> **Source:** https://permissionless.fi/en/19-market-making
> From *Permissionless Finance* (Permissionless Finance: From Perpetual Futures to the On-Chain Global Market) by Eric Cheung. Licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/).

# Chapter 19: Liquidity Provision and Strategic Games

Consider the operational practice of a global cryptocurrency market maker. (The opening scenario of this chapter is constructed from public market data and industry interviews; it is intended to illustrate the multidimensional pressure structure that perpetual-futures market makers face, and it does not refer to any specific institution. The concrete figures in the text—$50 million of net long inventory and a 150% annualized funding rate—are illustrative settings rather than measured statistics.) Within a single trading window, its trading system faces three types of structural pressure at once. On the centralized exchange (CEX) perpetual-futures order book, a sudden drop in ask-side depth triggers a toxic-order-flow alert, and distinguishing liquidation flow from informed flow becomes the key to staying profitable. At the funding-rate management level, the annualized BTC funding rate spikes to 150% (referencing the extreme rates during BTC's break to an all-time high in March 2024; here the annualization convention follows "annualized = per-settlement rate × settlements per day × 365," so 150% annualized corresponds to about 0.137% per settlement); combined with roughly $50 million of net long inventory, this generates a substantial holding cost in every settlement period, and unlike the overnight interest of traditional markets, this cost variable is highly random and directionally uncertain. In on-chain markets, the net asset value of the liquidity pool declines steadily in the absence of any obvious triggering event, revealing an additional risk dimension that market makers on decentralized exchanges (DEXs) face under maximal extractable value (MEV) attacks. This scenario reveals the core complexity of the perpetual-futures market-making business: its challenge arises not from a single dimension but from the simultaneous superposition of adverse-selection risk, inventory-cost risk, and institutional risk, and it takes fundamentally different forms across the two market architectures of CEX and DEX.

This is the typical operating environment confronting perpetual-futures market makers in 2026. Cross-platform arbitrage and the fragmented trading structure of crypto markets [1], together with the pricing mechanism of perpetual futures [2], jointly shape this environment: market makers must operate continuously for 72 hours across two core market architectures—the CEX central limit order book (CLOB) and the on-chain order book of DEXs. They must contend with "toxic" order flow from four distinct sources—informed trading, arbitrage extraction, liquidation cascades, and MEV attacks—while managing a risk inventory subject to periodic funding-rate shocks, all within a market that runs 24/7. Compared with market making in traditional financial markets, this difference in difficulty is not one of degree but one of dimension.

In Part Six of this book, we have repeatedly treated "liquidity" as a key background variable, using it to explain the boundaries of arbitrage opportunities and the causes of market anomalies—whether the "liquidity black hole" examined in Chapter 17 or the altcoin "efficiency wasteland" analyzed in Chapter 18. In those chapters, however, liquidity was a macro-level, given presence. This chapter shifts the analysis from the demand side of liquidity to the supply side and investigates a core question: where does liquidity come from? By whom, and at what cost and risk, is it "produced"? Why are market makers willing to bear the risks of adverse selection and inventory fluctuations on an ongoing basis? And what strategic framework have they developed within this game?

These questions are foundational. Understanding how liquidity is produced is the prerequisite for understanding why it fails, how it might be repaired, and why it vanishes precisely when it is most needed—and thus for designing more resilient liquidity-incentive mechanisms. Market making in perpetual futures rests on a multidimensional interplay of risk pricing, information games, and institutional constraints, and its cost structure differs fundamentally from that of traditional markets.

## 19.1 The economics of liquidity provision

In any active financial market, traders expect to be able to buy or sell an asset at a "reasonable" price at any time. Behind this expectation lies a market quality called "liquidity," which directly determines whether traders can complete their transactions at a reasonable cost. When liquidity is abundant, the market runs smoothly, and traders can easily execute their intentions without exerting excessive impact on prices; when liquidity dries up, the market becomes fragile and sluggish, and even small trades can trigger sharp price swings. Yet liquidity does not arise from nothing. It is not a natural state but a product that is carefully "manufactured." The professional participants who manufacture this product are market makers.

The market maker's role is often misunderstood. In the narratives of some market participants, they are mysterious "manipulators" with an information advantage who profit from the losses of ordinary traders. In other, overly simplified academic models, they are portrayed as "automata" that passively respond to order flow and mechanically adjust quotes. Both portrayals depart from reality. The essence of market making is neither a predation based on information asymmetry nor a passive public service. It is a rigorous, high-risk business whose core lies in the trade-off between risk and return. Market makers are willing to act as counterparty on an ongoing basis because they expect to earn a statistically positive spread return by bearing specific, manageable risks. Their presence allows those liquidity demanders who are unwilling to wait and want to trade immediately to externalize their "cost of waiting" and their "cost of finding a counterparty."

This section analyzes the economics of the market-making business and answers a central question: why is anyone willing to bear risk in order to supply liquidity to the market continuously? The analysis begins with the revenue and cost structure of the business and then examines two classic risks: the adverse-selection risk that stems from information asymmetry and the inventory risk that stems from price fluctuations. These two risks form the general theoretical pillars for understanding all market making.

Yet the perpetual future, as a distinctive financial instrument—with its lack of an expiration date, high leverage, funding rate, and forced-liquidation mechanisms—adds a new dimension to the market maker's traditional risk profile. One of this chapter's integrative frameworks addresses exactly this: it consolidates and names the "third pillar" of perpetual-futures market making, institutional cost. We analyze in detail how the randomness of the funding rate, the punitive nature of the liquidation mechanism, the lag of the mark price, and the operational pressure of 24/7 continuous trading together constitute a set of structural costs that are endogenous to the product's design and difficult to hedge away completely. Finally, we integrate these three pillars into a unified profit equation, thereby revealing the market maker's "survival condition." This equation not only explains how market makers profit but, more importantly, foreshadows the conditions under which market makers will choose to contract or even abandon market making, giving rise to the "liquidity black hole" phenomenon we will examine in later chapters. Understanding this profit equation is the key to understanding liquidity-provision behavior and, indeed, the stability of the entire market microstructure.

### 19.1.1 The nature of market making

The market maker's core business model can be distilled as follows: by continuously posting competitive bid and ask prices to the market, it profits from the spread between the two. Its function resembles that of an intermediary in financial markets—sourcing assets from deeper liquidity sources and offering them, at a price that includes a spread, to traders who need immediate execution. This seemingly simple model in fact performs several essential market functions.

First, the market maker supplies immediacy. In a purely order-driven market without market makers, a trader wishing to buy immediately must wait for a counterparty willing to sell at an acceptable price to appear. This wait can be long, and its outcome is uncertain. By committing to be ready to trade at any time, the market maker greatly shortens this matching time, connecting an asynchronous market into a seemingly synchronous whole. It links buyers and sellers who arrive at different moments along the dimension of time.

Second, the market maker reduces transaction costs. Competition among market makers is expressed primarily in who can offer a narrower bid-ask spread and greater quoting depth. A market with a narrower bid-ask spread means lower costs for liquidity demanders seeking immediate execution. A market with greater depth means that large trades can be executed with smaller price impact. The existence of and competition among market makers therefore directly determine the trading efficiency and capacity of the market.

Third, the market maker participates in and contributes to price discovery. The prices it posts are not set arbitrarily but reflect an integrated judgment of current market information, order-flow pressure, its own inventory level, and expectations of future volatility. The dense queue of quotes on an order book itself embodies the market's collective assessment of an asset's fair value. When new information enters the market, market makers quickly adjust their quotes, thereby guiding the price toward a new equilibrium. Their quoting behavior becomes an important link in the price-discovery process.

Understanding the market-making business hinges on distinguishing the source of its profit. Unlike position traders who depend on predicting an asset's long-term price direction, a pure market maker aims to profit from the turnover of two-way trading rather than from the price movement of a one-sided position. Its basic profit formula can be expressed as:

$$\text{Profit} \approx (\text{average bid-ask spread} \times \text{total volume}) - \text{transaction costs}$$

Here, "transaction costs" are not simply fees but a complex combination that mainly comprises two core risks: the loss from adverse selection by better-informed traders, and the risk of exposure to price fluctuations from holding inventory. A market maker's strategy, algorithms, and technology investments all revolve around how to price these two risks precisely and internalize their cost into the bid-ask spread. The core competitive advantage of a successful market maker lies not in predicting market moves more accurately but in measuring and managing risk more precisely than its competitors, so that it can offer better quotes at a given level of risk and achieve a statistically positive expected return across a large number of trades. From an economic standpoint, a market maker's profit is essentially the risk premium the market pays for two scarce services: immediacy and price certainty.

### 19.1.2 Adverse selection and inventory management

The market maker's profit model is, fundamentally, a search for certainty in a game against uncertainty. After decades of research, the academic literature has traced the core uncertainty faced by market makers to two sources, which constitute the twin pillars of modern market-making theory: adverse selection and inventory management. Understanding these two pillars is the basis for understanding how market makers price, how they manage risk, and why they sometimes choose to exit the market.

The first pillar, adverse selection, arises from the structural information asymmetry in the market [3]: the order flow a market maker faces is a mixture of two types of traders—"noise traders," whose behavior is random and driven by allocation or hedging needs, and "informed traders," who possess undisclosed information about an asset's future price. In nearly every trade with an informed trader, the market maker stands on the wrong side and bears a nearly certain loss—the adverse-selection cost. Its response is to spread this expected loss across every trade through a wider bid-ask spread, so that the width of the spread reflects the market maker's judgment about how active informed trading is in the market. A full characterization of this risk—the classic models of Glosten and Milgrom (1985) [4] and Kyle (1985) [5]—is developed in Sections 19.2 and 19.2.1.

