reflecting it, and suddenly there's value in acquiring it again. You can't escape the loop. The information hierarchy is self-sustaining — not a bug, not a market failure. A permanent structural feature.
Lila Soto: Wait — so an informationally perfect market would destroy itself.
Hugo Vance: Precisely. Which means informed traders and the information asymmetry they create are — they're load-bearing. Remove them and price discovery collapses. The market needs the predator to function.
Lila Soto: And order flow is how the predator leaks. Because — mm, they have to execute through the same Central Limit Order Book as everyone else. They can't place a trade in some private dimension. So their intentions show up in the sequence and size of what hits the book.
Hugo Vance: Before price fully moves. That's the counterintuitive core. Order flow is a leading signal.
Lila Soto: And this is where — okay, the thermodynamic framing actually gets rigorous, not just poetic. Touzo, Marsili, and Zagier connected the Glosten-Milgrom model directly to something called the Szilárd information engine from physics. The claim is that informed traders extract value from market makers the way a Szilárd engine extracts work from information. Bounded by how much they actually know. It's not a metaphor — it's a formal correspondence.
Hugo Vance: Mathematically austere. I grant that. But — and Ochędzan and Antulov-Fantulin's 2026 agent-based work makes this explicit — markets are adaptive social systems. Traders reprice their own strategies. The Szilárd engine is a closed physical system. Real markets are not. So the analogy has genuine limits, even as a formal structure.
Lila Soto: Hm — so the analogy holds at the level of the transaction but breaks at the level of the ecosystem.
Hugo Vance: Yes. Which actually sharpens the unsettling part. Electronic markets made the order book visible to everyone — NASDAQ, NYSE, the whole CLOB laid bare. And the assumption was: visibility equals fairness. But by the time an ordinary participant reads the order flow pattern and recognizes 'informed trading is happening here,' the informed trader has already repositioned. The edge belongs structurally to whoever is closest to the data, fastest to the signal.
Lila Soto: Transparency made the hierarchy invisible rather than flattening it.
Hugo Vance: The R² above 0.9 between spread and per-trade volatility — that number isn't just a correlation. It's evidence that the spread has always been pricing this specific, persistent information gap. We mechanized the problem. We did not solve it. We just made it look like physics.
Lila Soto: And GameStop is the proof. Not the metaphor — the actual proof. Glosten-Milgrom was published in 1985, and what it said was: when adverse selection risk peaks, market makers rationally reduce quotes or exit entirely. January 2021 — spreads widened on the most-traded names during that thousand-percent spike. Liquidity didn't fail. It did exactly what the model said it would do. It just did it to people who thought the market was a public utility.
Hugo Vance: Which is the thing I cannot stop turning over. Liquidity isn't structural. It's — I mean, it's a permission. Granted by whoever has the capital and the risk tolerance to keep quoting into chaos. When that permission is withdrawn — rationally, defensibly withdrawn — the book thins, spreads widen, and the people most desperate to trade pay the largest tax.
Lila Soto: A tax on the chaos they were supposed to be smoothing.
Hugo Vance: Yes. And that permission — it's concentrated. A small number of professional liquidity providers hold it. Which means the next stress event, whatever it is, runs through the same chokepoint.
Lila Soto: I don't know what to do with that, honestly. Like — I came into this wanting to understand the mechanism, and I think I do now. The Central Limit Order Book, the adverse selection problem, the spread as the confession. But understanding the mechanism makes the fragility feel more concrete, not less.
Hugo Vance: Yes. That's where I land too. We mechanized price discovery. We didn't make it robust. We just made it legible — to some.
Lila Soto: To some. That's the whole sentence right there.
Hugo Vance: Well. Good conversation. Genuinely.
