Toxic Flow: The Hidden Cost of Providing Liquidity

The compensation for doing this is the spread, the small gap between the buy price and the sell price, earned over and over again.

It’s a relatively simple business model at first glance, and a fragile one. It only works if most of the people trading are not trying to outsmart the price in front of them.

Most traders are what the industry calls "uninformed", which is an unflattering way of describing normal behaviour. They trade because they got a bonus, need cash, or because an influencer told them they could win big by buying the latest memecoin. 

Their trades aren’t predictions, and they don’t carry information about where prices are about to go. If trading mostly looked like that, market making would be dull and reliably profitable.

The problem is the other type of trader, the one who trades with liquidity providers (LPs), knowing that the price is about to move. Moments later, it does. The LP’s quote wasn’t unlucky. It was wrong.

That’s toxic flow.

Su Zhu, the co-founder of the now collapsed crypto hedge fund Three Arrows Capital, has compared this dynamic to the gambling industry. A casino happily accepts bets from millions of ordinary players, most of whom lose. The house edge works because the flow of bets is mostly uninformed. 

But a small number of skilled card counters can still threaten the casino, even though the odds are stacked against them. Casinos deal with this by identifying and removing those players. 

Market makers in traditional financial markets don’t have that option. They have to stand ready to trade with anyone who shows up.

What toxic flow really is

Toxic order flow isn’t "toxic" because it’s unethical (though it sometimes is). 

It’s toxic because it exploits information asymmetry. One side knows something the other side does not, and trades before prices adjust.

Dr Rand Low, a professor of quantitative finance at Bond University, told Sandmark that toxic order flow arises when traders have private information, or an informational advantage, sometimes illegal, over LPs. 

Market makers, who generally trade without that information, face what economists call adverse selection. They end up trading most often with the people who know the price they are quoting is already wrong.

The market maker thinks it’s offering a fair price. The counterparty knows it isn’t.

This dynamic has been studied for decades. Economists have long warned that markets struggle when one side consistently knows more than the other.

George Akerlof, who later shared the 2001 Nobel Prize in Economics, illustrated the problem of information asymmetry using the used car market. 

Sellers know which cars are good and which are bad. Buyers can’t tell. Buyers thus offer an average price. Sellers of good cars obviously refuse to sell at that price and leave the market. The market is then increasingly made up of bad cars, until little else remains. 

In crypto, the result is a market shaped by defensive pricing and dominated by traders who only engage when they have an edge.

How market makers get picked off

When order flow becomes toxic, market makers often provide liquidity at a loss without realizing it at the moment of execution. The response is straightforward: they either widen their spreads or stop quoting altogether.

Widening spreads is the quieter response. It raises the cost of trading for everyone, informed or not, because the market maker needs compensation for the risk of being wrong. Toxic flow thus becomes a cost that ordinary traders end up paying.

Pulling liquidity is the louder response. It happens when quoting prices no longer makes sense. Liquidity vanishes, execution worsens, and markets become brittle.

The simplest example of toxic flow is the stale quote. A high-frequency trader detects a price change on one exchange milliseconds before a market maker updates its quotes elsewhere. The trader trades against the outdated price. The market maker fills the trade and immediately loses money.

Repeated thousands of times, this becomes decisive. The market maker is no longer earning the spread, and value is systematically transferred to better-informed traders.

Zhu has argued that this intuition is backwards. Market makers are not businesses serving customers so much as standing offers to trade, while the sharpest traders hold the advantage. They are under no obligation to act. They can wait for the moment when prices are wrong, and trade only then.

How crypto makes this worse

Crypto sharpens the problem in two ways.

First, liquidity provision is mostly voluntary. In traditional equity markets, certain firms are formally designated as market makers and required to maintain two-sided quotes, even in volatile conditions. In crypto, firms quote when it is profitable and disappear when it is not.

Second, crypto markets are split across systems that update at different speeds. Prices on large centralized exchanges, or CEXs, often move first. Prices on decentralized exchanges, or DEXs, update mechanically and can lag behind.

Most toxic flow still comes down to information asymmetry, but it usually appears in two concrete forms.

• Speed - Some traders see the price move elsewhere first, and trade before slower venues or slower market makers update.
• Coverage - Some traders see more of the market. They observe multiple exchanges, OTC desks, correlated assets, liquidations, or large flows that have not yet shown up publicly.