The second pillar, inventory management, arises from the fact that a market maker inevitably accumulates net inventory through passive fills: buy and sell orders are hard to match perfectly over the short term, so inventory departs from neutral and forms a net long or net short position, turning the market maker partly from a pure spread trader into a position holder exposed to price fluctuations. The core response tool is "quote skewing"—dynamically adjusting quotes on the bid and ask sides according to inventory direction so as to steer order flow and accelerate the return of inventory to neutral. In essence, this is a trade-off between "closing the position quickly to reduce risk" and "waiting for a better price to profit"; the longer the holding period, the greater the risk, so market makers generally favor high turnover. The classic modeling of this mechanism—the inventory models of Ho and Stoll (1981) [6] and Amihud and Mendelson (1980) [7]—is developed in Sections 19.3 and 19.3.1.

Together, the two risks of adverse selection and inventory management constitute the core of the market maker's cost structure. The market maker's quoting decision is, in essence, the solution to an optimal spread and depth subject to the constraints of these two risks. The spread must be wide enough to compensate for the expected adverse-selection loss and the risk premium of holding inventory, yet narrow enough to attract sufficient order flow in competition. This is a dynamic optimization process that continually seeks a moving balance between risk and return, safety and growth.

### 19.1.3 Institutional cost

The classic twin-pillar theory provides a solid foundation for understanding market-making behavior in all markets. Yet when we turn to the perpetual future—a financial instrument native to the crypto world—we find that adverse selection and inventory risk alone are no longer sufficient to explain market-maker behavior. The distinctive institutional design of perpetual futures—including the funding rate, leverage and liquidation, the mark-price mechanism, the absence of an expiration date, and 24/7 uninterrupted trading—together gives rise to a third systematic source of cost, which we term "institutional cost." This consolidated naming reveals why market making in perpetual futures differs in kind from market making in traditional equities or futures.

Institutional cost arises not from information asymmetry or price fluctuations but is embedded in the "rules of the game" of the perpetual future as a product. As long as one participates in this market, one must bear these costs. They comprise the following main components.

The first is funding-rate exposure. The funding rate is the core mechanism by which perpetual futures anchor to the spot price [8]; it can be positive or negative and fluctuates sharply. Two related but distinct concepts must be distinguished: the spot-futures basis between the perpetual future and the spot (referred to in the academic literature as "crypto carry"), and the funding rate itself. Schmeling et al. (2023) show that crypto carry sometimes exceeds 40% annualized, far above the level of traditional markets, largely owing to the structural friction between leveraged retail demand and scarce arbitrage capital [9]. The basis can be captured by arbitrageurs through a cash-and-carry strategy, whereas the funding-rate exposure a market maker faces is of a different nature: it is the funding payment or receipt in which the market maker is forced to participate every settlement period after accumulating inventory through passive fills. The funding rate fluctuates far more than traditional funding costs: during BTC's break to an all-time high in March 2024, a single 8-hour settlement rate once exceeded 0.1% (using the convention "annualized = per-settlement rate × settlements per day × 365," a single rate of 0.1% per 8 hours corresponds to about 109.5% annualized, while an annualized figure of over 400% would correspond to about 0.365% per settlement), which means that a market maker holding $10 million of net long inventory would pay more than $10,000 in a single settlement period. More critically, the direction and magnitude of the funding rate are highly random and can flip from an extreme positive value to a negative one within hours. This stands in sharp contrast to the relatively stable, predictable funding costs (such as the overnight rate) in traditional futures market making. A market maker must manage not only the risk of price fluctuations but also the risk of funding-rate fluctuations—a risk dimension unique to perpetual futures.

The second is the liquidation risk premium. Because of the high leverage of perpetual futures, the forced-liquidation mechanism is triggered frequently during sharp market swings. A market maker's own inventory also faces the risk of forced liquidation. Forced liquidation differs fundamentally from a market maker's active inventory management: it occurs at the moment when market liquidity is worst and prices are most unfavorable, a punitive, passive purging of risk. To avoid this, a market maker must maintain a higher margin level than theory would require, or be forced to close positions actively at very poor prices as it approaches the liquidation threshold. The additional cost incurred, or potential return forgone, in order to avoid forced liquidation constitutes the liquidation risk premium.

The third is mark-price deviation risk. Margin and profit-and-loss (P&L) calculations for perpetual futures are based on a smoothed mark price (typically a moving average of a multi-exchange index price plus the basis). The primary design purpose of this smoothing mechanism is to prevent erroneous liquidations caused by short-term price manipulation; in this sense it is protective. However, in a sustained one-directional price move, the mark price systematically lags the true, latest market price, and this lag produces two risks of opposite direction but equal harm. First, liquidation-delay risk: when the price moves continuously in one direction, the lag in the mark price means that positions that should have triggered liquidation earlier are "tolerated" for longer, so that the eventual liquidation loss is far greater than it would have been under timely liquidation; this additional loss must be borne by the insurance fund or through a socialized-loss mechanism, indirectly affecting all market participants, including market makers. Second, hedge-basis risk: when a market maker hedges its perpetual-futures position, the price of the hedging instrument (such as spot) moves in real time with the latest traded price, whereas the P&L calculation of the perpetual future is anchored to the lagging mark price. In fast markets, the deviation between the two can widen significantly, so that a market maker's hedge appears balanced in nominal terms while the actual P&L shows unexpected deviations caused by the mark-price lag.

In addition, the indefinite exposure created by the lack of an expiration date increases the difficulty of inventory management: perpetual futures lack the time anchor of traditional futures' "natural settlement at expiration," so an imbalanced inventory can in theory be held indefinitely, remaining continuously exposed to price and funding-rate risk, forcing market makers to rely more heavily on costlier active hedging (the full mechanism of its missing "center of gravity" is discussed in Section 19.3.2).

Finally, 24/7 uninterrupted trading brings enormous operational costs and risk-monitoring pressure. Traditional markets have closing and holiday hours, which give market makers necessary downtime for system maintenance, model calibration, risk review, and strategy adjustment. In the perpetual-futures market, risk is ever-present. Market makers must invest heavily in building trading systems, risk-control systems, and monitoring teams capable of running stably around the clock, so as to respond to any market disruption or technical failure that may occur at any moment. This continuous operational pressure and labor cost constitute another important component of institutional cost.

Integrating the five subcomponents above, we arrive at a more complete cost structure for a perpetual-futures market maker. Institutional cost is a composite category, encompassing all costs induced by the institutional design of the perpetual future itself—from funding-rate exposure, the liquidation risk premium, and mark-price deviation risk to the indefinite exposure of no expiration and the pressure of 24/7 uninterrupted operation. Among these, the system-building, staffing, and continuous-monitoring expenditures brought by 24/7 operation, though "operational" in form, are rooted in the institutional design of perpetual futures' uninterrupted trading rather than in the expansion of a market maker's own business scale; they are therefore classified under institutional cost rather than as a separate operational-cost category in this chapter's analytical framework. Compared with spot or delivery-futures market making, a perpetual-futures market maker must not only face more severe adverse-selection and inventory risks (amplified by leverage and high volatility) but also bear this set of institutional costs, brought by the product's institutional design itself, that cannot be fully eliminated through traditional risk-management tools. This explains why liquidity-provision behavior in perpetual-futures markets exhibits characteristics markedly different from those of traditional markets.

To give the concept of institutional cost a structured analytical expression, we can decompose it into the following four subcomponents:

$$C_{\text{inst}} = \mathbb{E}\!\left[\lvert FR_t\rvert \cdot \lvert Q_t\rvert\right] + P(\text{liq}) \cdot L_{\text{penalty}} + f(\Delta_{\text{mark}}) + C_{\text{ops}}$$

Here, the first term is the expected cost of funding-rate exposure, where $FR_t$ is the funding rate at time $t$ and $Q_t$ is the size of the market maker's net inventory; the interaction of the two determines the funding payment or receipt in each settlement period, and taking the expectation of the absolute values reflects the risk exposure under directional uncertainty of the rate. The second term is the liquidation risk premium, where $P(\text{liq})$ is the probability of triggering forced liquidation at a given inventory and leverage level, and $L_{\text{penalty}}$ is the additional loss from forced liquidation relative to active position closing. The third term is a function of mark-price deviation risk, where $\Delta_{\text{mark}}$ is the magnitude of the deviation between the mark price and the latest traded price; this function grows nonlinearly when prices move rapidly in one direction. The fourth term, $C_{\text{ops}}$, is the fixed and semi-fixed cost of 24/7 uninterrupted operation. This expression is a conceptual classification framework, not a directly calibratable quantitative model: the parameters of each subterm (such as the computation of $P(\text{liq})$) depend heavily on a specific exchange's liquidation-engine rules, margin system, and microstructural state, and in practice are usually estimated by Monte Carlo simulation rather than by an analytical expression. Its value lies not in providing precise numbers but in revealing the multiple sources of institutional cost and the structural relationships among its subcomponents; in particular, the randomness and state-dependence of the first three terms mean that institutional cost can expand sharply under extreme market conditions.

### 19.1.4 The profit equation and survival conditions

Combining the above analysis, we can construct a more comprehensive profit equation for the market maker that systematically characterizes its sources of revenue and its cost components. This equation is not merely an accounting identity but an analytical framework that guides a market maker's day-to-day decisions and strategic choices.

$$\text{Total profit} = \text{spread revenue} - (\text{adverse-selection cost} + \text{inventory-risk cost} + \text{institutional cost})$$

Here, "spread revenue" is a broad concept in this chapter's framework: it includes not only the profit from the market-making spread itself but also the fee-rebate revenue that exchanges pay to market makers (detailed in Section 19.5.4) and the funding-rate revenue that market makers *collect* by strategically managing their inventory direction (Section 19.3.3 will analyze the dual nature of the funding rate as both cost and revenue). The accounting attribution of the funding rate must be clearly distinguished: the funding revenue a market maker strategically *collects* at the margin through quote skewing is booked under the broad "spread revenue," whereas the funding payment it is forced to *bear* after accumulating inventory through passive fills is booked under "institutional cost." The two are in fact manifestations of the same variable in opposite directions (single-period funding P&L $= -FR_t \cdot Q_t$, with a positive value indicating net revenue to the market maker): when the inventory direction is opposite to the funding direction, it is positive revenue; when it is in the same direction, it is a cost. On mainstream pairs, rebate revenue can account for a significant proportion of a market maker's total revenue. The three costs in parentheses—corresponding to information risk, position risk, and product-institutional risk, respectively—constitute the cost threshold that a market maker must clear. This equation reveals the market maker's "survival condition": the revenue it captures through spreads must, in a statistical sense, be sufficient to cover all the risk costs it faces. When this inequality cannot be satisfied, a rational market maker has only three options: widen the bid-ask spread to increase revenue per trade, reduce quoting depth to lower risk exposure, or, in the most extreme case, withdraw from the market entirely and stop quoting.

The shifting balance of this profit equation explains the dynamic changes in market liquidity. In an environment that is calm, information-transparent, and highly competitive, a market maker's various costs are low, and it has both the ability and the willingness to offer narrow spreads and deep liquidity, so that the market appears efficient and stable. However, when the market enters a period of sharp volatility, when major news breaks frequently, or when the institutional mechanisms of perpetual futures (such as the funding rate or liquidation pressure) become extreme, a market maker's costs surge. Adverse-selection risk rises as information asymmetry intensifies, inventory risk increases as price volatility is amplified, and institutional cost becomes salient as funding rates turn extreme and liquidation risk looms. At this point, to maintain its survival condition, market makers are forced to collectively and defensively widen spreads and withdraw depth. This is precisely the microeconomic root of the "sharp liquidity contraction" phenomenon we will examine in depth in later chapters.

Recent empirical research further corroborates the theoretical analysis above. Using the natural experiment of Binance's July 2022 removal of maker-taker fees on BTC pairs, Galati (2024) finds that zero fees actually led market makers to widen bid-ask spreads and reduce quoting depth, raising total market transaction costs [10]. The mechanism is that, once the maker rebate was removed, market makers lost the explicit incentive to maintain narrow spreads and instead compensated with wider spreads, so that the implicit spread cost exceeded the explicit fees that had been eliminated. Easley et al. (2024), using machine-learning methods to analyze microstructure data for several mainstream crypto assets on a major crypto exchange, demonstrate that microstructure indicators such as liquidity and price discovery have significant predictive power for crypto-asset price dynamics [11]. These studies show that the impact of market-maker behavior on market quality is far more complex than a simple fee incentive would suggest.

A market maker's survival condition varies markedly across asset types. For highly liquid, high-market-cap assets such as Bitcoin or Ethereum, market depth is sufficient to absorb large volumes of trading, and information is relatively dispersed and transparent. Although informed trading still exists, its share of total order flow is low. A market maker can therefore be profitable at a relatively narrow spread through a model that combines lower profit margins with high volume. By contrast, for the large universe of "altcoin" perpetual futures, markets are typically thin in liquidity, highly concentrated in information (with project teams or whales holding an absolute information advantage), and more volatile in price. This means market makers face far higher adverse-selection and inventory-risk costs in these markets. To survive, they must set very wide bid-ask spreads. This in turn suppresses trading activity, forming a negative feedback loop of insufficient liquidity. This explains the microeconomic mechanism behind the "altcoin efficiency wasteland" phenomenon observed in Chapter 18.

Using a real-time order-book snapshot of ten major perpetual-futures pairs on Binance, the following figure illustrates the negative correlation between market depth and the bid-ask spread: BTC has the greatest depth and the lowest spread, ETH follows, and for small- and mid-cap assets such as ADA and DOT, depth drops sharply while spreads surge (see the figure for specific values). Over this sample period, liquidity is highly uneven across assets, consistent with the direction of the theoretical analysis above.

![Figure 19-1](./images/fig-19-1-en.png)

**Figure 19-1.** Comparison of market depth and bid-ask spread across perpetual-futures pairs (Data source: real-time order-book data for Binance USDT-M perpetual futures, March 2025)

Ultimately, we can understand a market maker's profit as the fee the market pays to obtain liquidity as a "public good." Liquidity is not free precisely because it is "produced" by professional risk-bearers who commit capital, technology, and labor and who bear the multiple risks of adverse selection, inventory, and institutional (including operational) cost. A market maker's profit is the reasonable compensation for this production process. Only by understanding this can we appreciate why liquidity is valuable and why, under specific conditions, it becomes fragile and scarce.

## 19.2 Adverse selection

In the market maker's profit equation, inventory risk is a persistent risk exposure, whereas adverse selection is an episodic risk factor with pronounced impact. It occupies a central place in the market maker's cost structure. The essence of adverse selection stems from the ever-present information asymmetry in markets. In an ideal world where all participants held exactly the same information, a market maker's risk would be confined solely to the price fluctuations of its inventory. In the real world, however, there is always a subset of traders (whom we call "informed traders") who, sooner or later and to a greater or lesser degree, possess nonpublic information about the future direction of an asset's price. When these informed traders enter the market, they do so not to satisfy a liquidity need but to profit from their information advantage. And the market maker, as the liquidity supplier that always stands ready to quote, inevitably becomes the primary counterparty through which informed traders monetize the value of their information.

This game is asymmetric from the outset: the market maker cannot tell before a trade whether its counterparty is an informed trader or a noise trader, yet it always ends up on the wrong side in trades with the former—the price it pays for information asymmetry is the adverse-selection cost. How great the adverse-selection risk of a market is therefore directly determines the "floor" of its liquidity cost; the bid-ask spread is, in a sense, the insurance premium the market pays for the expensive commodity of "information."

In the distinctive ecosystem of perpetual futures, this classic problem takes on new dimensions and greater intensity. The classic models used to analyze adverse selection in traditional markets, such as the Glosten-Milgrom model, retain their core insight, but the model's parameters and dynamics are greatly amplified by the institutional features of perpetual futures (especially leverage, 24/7 uninterrupted trading, and the distinctive on-chain information environment). Market makers must contend not only with informed traders from traditional channels but also with new sources of information advantage arising from the on-chain world. Their risk-monitoring systems must be more sensitive—able not only to detect the occurrence of a risk event but also to distinguish its type. This section analyzes in depth the particularities of adverse selection in perpetual-futures markets: it begins by reinterpreting the classical theory, identifies the sources of informed trading, analyzes its amplifiers, and finally examines the quantitative tools market makers use to protect themselves, revealing the full picture of this ongoing game.

### 19.2.1 The applicability of the Glosten-Milgrom model

One of the cornerstones of market-microstructure theory is the information-asymmetry model proposed by Glosten and Milgrom (1985) [4]. Although this model was born in the pre-digital-currency era, its profound insight remains the starting point for understanding market-maker behavior today. The core intuition of the Glosten-Milgrom model can be depicted as a simple probability game. When setting bid and ask quotes, a market maker faces two types of counterparties: noise traders, who trade for various reasons unrelated to the asset's fundamentals, such as liquidity needs, portfolio rebalancing, or irrational market sentiment; and informed traders, who possess private information about the asset's future value.

The market maker's dilemma is that it cannot tell a counterparty's true identity before a trade occurs. It can only estimate, based on historical experience, the probability that the next order comes from an informed trader (usually denoted by the Greek letter $\mu$). When a buy order arrives, the market maker knows there is a probability $\mu$ that it comes from an informed trader who knows the asset is about to appreciate, and a probability $1-\mu$ that it comes from a random noise trader. If the trade goes through and the counterparty happens to be informed, then the price of the asset the market maker sold will rise in the future, and it will suffer a loss. If, on the other hand, the counterparty is a noise trader, the future direction of the price is random, and the market maker earns the bid-ask spread securely. To avoid going bankrupt in a repeated game against informed traders, a rational market maker must set a bid-ask spread wide enough that the profit earned from noise traders can offset the expected loss suffered on informed traders. The size of the spread therefore directly reflects the market maker's judgment about the degree of information asymmetry in the market: the higher $\mu$, the wider the spread.

Applying this classic framework to perpetual-futures markets, we find that its basic logic still holds, but the model's key variables are given new, more violent dynamics by the institutional design of perpetual futures. First, the leverage mechanism becomes a powerful amplifier of adverse selection. In the original Glosten-Milgrom model, the impact an informed trader can inflict using an information advantage is limited by the size of its own capital. But in the perpetual-futures market, an informed trader with $1 million of capital can easily open a position at 10x or even 100x leverage, instantly scaling its notional principal to $10 million or even $100 million. This means the same piece of information can exert an impact on the order book several times greater in the perpetual-futures market than in the spot market. The market maker no longer faces a counterparty of limited capital but a trader whose impact, magnified by leverage, is several times its original capital. This forces the market maker to treat the leverage level as a core variable in its assessment of adverse-selection cost, systematically setting wider spreads in high-leverage markets or on high-leverage assets.

Second, 24/7 uninterrupted trading fundamentally changes the pattern of information arrival and release. In traditional equity markets, information typically accumulates after the close and is released collectively at the next day's open. Market makers (or specialists) can make a preliminary judgment about the degree of information asymmetry during the opening call auction by observing order flow, and adjust the opening spread accordingly. In the perpetual-futures market, however, trading runs without interruption, and information can pour in at any moment from anywhere in the world. Whether U.S. regulatory news, project developments in Asia, or macroeconomic data from Europe, any of these can be instantly translated into trading instructions that hit the order book. Market makers lose the important assessment-and-adjustment window of traditional markets; they must remain vigilant at all times, and their risk-management systems must be automated and run around the clock. This continuous state of alertness is itself a high operational cost and risk premium.