In both cases, the trader is exploiting the fact that the price in front of them is wrong.

Zhu frames the asymmetry in practical terms. Takers, who trade against posted prices, can afford to be patient. They only trade when the odds are already in their favour. 

Market makers, by contrast, must keep their prices available to everyone, all the time. 

No spread is too wide if the taker has already priced it into their model.

DeFi and structural toxic flow

Automated market makers, or AMMs, such as Uniswap or PancakeSwap, price assets using a fixed, transparent, and predictable formula. There is no discretion and no judgment. The AMM can't recognize informed trading and adjust its price in response.

As a result, toxic flow is not an occasional problem for AMMs, but a built-in cost.

Low gives a simple example. When the market price of Ether moves sharply on a highly liquid centralized exchange, the AMM price does not update instantly.

• Ether trades at $2,000 on a centralized exchange.
• The AMM pool still prices it at $1,990.
• Liquidity providers sell below fair value, and the bot captures the difference.

Liquidity providers do not receive a bill for this. They simply own a worse pool.

This cost is known as loss-versus-rebalancing, or LVR. It is the value transferred to informed traders who rebalance the pool by trading against it as soon as the price becomes stale. Unlike impermanent loss, which can reverse, LVR is permanent.

Recent research on Uniswap v3 pools quantifies this effect. When you compare the fees liquidity providers earn with the value lost to arbitrage, the gap is persistent and, in many cases, growing.

Liquidity providers on major Uniswap v3 pools often lose more to arbitrage than they earn in fees. Left: cumulative difference between fees earned and arbitrage losses. Right: fees as a share of arbitrage losses, with 100% meaning fees fully offset losses.

Source: Canidio & Fritsch (2024)

Zhu draws a direct comparison to foreign exchange markets, where liquidity providers eventually learned to protect themselves by segmenting order flow and rejecting trades that looked informed. AMMs cannot do this, as their pricing is deterministic, and anyone can trade against them at any time.

Vertical integration

In crypto, toxic flows have not only persisted but also become organized.

Low describes the emergence of specialized actors, known as searchers, who monitor public mempools for profitable transactions. Evidence of how concentrated this activity has become comes from a recent study of Ethereum arbitrage by Wu et al. (2025), which finds that a small number of searcher entities dominate the landscape. 

In their data, Wintermute and two large entities labelled SCP and Kayle jointly accounted for roughly 90% of the value extracted in the first quarter of 2025.

To ensure their trades are executed, the most successful searchers increasingly integrate with block builders, the entities responsible for assembling blocks of transactions. This vertical integration concentrates power, giving a small number of players priority access to block space and allowing them to capture a disproportionate share of arbitrage profits.

What can be done

Every market tries to solve the same problem: how do you keep liquidity provision attractive when a subset of traders is systematically better informed?

The honest answer is that you can’t eliminate toxic flow. You manage it and design around it.

Low, Li, and Marsh (2016) have shown that changes in order flow often appear before volatility spikes and liquidity withdrawals. When informed trading intensifies, market conditions tend to deteriorate. Tracking these patterns can provide an early warning that risk is rising.

There are also design responses that try to reduce the value of being first.

• Dynamic, asymmetric fees that rise when trading appears informed, compensating liquidity providers when risk is highest.
• External price references or delayed settlement to reduce stale pricing.
• Batch auctions that execute trades at a single clearing price, reducing the advantage of tiny speed differences.
• Intent-based architectures and private order flow that keep transactions out of the public mempool.

The punchline

Toxic flow is what happens when the market maker ends up funding the market.

As Zhu puts it, liquidity will always be used to generate alpha for better-informed traders. The only open question is who pays for it.

It begins with a millisecond advantage on another venue and ends with wider spreads, poorer execution, and liquidity that disappears when it is needed most. In DeFi, the process is especially clear. Value is extracted onchain, block by block, as a direct consequence of how prices are set.

Market making is often framed as a public service, but in reality, it only works when most participants are not trying to beat the price they are offered.

Liquidity is easiest to provide in a market full of people who are simply trying to trade. 

When too many participants know more than the price implies, the shop raises its prices or shuts its doors.