Finally, some mechanisms unique to perpetual futures, such as funding-rate settlement, also create "micro-hotspots" of information trading. The funding rate settles every 8 hours (or more frequently), and around the settlement point, market participants trade based on their expectations of the future rate, which itself can attract arbitrageurs and informed traders. For example, if the market broadly expects the funding rate to flip from positive to negative, traders holding long positions have an incentive to close before settlement, while informed traders who expect the price to come under pressure as a result may preemptively build short positions. These games around funding-rate settlement make the periods before and after the settlement point a hotbed of information trading, and the adverse-selection risk a market maker faces during these periods spikes periodically. This mechanism makes information-risk concentration periods—analogous to the "open" and "close" of traditional markets—recur every 8 hours.

The reinterpretation of the Glosten-Milgrom model in perpetual futures is therefore no longer a static probability calculation but a dynamic, multivariate risk-assessment process. A market maker's quoting decision must reflect, in real time and in an integrated way, factors such as the leverage level, global information flow, and the funding-rate cycle, which makes the asymmetric game in perpetual-futures markets significantly more complex than in traditional markets.

### 19.2.2 Sources of informed trading in perpetual futures

For a perpetual-futures market maker, the core problem is to identify the order-flow attributes of every fill. Identifying the intent behind order flow—distinguishing "toxic" informed trading from "harmless" noise trading—is the key to its survival. Compared with the relatively mature and well-regulated traditional financial markets, the spectrum of "informed traders" in perpetual-futures markets is broader and more complex, and their information sources are more diverse. Classifying these informed traders helps us map the threats a market maker faces.

The first type, and the most traditional, is the macro- or event-driven informed trader. These traders possess nonpublic information about to have a major impact on the market. Such information may come from the regulatory level, such as inside information that the U.S. Securities and Exchange Commission (SEC) is about to approve or reject a crypto exchange-traded fund (ETF); it may also come from project fundamentals, such as a core developer of a major public chain discovering a serious vulnerability, or a DeFi protocol about to announce a major partnership. In the brief window before this information becomes public, informed traders use the high leverage of perpetual futures to build large positions quickly. Their trading is decisive and firm, typically taking the form of continuous, large market orders aimed at completing the position build as fast as possible rather than quibbling over small slippage. For a market maker, this type of order flow is the most toxic, because it stems directly from a certain future change in price. Once it trades with such a trader, the market maker is almost destined to be on the wrong side.

The second type is the cross-market information-relay trader. The "information" advantage of these participants stems not from inside information but from speed. The crypto market is a global, fragmented market, and the same asset may trade simultaneously on hundreds of spot exchanges, futures exchanges, and ETF markets. Because of information-transmission delays and microstructural differences among markets, price changes do not occur synchronously across all markets. For example, when the BTC spot price on Coinbase suddenly jumps because of a large institutional buy order, a high-frequency trading firm can, using its low-latency servers and optimized trading algorithms, capture this signal within milliseconds and immediately build a long position in the BTC perpetual-futures market on Binance. For market makers on Binance, this order flow is likewise "informed," because it foreshadows a follow-on rise in price. These traders act as "information relays" across markets, earning the profit from differences in the speed of price discovery among markets, while market makers pay the adverse-selection cost for that speed difference.

The third type, unique to the crypto world, is the on-chain-information-parsing trader. The transparency of the blockchain means that a large amount of valuable information is publicly recorded, but interpreting this information requires expertise and powerful data-analysis capabilities. These informed traders predict future price movements by monitoring on-chain activity. For example, they can track large transfers by "whale" addresses; when a long-dormant whale address suddenly transfers a large quantity of tokens to an exchange, this may foreshadow imminent selling pressure. They can also analyze the liquidation queues of DeFi protocols; when the number of under-collateralized positions surges, this signals that a wave of liquidations may be imminent, leading to a sharp price drop. In addition, analyzing changes in a protocol's total value locked (TVL), the voting trends of governance tokens, and even the activity patterns of MEV bots can all provide clues for price prediction. Once effectively parsed, these on-chain signals constitute a powerful information advantage, allowing such traders to conduct adverse selection against market makers with weaker on-chain-data-analysis capabilities in the CEX perpetual-futures market.

The fourth type is the whale or large holder with the intent to manipulate the market. Unlike the first three types, who passively exploit information, these participants actively create information. Using their ample capital, they artificially manufacture sharp price swings—through continuous large market orders during relatively thin trading periods (such as the early hours of a weekend) or on small-cap assets such as altcoins. The purpose may be to dump spot holdings on retail traders after pushing the price up, or to trigger other traders' stop-losses or liquidations and profit amid the chaos. Although this behavior constitutes market manipulation in many jurisdictions, it remains common in the still-underregulated crypto market. For a market maker, this type of order flow is extremely hard to distinguish from genuinely fundamentals-driven informed trading, yet its risk is equally large. A market maker's quoting depth is easily punched through under such large impacts, leading to a significant accumulation of inventory risk.

The four sources of informed trading above differ markedly in market share, type of information advantage, and dependence on speed. Macro-event-driven informed traders are the fewest in number but have the greatest single-shot impact; their information advantage stems from nonpublic fundamental information, their demand for speed is relatively low, and the information window may last for hours. Cross-market information-relay traders are more numerous and highly dependent on speed; their profit depends entirely on a millisecond-level latency advantage, but the information content of each trade is low. On-chain-information-parsing traders are growing rapidly in number as on-chain analytics tools become widespread; their information advantage stems from the specialized interpretation of public data and lies between the first two. Manipulative large holders are the fewest but the hardest to predict, because they not only exploit information but actively create it. Together, these four sources constitute the multi-source threat network a market maker faces.

Understanding these diverse sources of informed trading directly shapes the design of a market maker's defensive strategy. Adverse selection is not a single problem but a composite of risks of different natures. A market maker's countermeasures must therefore be multilayered and differentiated. Countering macro-event-driven traders may require more macro-level management of risk exposure; contending with cross-market information relays is an ongoing competition over technology and speed; and defending against on-chain-information parsers requires the market maker itself to possess powerful on-chain-data-analysis capabilities. In this ongoing game, a market maker's survival strategy lies in continuously optimizing its risk-monitoring system so as to identify the true state of the market faster and more accurately than its opponents.

### 19.2.3 Amplification mechanisms of adverse selection

If the diverse sources of informed trading lay a complex web of threats before a market maker, then several core institutional features of perpetual futures act as powerful amplifiers, significantly heightening the impact of each risk in that web. Leverage, 24/7 uninterrupted trading, and structural information asymmetry together push the adverse-selection problem of perpetual-futures markets to a sharper edge than in any traditional financial market.

Leverage is the first and most direct amplifier. It fundamentally alters the cost-benefit structure of informed trading. An informed trader with a $1 million information advantage, in an unleveraged spot market, has both potential profit and market impact capped at its principal. But in a perpetual-futures market offering 10x leverage, it can use the same $1 million in margin to control a position with $10 million in notional value. This means its ability to attack the market maker's quotes using its information advantage is amplified tenfold. The market maker's bid and ask quotes, originally set to withstand million-dollar-scale impacts, must now face ten-million-dollar-scale impacts. This disproportionate risk exposure forces market makers to treat the leverage level as a key parameter in their pricing models. As a result, all else equal, the bid-ask spread of perpetual futures (measured in basis points) is almost always systematically wider than that of the spot market for the same asset. This extra spread is precisely the risk premium the market pays for leverage's "double-edged" effect.

This amplification is not linear but exhibits superlinear (convex) growth. As the market's average effective leverage rises from low (1–5x) to medium (5–10x) and then to high (10–20x and above), the implicit cost markup a market maker requires to compensate for the amplified impact of informed trading rises faster than the linear increase in the leverage multiple itself. This convexity arises from a key dual mechanism: higher leverage not only amplifies the notional impact of a single informed trader (a 10x-leveraged informed trader's impact on the order book is 10 times its principal) but also lowers the capital threshold for informed traders to enter the market, thereby attracting more potential informed traders. The market maker faces not only a greater impact per informed order but also a rise in the probability of informed trading, $\mu$, itself. The amplifying effect of leverage therefore has a "both-rise" character: each informed trader's impact is larger, and the total number of informed traders is greater. This structurally explains why high-leverage perpetual-futures markets have higher transaction costs.

Second, 24/7 uninterrupted trading, while convenient for users, also creates structural windows of fragility for market makers—the so-called "trough periods." The intensity of activity in global financial markets follows an intraday pattern dominated in turn by the Asian, European, and North American sessions. In the gaps where trading activity in different time zones hands off—for example, the deep night of the Asian time zone and the early morning of the North American time zone—overall market volume drops significantly, and the number of market makers participating in quoting also falls. This causes order-book depth and liquidity to decline sharply, forming "liquidity troughs." These trough periods become ideal operating windows for informed traders and market manipulators. The same large order that might have only limited impact during peak trading hours can, during a liquidity trough, exhaust one side of the order book's depth and cause a sharp price impact. The adverse-selection risk a market maker faces during these periods is sharply amplified. In response, a rational market maker systematically widens spreads, reduces quoting depth, and even withdraws from the market entirely during trough periods to avoid risk. This behavioral pattern ultimately manifests as a clear time-zone-based intraday pattern in the bid-ask spread, which peaks during liquidity troughs.

The following figure simulates the typical variation of the BTC perpetual-futures bid-ask spread over a 24-hour cycle: the spread is widest in the early Asian morning—the period of thinnest global liquidity—and narrowest when the European and North American sessions overlap and quoting competition is most intense, so the entire curve exhibits a single-peak, single-trough shape. This corroborates the earlier mechanistic account that "liquidity falls during trough periods and spreads widen accordingly," and provides an intuitive illustration that "24/7 trading increases adverse-selection risk."

![Figure 19-2](./images/fig-19-2-en.png)

**Figure 19-2.** Intraday variation pattern of the BTC perpetual-futures bid-ask spread (Data source: Kaiko Research, Binance BTCUSDT perpetual-futures order-book statistics, December 2024–March 2025; this figure shows an illustrative intraday pattern, and the plotted curve is a representative shape rather than an hour-by-hour empirical sample)

Finally, the very position of the perpetual-futures market within the crypto ecosystem also creates a form of structural information asymmetry. As argued in Chapter 13 of this book (price discovery), because of their high liquidity, low transaction costs, and high leverage, the perpetual-futures markets of mainstream crypto assets are often the center of price discovery for the entire crypto world. This means that the latest information about these assets is usually reflected first in the perpetual-futures market. Those who trade first in the perpetual-futures market therefore hold a systematic information advantage relative to traders in other markets (including spot market makers). Perpetual-futures market makers, though situated at the heart of price discovery, are passive liquidity suppliers by role, which dictates that they are always the ones who learn information "second," while the first to know are the informed traders who initiate the trades. This information disadvantage, endogenous to the level of market microstructure, is a structural constraint on perpetual-futures market makers and constitutes a stable, persistent component of their adverse-selection cost.

Taking the three amplification mechanisms together, we can decompose a market maker's total adverse-selection cost into three layers: the baseline adverse-selection cost (the information-asymmetry cost that exists even in an unleveraged market with fixed trading hours), the leverage-amplification layer (the additional cost induced by the high leverage of perpetual futures), and the structural-information-asymmetry layer (the cost jointly produced by 24/7 trading, the funding-settlement cycle, and the perpetual future's position as a center of price discovery). When the market operates calmly, the relative weights of these three layers are roughly balanced, and the total adverse-selection cost remains at a level a market maker can cover with normal spreads. However, when the market enters a high-volatility state—especially the extreme conditions of a liquidation cascade—the costs of the leverage-amplification layer and the structural-information-asymmetry layer expand sharply and nonlinearly. The leverage-amplification effect is further intensified by the forced liquidations that liquidation triggers (liquidation itself creates new one-directional order flow), and information asymmetry also widens as market participants' ability to interpret information diverges amid panic. The resonance of these three layers pushes the total adverse-selection cost to extreme values far above normal levels.

In summary, leverage, 24/7 trading, and structural information asymmetry together constitute the triple amplification mechanism of adverse-selection risk in perpetual-futures markets. They increase the absolute cost of adverse selection and make it vary in complex, dynamic ways, posing a systematic challenge to a market maker's risk-management and pricing capabilities beyond anything in traditional markets.

### 19.2.4 Toxicity quantification and risk monitoring

Faced with an adverse-selection threat that is amplified by multiple mechanisms and complex in origin, a market maker relying on intuition and manual adjustment alone cannot effectively cope with these high-frequency, multi-source risks. To survive, modern market makers—especially institutions engaged in high-frequency market making—must build a sophisticated quantitative risk-monitoring system to detect and assess the "toxicity" of order flow in real time and adjust their quoting strategy accordingly. Order-flow toxicity is, in essence, another way of describing adverse-selection risk; it measures the likelihood that trading with the current order flow will lead to a loss over some future period. This section introduces several core toxicity metrics, particularly VPIN, and examines how they are integrated into a market maker's decision framework.

Among the many toxicity metrics, the volume-synchronized probability of informed trading (VPIN), formally developed by Easley, López de Prado, and O'Hara (2012), is the most widely studied and applied [12]. The core idea of VPIN is to modify the traditional PIN (probability of informed trading) model to make it more suitable for the high-frequency world. The traditional PIN model infers the probability of informed trading by analyzing the arrival rates of buy and sell orders. But this "clock-time"-based sampling method runs into problems in the high-frequency world: trading activity is extremely uneven over time, with a large volume of trades possibly erupting within seconds and calm running for minutes. Analysis based on fixed time intervals can easily make wrong judgments during sparse-trading periods or be overwhelmed by information during dense-trading periods.

VPIN's core methodological innovation is that it replaces "clock time" with "volume time." Rather than partitioning data by fixed time intervals (such as every minute), it partitions data by fixed volume, forming so-called "volume buckets." Within each volume bucket, VPIN computes the absolute value of the difference between buy-side volume and sell-side volume—the "order-flow imbalance." The logic behind this is that informed traders, in order to build positions quickly, typically consume liquidity on the order book continuously and in one direction, thereby causing a significant order-flow imbalance. Conversely, the buying and selling of noise traders is more likely to offset over the short term, resulting in a smaller order-flow imbalance. By continuously computing the average order-flow imbalance across a series of volume buckets, VPIN ultimately produces a value between 0 and 1. A high VPIN value means a high degree of order-flow imbalance in the recent market, a high probability of informed trading, and strongly "toxic" order flow. Conversely, a low VPIN value indicates that the market is dominated by noise trading and the environment is relatively "safe."

The power of VPIN is that it has been shown to be an effective leading indicator of short-term market volatility, especially the "flash crashes" triggered by liquidity evaporation. When the VPIN value climbs steadily and exceeds a certain threshold, it usually foreshadows an imminent sharp price move. For a market maker, VPIN functions as a toxicity-warning indicator. When the VPIN reading is low, the market maker can offer narrow spreads and deep liquidity and earn a steady spread income. But when VPIN begins to surge, the system issues an alert, and the market maker's algorithms respond immediately: rapidly widening bid-ask spreads to compensate for the increased adverse-selection risk, cutting quoting depth to reduce risk exposure, and even, in extreme cases, withdrawing quotes entirely and waiting for market pressure to ease. The VPIN value is positively correlated with order-flow toxicity, and when VPIN enters a high range, market makers typically make more aggressive defensive quoting adjustments.

Nevertheless, VPIN as a measurement tool also has significant limitations. Its predictive power is contested in the academic literature, with some arguing that after controlling for volatility, VPIN's independent predictive value may drop substantially. In the special environment of crypto markets, the fake volume generated by wash trading may further weaken the accuracy of VPIN's buy-sell classification algorithm, because fake trades may artificially create a symmetric order flow that masks the true order-flow-imbalance signal. This also explains why, in practice, market makers do not rely on the single metric of VPIN alone. In the distinctive perpetual-futures market, a market maker's risk-monitoring system typically integrates more dimensions of targeted toxicity signals. One important class of signals stems from the mechanisms of perpetual futures themselves. For example, a rapid, large change in the funding rate often signals a sharp imbalance between long and short forces in the market, behind which strongly directional informed trading may well be hidden. Likewise, a swelling of the liquidation-queue size published by an exchange is a clear toxicity-warning signal, foreshadowing an imminent, high-impact "toxic flow" driven by forced liquidations. Astute market makers watch these indicators closely and cut their risk exposure before the liquidation wave arrives.

Another class of signals comes from a deeper analysis of order-book microstructure. For example, the frequency and size of large market orders is a key indicator. Informed traders, to ensure immediacy of execution, tend to favor market orders. Thus, when the ratio of market-order volume to total volume per unit of time rises sharply, it usually means that order-flow toxicity is increasing. In addition, analyzing the "order-taking" behavior patterns of the order book—for example, whether a large order "sweeps" through multiple price levels or merely consumes the liquidity at the best price—can also provide clues for judging the intent behind it.

In practice, a cutting-edge market-maker risk-monitoring system is a complex decision matrix that fuses dozens or even hundreds of indicators. It includes not only core toxicity metrics such as VPIN, funding rates, liquidation data, and the market-order ratio, but may also integrate cross-exchange spread data, anomalies in on-chain whale addresses, social-media sentiment indices, and even the volatility indicators of related assets (such as U.S. equity-index futures). All these data streams are fed in real time into a machine-learning model, which dynamically assesses the current market's composite "toxicity score" and, according to preset risk parameters, adjusts the width, depth, and skew of quotes at the millisecond level. This is no longer a simple "quote-fill" loop but an ongoing process of information processing, data analysis, and risk control. The sensitivity, comprehensiveness, and response speed of the monitoring system directly determine a market maker's probability of survival in the asymmetric game against informed traders.

The analytical framework of this section has consistently taken classic "informed trading"—the order flow driven by information asymmetry that causes permanent price impact—as its core object. Yet the harmful order flow a perpetual-futures market maker faces in actual operation is far from this alone. The extractive behavior of arbitrageurs exploiting cross-platform spreads, the forced-liquidation instructions that pour in during a liquidation cascade, and MEV attacks in the on-chain environment all likewise inflict losses on market makers on the surface, but they differ fundamentally from informed trading in information content, the persistence of price impact, and the optimal response strategy. Lumping these risk sources of very different natures together under the category of "adverse selection" would lead to systematic bias in strategy design. Section 19.4 of this chapter systematically extends the discussion, proposing a "toxic-flow spectrum" classification framework that identifies and analyzes, beyond adverse selection, three other toxicity sources with different dominant causes.

## 19.3 Inventory risk

Adverse selection constitutes a market maker's core risk on the information dimension, whereas inventory risk is its persistent risk exposure on the position dimension. As a passive participant in the market, a market maker is by its very nature unable to choose its counterparties or trading direction. When buy orders outnumber sell orders, it is forced to accumulate a short position; when sell orders outnumber buy orders, it is forced to accumulate a long position. This net holding, passively formed in the course of trading, is the market maker's "inventory." Every unit of inventory is an exposure subject to the risk of market price fluctuations. Even if a market maker has no view whatsoever on the market's long-term direction, this holding, produced purely by business activity, itself constitutes one of its core sources of risk. Managing inventory, keeping it within a controllable range of risk, and ultimately "clearing" it through subsequent offsetting order flow is the central task of a market maker's daily operations.

In any market, inventory management is a challenge. Yet the distinctive institutional design of perpetual futures raises the difficulty of this challenge to an entirely new magnitude. Market makers in traditional financial markets, though they also face inventory risk, usually have a reliable "safety anchor." For example, an equity market maker can hedge after the close through various cross-market instruments, while a futures market maker enjoys an ultimate certainty—the expiration and delivery date, at which all positions are forcibly closed, the basis converges to zero, and inventory risk is naturally lifted. A perpetual-futures market maker, by contrast, operates in a risk environment with no definite endpoint: the indefinite risk exposure brought by no expiration, the funding rate as a randomly fluctuating holding cost, and the highly nonlinear inventory tail risk under fat-tailed volatility together push inventory management to an entirely new level of difficulty. This section analyzes these three challenges in depth, revealing why and how, in the context of perpetual futures, inventory evolves from a routine business risk into a major source of risk for the market maker.

### 19.3.1 Adapting classical inventory models to perpetual futures

The academic study of market-maker inventory risk has a long history, and the most representative work is the model proposed by Ho and Stoll (1981) [6]. The core idea of the model is that, to manage its inventory level, a market maker dynamically adjusts its quotes. Specifically, a market maker's optimal quotes are not simply symmetric on either side of the market mid-price but are "skewed" according to its current inventory level. When a market maker has accumulated too much long inventory, it shifts its entire quote range downward: lowering the ask to attract buyers, while lowering the bid more sharply to suppress sellers, thereby incentivizing the arrival of offsetting order flow and helping to reduce its inventory. Conversely, when it has too much short inventory, it shifts its quote range upward. The degree of this quote skew depends on the market maker's inventory level, degree of risk aversion, and judgment about future price volatility.

The Ho-Stoll model provides a systematic theoretical framework for understanding inventory management, describing a market maker's behavior as an optimization process that seeks a balance between "earning the spread" and "controlling inventory risk." Building on this, Avellaneda and Stoikov (2008) formalized the problem of optimal bid and ask quotes under inventory aversion, proposing a standard framework in which quotes are set symmetrically around a "reservation price" and skewed by an amount proportional to the risk-aversion coefficient, the price variance, and the inventory level (i.e., $\delta \propto \gamma\sigma^2 Q$) [13]; this framework became the benchmark for subsequent high-frequency market-making quote optimization. However, when we apply this classic framework directly to perpetual-futures markets, we find that its underlying assumptions are significantly disconnected from market reality. This disconnect is reflected mainly in three respects, which force us to make key adaptations and extensions to the model.

First, the traditional model defines holding cost too simply. In the classic framework, the holding cost of inventory comprises mainly two parts: the opportunity cost of capital (i.e., funding interest) and the risk exposure from price fluctuations. The former is relatively stable and predictable, and the latter is usually assumed to follow a normally distributed random walk. But in perpetual futures, a brand-new cost item with a significant impact on P&L is introduced: the funding rate. The net inventory a market maker holds must pay or receive a fee in every funding-settlement period. Unlike relatively stable interest, the funding rate is a highly uncertain random variable whose direction and magnitude may change sharply over short periods. This means that the inventory holding cost of a perpetual-futures market maker contains an additional, dynamic stochastic process, which greatly increases the complexity of cost estimation and risk management.

Second, the traditional model implicitly assumes that "inventory is disposable." Whether the end-of-day settlement of the equity market or the expiration and delivery of the futures market, each provides a market maker with a mechanism to manage or clear its inventory at a specific point in time. The "no expiration" feature of perpetual futures completely breaks this assumption. A market maker's inventory management no longer has any "natural endpoint" and must rely entirely on active, continuous hedging—for example, attracting offsetting flow through quote skewing within the same exchange, or building an offsetting position on another exchange or in the spot market. These active hedging actions themselves carry additional costs and risks, such as trading fees, cross-market basis risk, and execution-latency risk.

Third, the traditional model's assumption about price-volatility risk is too benign. The model usually assumes that price fluctuations follow a normal distribution or a similar "well-behaved" distribution. However, as repeatedly emphasized elsewhere in this book, crypto markets—especially high-leverage perpetual-futures markets—exhibit pronounced "fat-tailed" characteristics in their price distributions. The "jump" risk triggered by mechanisms such as liquidation cascades is far beyond what a normal distribution can describe. This means that the tail risk facing a market maker's inventory (i.e., the possibility of an extreme loss) is systematically underestimated. The relationship between inventory cost and volatility is also not linear but exhibits a convex relationship that grows rapidly in the high-volatility region.

Comparing the inventory-cost structures of a traditional futures market maker and a perpetual-futures market maker makes the urgency of these adaptation needs clearer. For a traditional delivery-futures market maker, the composition of the inventory holding cost is relatively simple: price-volatility risk dominates (usually accounting for the larger share of total cost), followed by the opportunity cost of capital, with the remainder being trading fees and operational costs. Because of the existence of an expiration date, the time boundary of these costs is definite, and the total can be estimated fairly accurately. For a perpetual-futures market maker, the cost structure is fundamentally reconstructed: although price-volatility risk remains important, its relative share of total cost falls with the introduction of institutional cost; funding-rate exposure becomes an independent cost item comparable in scale to price-volatility risk, and under extreme market conditions it may even exceed the latter; and the liquidation risk premium and mark-price deviation risk each contribute a non-negligible incremental cost. More critically, these institutional-cost items are not independent of one another: high volatility simultaneously pushes up the extreme values of the funding rate and the probability of triggering liquidation, forming a positive correlation among the cost items that makes the total cost exhibit superlinear growth in extreme environments.

Therefore, an inventory-management model adapted to the perpetual-futures environment must incorporate these three new variables—the randomness of the funding rate, the indefinite risk exposure brought by no expiration, and the nonlinear risk under a fat-tailed distribution—into its core equation. If Section 19.1.3 defined "what" institutional cost is from the perspective of cost classification, then the task of this section is to answer, from the perspective of modeling, "how" institutional cost affects inventory-management decisions. In the classic Ho-Stoll and Avellaneda-Stoikov frameworks, a market maker's optimal quote-skew amount $\delta^*$ is determined mainly by the inventory level $Q$ and the price volatility $\sigma$: $\delta^* = f(Q, \sigma, \gamma)$, where $\gamma$ is the risk-aversion coefficient. In the perpetual-futures environment, this decision function must be extended to $\delta^* = f(Q, \sigma, \gamma, FR_t, T_{\text{eff}}, \kappa)$, where $FR_t$ is the current funding rate (affecting the directional preference of inventory holding), $T_{\text{eff}}$ is the effective time horizon (replacing the now-vanished expiration-date constraint—this is not an exogenously given parameter but a decision variable the market maker must optimize endogenously; in practice, many market makers use the 8-hour funding-settlement cycle as a natural time anchor, or adopt a multi-timescale nested risk-control framework to manage short- and medium-term inventory risk simultaneously), and $\kappa$ is a tail-risk adjustment coefficient (capturing the nonlinear amplifying effect of fat-tailed events such as liquidation cascades on inventory cost). This extended decision function is analyzed term by term in the sections that follow.

### 19.3.2 The inventory dilemma of no expiration

For a market maker operating in a traditional delivery-futures market, the expiration date constitutes a key anchor for risk management. However tortuous the trading within a contract cycle, and however out of control the inventory level, the approach of the expiration date signals the return of certainty. As the delivery date arrives, the basis between the futures price and the spot price inevitably converges to zero. All long or short positions the market maker holds are delivered or cash-settled at the final settlement price, and inventory is "naturally" cleared to zero. This process provides a definite time boundary for a market maker's risk management. It can plan its inventory exposure within a definite time frame, and even if inventory sometimes deviates substantially, it knows that the term of its risk exposure is finite. The expiration date sets a definite ceiling on a market maker's risk exposure.

A perpetual-futures market maker does not have this institutional safeguard. One of the core designs of the perpetual future is precisely that it is "perpetual"—it has no expiration date. While convenient for traders, this feature creates enormous trouble for market makers: their inventory risk exposure is, in theory, indefinite. If the market exhibits a sustained one-directional trend—for example, in an uptrend lasting weeks—a market maker's sell orders are continuously hit by buyers, causing its passively accumulated short inventory to keep swelling. Because there is no expiration date as a "terminus," this short position does not close automatically; it remains exposed day after day to the risk of continuously rising prices, and losses can expand without limit. The market maker's only way out is to conduct active, continuous hedging.

The cost of such active hedging is high and varied. The most direct approach is to attract offsetting order flow within the same exchange through "quote skewing" (the mechanism by which it adjusts quotes according to inventory direction is described in Section 19.3.1). But the price of this approach is that it may forgo opportunities to trade at more favorable prices, and if the market keeps rising, the act of high-priced hedging itself causes losses. Another approach is cross-market hedging—building an opposite position in the perpetual-futures market of another exchange or in the spot market. For example, if it has accumulated a short inventory of 100 BTC on exchange A, the market maker can buy 100 BTC on exchange B or in the spot market. Although this can neutralize the nominal risk exposure, it introduces new risks and costs: first, double trading fees; second, cross-exchange or cross-market basis risk, i.e., the spread between the two markets may move unfavorably; and finally, execution-latency risk, in that a delay in the hedging operation in a fast-moving market can cause huge slippage losses.

There is a fundamental difference in the inventory-evolution path between delivery futures and perpetual futures. Take a market maker that initially holds a short inventory of 100 BTC. In the delivery-futures market, even if this 100-BTC short position at one point expands to 200 or even 300 during the contract cycle, as the expiration date approaches, the basis-convergence effect causes the market to spontaneously generate convergent order flow, and the market maker's inventory level necessarily returns toward zero. The expiration date is like a "center of gravity": however far the inventory departs along the way, it will ultimately be pulled back to zero. This is a deterministic process embedded in the institutional design. In the perpetual-futures market, by contrast, the same 100-BTC short inventory has no such "center of gravity." If the market rises continuously for a month, this 100-lot short position not only fails to shrink automatically but may even swell further to 200 or 300 as the market maker keeps passively selling during the rise. The evolution of the inventory level depends entirely on the degree of the market's one-directional move and the efficiency of the market maker's active hedging. In the absence of an expiration-date constraint, a market maker's risk exposure has no natural convergence mechanism and can only rely on costlier active hedging to control risk. The practical implication of this path difference is significant: a delivery-futures market maker can treat an inventory deviation as "temporary," whereas a perpetual-futures market maker must treat every inventory deviation as "potentially permanent."

At a deeper level, "no expiration" also brings a cognitive dilemma—the "time horizon" problem of inventory management. Traditional market makers usually have clear risk-assessment cycles, such as end-of-day or end-of-week, at which they must ensure that inventory levels and risk exposures are within an acceptable range. The 24/7 uninterrupted trading feature of perpetual futures makes this kind of periodic assessment blurry. Risk is continuous, with no rest window. A market maker faces a thorny question: "On what timescale is my inventory level 'healthy'?" Is it one hour, eight hours, or a day? Too short a timescale may lead to overly frequent and expensive hedging ("overreaction"), while too long a timescale may leave risk exposed for too long ("underreaction"). This blurriness of the time horizon makes an optimal inventory-management strategy extraordinarily difficult to formulate and places heavy demands on a market maker's risk-control system.

### 19.3.3 Funding rates and inventory holding costs

If the no-expiration feature defines the "time" dimension of inventory risk, then the funding rate defines its "cost" dimension. The funding rate anchors to the spot price by periodically transferring fees between longs and shorts. For a market maker that passively holds inventory, however, this mechanism turns into a highly uncertain key variable that affects its profitability. The causal chain is clear: a market maker accumulates net inventory through passive fills → it is forced to participate in the transfer of fees between longs and shorts at every funding-settlement point → that inventory therefore incurs a payment or receipt determined by the direction of the rate. Whatever net inventory it holds, long or short, must participate in the payment or receipt of the funding rate in every settlement period. This makes the funding rate a new, dynamic, and hard-to-predict component of a market maker's inventory holding cost.

This mechanism can be precisely characterized with a minimal formalization: let $Q_t$ be the market maker's signed net inventory at the settlement point (positive for long, negative for short), and $FR_t$ be the funding rate at that point (a positive value means longs pay shorts); then the market maker's funding P&L in a single settlement period is

$$\text{Per-period funding P\&L} = -FR_t \cdot Q_t$$

When $FR$ and $Q$ have the same sign (e.g., holding long inventory under a positive rate), this term is negative—that is, the market maker pays; when $FR$ and $Q$ have opposite signs (e.g., holding short inventory under a positive rate), this term is positive—that is, the market maker receives. The following table is precisely the natural expansion of this expression across the four quadrants:

| Funding rate | Market maker's net inventory | Sign of $-FR_t \cdot Q_t$ | Effect on the market maker |
| :--- | :--- | :--- | :--- |
| Positive (longs pay) | Long (Q > 0) | Negative | Pays the fee; cost increases (implicit tax) |
| Positive (longs pay) | Short (Q < 0) | Positive | Receives the fee; cost decreases (implicit subsidy) |
| Negative (shorts pay) | Long (Q > 0) | Positive | Receives the fee; cost decreases (implicit subsidy) |
| Negative (shorts pay) | Short (Q < 0) | Negative | Pays the fee; cost increases (implicit tax) |

**Table 19-1.** The correspondence between funding-rate direction and a market maker's inventory holding cost (Data source: derived by the author from the single-period funding P&L $-FR_t \cdot Q_t$)

The effect of the funding rate on a market maker's inventory is two-sided: it can be either an "implicit tax" or an "implicit subsidy." When the funding rate is positive, market sentiment leans long, and longs must pay shorts. In this case, if the market maker holds net long inventory through passive fills, it must pay this funding fee, which directly increases the holding cost of its inventory. Conversely, if it holds net short inventory, it can receive this fee, and this income can partly or wholly offset the risk cost of its position, becoming a subsidy. When the funding rate is negative, the situation is exactly reversed. Table 19-1 summarizes this correspondence.

The volatility of the funding rate varies significantly across assets, further compounding the difficulty of inventory management for market makers. The following figure compares the 8-hour funding-rate distributions of four assets—BTC, ETH, SOL, and DOGE—whose cross-asset differences in standard deviation and share of positive rates are pronounced (BTC is the most concentrated, while SOL's volatility is nearly three times as high; see the statistics box in the figure). These differences directly affect market makers' pricing strategies and inventory management across pairs.

![Figure 19-4](./images/fig-19-4-en.png)

**Figure 19-4.** Comparison of 8-hour funding-rate distributions across assets (Data source: Binance perpetual-futures funding-rate history, December 2024–March 2025; the mean, median, standard deviation, and share of positive rates in the statistics box are authoritative values, while the histogram shape is an illustration reconstructed from the published means and standard deviations rather than a tick-by-tick measurement)

> Note on figure numbering: Figures in this chapter are numbered starting from Figure 19-1, in the sequence Figure 19-1, 19-2, 19-4, … 19-13. Figure 19-3 was reserved during layout and ultimately merged and omitted, so it does not appear in this chapter; to keep the figure numbers stable, the remaining numbers are not renumbered sequentially, and all original numbers are retained unchanged. The cross-sectional comparisons of the two paradigms and the three markets are instead carried by newly added tables (Table 19-3 and Table 19-4); see Sections 19.5.5 and 19.6.3.

The funding rate is itself a stochastic process, driven by the combined influence of market sentiment, leverage levels, arbitrage activity, and other factors, and it exhibits pronounced volatility and unpredictability. As shown in the following figure, a typical time series switches frequently between positive and negative values, with occasional extreme highs or lows. For market makers, this means that the inventory holding cost embeds a random term that cannot be determined in advance—a stark contrast with the stable, predictable funding costs of traditional finance (such as the overnight rate).

![Figure 19-5](./images/fig-19-5-en.png)

**Figure 19-5.** Time series of the 8-hour funding rate for BTC perpetual futures (Data source: CoinGlass Funding Rate History, January–December 2024; this figure is a representative, illustrative time series meant to convey the qualitative pattern of positive-negative switching and extreme values over the year rather than a period-by-period empirical sample, and the vertical-axis magnitude has been scaled to the realistic 8-hour funding-rate range of about ±0.01% to ±0.375%)

Beyond market-internal sentiment, the overall level of the funding rate is also driven structurally by the global macro interest-rate environment. In a low-interest-rate environment (such as 2020–2021), the large spread between the low funding costs of traditional financial markets and the high leverage demand of crypto markets pushed the positive funding rate of perpetual futures steadily higher; whereas during a global rate-hiking cycle (such as 2022–2023), the rise in traditional funding costs was partly transmitted to crypto markets, compressing the carry space [9]. This means that a market maker's inventory holding cost embeds not only the microstructural randomness of the funding rate but also the slow-variable influence of the macro interest-rate cycle.

This randomness introduces new complexity into a market maker's strategy. On one hand, a market maker has an incentive to actively manage its inventory direction so that it aligns with the expected direction of the funding rate. For example, if the market broadly expects the funding rate to stay positive for some time (such as in a bullish atmosphere), a market maker will lean toward holding net short inventory, because this allows it to steadily "collect rent." It may adjust its quote-skewing strategy to make it easier to accumulate short positions. However, this introduces an entirely new risk dimension: a "directional bet" on the funding rate. If market sentiment suddenly reverses and the funding rate flips from positive to negative, then the short inventory that had been a source of subsidy instantly becomes a burden requiring high fee payments, causing a double loss from both "price fluctuations" and "the funding rate."

Furthermore, there is a subtle "reflexivity" between a market maker's behavior and the funding rate. We must first clarify the tension between this reflexivity and the earlier point that "market makers passively hold inventory and cannot choose direction" (see Section 19.1.2 and the opening of Section 19.3): on the basis of the inventory formed by passive fills, a market maker can still, at the margin, actively lean toward a particular inventory direction through quote skewing; reflexivity operates precisely on this marginal active portion and does not negate its fundamental nature as a passive liquidity supplier. As the analysis of the funding-rate game in Chapter 10 of this book shows, the overall long-short position imbalance in the market is the key factor determining the direction and magnitude of the funding rate. As the most important liquidity providers in the market, market makers' marginal choice of inventory direction (i.e., whether they lean toward accumulating long or short inventory) affects, to some degree, the market's net-position balance. For example, if most market makers lean at the margin toward holding short inventory in anticipation of a positive rate, their collective behavior itself may intensify the market's short lean and thereby, in turn, push the funding rate down. This reflexive feedback loop is a theoretical inference drawn from Chapter 10; this chapter does not test it independently or empirically. This feedback loop between individual decisions and macro market indicators makes the dynamics of the funding rate even more complex, and it also means that any misjudgment of the rate direction by a market maker can turn the holding cost from a subsidy into a penalty, causing significant additional losses. The following figure shows the dynamic relationship among open interest, price, and the funding rate for BTC perpetual futures: rapid increases in open interest often accompany one-directional price movements, while the funding rate exhibits a pronounced clustering effect—positive and negative funding rates each form continuous "clusters" rather than a random distribution. This clustering is directionally consistent with the theoretical inference about funding-rate reflexivity and can be regarded as observational corroboration.

![Figure 19-6](./images/fig-19-6-en.png)

**Figure 19-6.** The dynamic relationship among open interest, price, and the funding rate for BTC perpetual futures (Data source: Binance BTCUSDT perpetual-futures open-interest and funding-rate data, February–March 2025; this figure is a representative, illustrative series in which the magnitudes of open interest, price, and funding rate follow public baselines rather than a period-by-period empirical sample)

### 19.3.4 The nonlinear relationship between volatility and inventory cost

In classical inventory-management models, the holding cost of inventory is usually assumed to be proportional to market volatility. This linear assumption is a reasonable and effective approximation in low-volatility, "normal" market environments. It means that if the market's annualized volatility doubles, the risk compensation a market maker requires per unit of inventory also roughly doubles. However, in perpetual-futures markets, where leverage, liquidation cascades, and information-transmission effects act together, this linear assumption deviates significantly from actual data. The relationship between a perpetual-futures market maker's inventory cost and volatility is nonlinear—specifically, convex.

A convex relationship means that as volatility increases, the inventory cost grows at an ever-accelerating rate. To make the magnitude of this convexity intuitive, consider the following illustrative numbers (a scenario constructed from market makers' empirical characteristics for explanatory purposes, not a measured-sample statistic): when the annualized volatility of BTC perpetual futures is in the low-volatility range below 30%, a market maker holding $1 million of net inventory incurs a daily risk cost roughly on the order of $200–500, broadly consistent with the linear assumption's prediction. But when volatility climbs to the medium-high range of 50–70%, the daily risk cost of the same-sized inventory may grow to $1,500–3,000, an increase far exceeding the growth in volatility itself. And in an extreme environment where volatility surges above 100% (such as during the global risk-asset sell-off of August 2024), the inventory cost may grow in a superlinear (convex) manner to more than $10,000 per day, on the order of 20 times the low-volatility level. The three tiers of figures above are meant to convey the qualitative conclusion that "cost rises at an accelerating pace with volatility," not to serve as precise, calibratable estimates. The traditional linear assumption and the convex reality of perpetual futures differ little in the low-volatility region but diverge rapidly in the high-volatility region. The ever-widening gap between the two is caused precisely by the distinctive risk-amplification mechanisms of perpetual futures (i.e., liquidation cascades, liquidity contraction, and the superposition of institutional cost).

The core reason for this convex relationship is that, in perpetual-futures markets, a high-volatility environment is not an isolated variable; it is often activated simultaneously with multiple negative-feedback mechanisms. First, high volatility is an important trigger of large-scale "liquidation cascades." When prices swing sharply, the margin of many high-leverage positions is punched through, triggering forced liquidations by the exchange. These forced-liquidation market orders further impact the market, driving prices further in the same direction and thereby triggering the liquidation of more accounts. In this positive feedback loop, if a market maker holds an opposing inventory, its losses are amplified disproportionately. It faces not a smooth random walk of prices but violent, cliff-like price "jumps."

Second, a high-volatility environment is always accompanied by a sharp contraction of liquidity. This mechanism is essentially identical to the "liquidity spiral" identified by Brunnermeier and Pedersen (2009) [14] in traditional financial markets: rising volatility → higher margin requirements → forced deleveraging → falling liquidity → amplified price impact → further rising volatility. In the high-leverage liquidation environment of perpetual futures, this spiral is further accelerated, because the forced-liquidation mechanism turns "forced deleveraging" from a gradual process into a threshold-triggered discrete event. As discussed earlier in this chapter, when adverse-selection risk and inventory risk rise sharply, a market maker's rational choice is to widen bid-ask spreads, reduce quoting depth, and even withdraw from the market entirely to protect itself. This collective "risk-avoidance" behavior makes the order book extraordinarily thin during high-volatility periods. For market makers that remain in the market and hold inventory, this means that when they try to actively close or hedge their risk positions, they face enormous slippage costs. Inventory that could originally be managed with low-cost operations becomes "toxic" and hard to dispose of in a high-volatility environment, and its value at risk rises sharply.

Finally, a high-volatility environment also amplifies other institutional costs. For example, the deviation between the mark price and the latest traded price becomes larger, increasing the risk of needless liquidation; and the funding rate may become extreme over a short period, imposing a heavy penalty on market makers holding inventory in the unfavorable direction. All these factors together constitute the "extra cost" beyond the linear assumption. This extra cost is nearly invisible in a low-volatility environment but expands sharply in a high-volatility environment, becoming the major component of a market maker's inventory holding cost.

Using historical data for BTC perpetual futures, the following figure divides the 14-day rolling volatility into low, medium, and high regimes: in the high-volatility regime (annualized volatility above 40%), the mean intraday price range for BTC is about 1.6 times that of the low-volatility regime, and the dispersion of fluctuations increases markedly with extreme values appearing frequently. This nonlinear volatility-amplification effect is directionally consistent with the claim that market makers' inventory costs grow convexly.

![Figure 19-7](./images/fig-19-7-en.png)

**Figure 19-7.** BTC volatility regimes and intraday price range (Data source: Binance BTCUSDT perpetual-futures daily candlestick data, March 2024–March 2025)

Going further, the following figure reveals the relationship between volatility and the funding rate: the time series on the left shows a temporal resonance between periods of surging volatility (such as August and December 2024) and extreme funding rates, while the scatter analysis on the right shows a weak linear correlation between the two (r = 0.076). This must be interpreted with caution: a near-zero correlation coefficient of r = 0.076 can only refute the existence of a simple linear mean relationship between volatility and the funding rate; it does not by itself constitute heteroskedasticity evidence for "the variance of the funding rate being amplified under high volatility." The causal chain "higher volatility → greater funding-rate variance → surging uncertainty in inventory holding costs" should therefore be understood as a hypothesis awaiting testing rather than a conclusion already confirmed by Figure 19-8; rigorously testing this hypothesis would require further bucketing the funding rate by volatility and examining its standard deviation and extreme-value statistics, a formal estimation this chapter does not undertake.

![Figure 19-8](./images/fig-19-8-en.png)

**Figure 19-8.** The dynamic relationship between BTC volatility and the funding rate (Data source: Binance BTCUSDT perpetual-futures daily candlestick and funding-rate data, March 2024–March 2025)

This nonlinear relationship directly explains market makers' behavioral patterns in high-volatility environments and their impact on market liquidity. It explains why, at the extreme moments when the market most needs liquidity (such as when prices crash or spike), liquidity instead contracts sharply. This is not because market makers lack willingness or harbor subjective malice, but because in that environment the inventory cost they face has grown to a level that normal spreads simply cannot cover. Continuing to provide liquidity would lead to unbearable losses. Withdrawing from the market is the only rational answer given by their profit equation. This also provides the most direct microeconomic-mechanism explanation for the theme we will examine in depth in the next chapter—why liquidity vanishes precisely when it is most needed. The nonlinearity of inventory risk is a key link in understanding the endogenous fragility of perpetual-futures markets.

## 19.4 The toxic-flow spectrum

In classical market-making theory, "toxic flow" (shorthand for the toxic order flow that inflicts systematic losses on market makers; the remainder of this chapter uses the term "toxic flow" uniformly) is a concept nearly interchangeable with "informed trading." It paints a clear but somewhat monotonous picture: some participants in the market hold nonpublic information—"the informed"—and they trade with market makers whose information lags relatively behind, causing market makers to incur systematic losses, namely the adverse-selection cost. A market maker's way to survive is to compensate for potential losses by adjusting its quoting spread in a high-risk environment composed of informed trading. However, when this analytical framework is applied to the perpetual-futures market—high-leverage, around-the-clock, and spanning both centralized and decentralized platforms—the sources and impact mechanisms of toxic flow exhibit far greater complexity. Lumping all order flow unfavorable to market makers together as "informed trading" oversimplifies reality and can lead to serious strategic errors.

This section proposes and systematically develops one of this book's integrative classification frameworks: the toxic-flow spectrum. We must first clarify the nature and boundaries of this framework: it does not claim to have "discovered" several entirely new risk types but instead places the harmful order flow perpetual-futures market makers already face into a unified coordinate system composed of "information content" and "permanence of price impact," so as to systematically compare their mechanistic differences and design differentiated responses accordingly. Within this coordinate system, we identify, by "dominant cause," four typical types of toxic flow: fundamentals-driven "informed-trading toxicity," cross-platform-spread-driven "arbitrage-extraction toxicity," "liquidation-cascade toxicity" endogenously determined by the leverage mechanism, and "MEV-attack toxicity" unique to the decentralized environment. These four are a perspective divided by dominant cause, not a mutually exclusive and collectively exhaustive (MECE) set of independent categories: their boundaries can overlap on the coordinate system—"informed-trading toxicity" is essentially the naming, within this coordinate system, of the adverse-selection risk of Section 19.2, while the latency-arbitrage form of "arbitrage-extraction toxicity" on an on-chain CLOB overlaps mechanistically with the second "cross-market information-relay" type of informed trading in Section 19.2.2. The term-by-term analysis of the four types of toxicity below should therefore be understood as a characterization of four dominant regions on the same coordinate plane, not an enumeration of four mutually disjoint sets.

This coordinate system is built on two core dimensions. The first dimension is "information content": whether the order flow embeds new information about an asset's fundamental value. Informed trading has high information content; arbitrage extraction and liquidation cascades have lower information content (the former exploits an already-public cross-market spread, the latter is an endogenous product of the leverage mechanism); and MEV attacks contain no fundamental information. The second dimension is "permanence of price impact": whether the price change triggered by the order flow persists after the fact. The price impact caused by informed trading is almost entirely permanent; the impact of arbitrage extraction has medium permanence (pushing the price toward fair value); the impact of a liquidation cascade contains a large temporary-overshoot component; and the price impact of an MEV attack is usually absorbed within the same block. The following table maps the dominant characteristics of the four types of toxic flow onto this two-dimensional framework:

| Toxicity type | Information content | Permanence of price impact | Temporal characteristics | Primary response strategy |
| :--- | :--- | :--- | :--- | :--- |
| Informed-trading toxicity | High | High (permanent) | Event-driven pulses | Widen spreads, reduce depth, avoid |
| Arbitrage-extraction toxicity | Low | Medium (toward fair value) | Continuous, positively correlated with volatility | Shorten oracle latency, dynamic fees |
| Liquidation-cascade toxicity | Low | Low (large temporary overshoot) | Sudden pulses, positive feedback | Real-time signal classification, controlled absorption |
| MEV-attack toxicity | None | Very low (absorbed within the block) | High-frequency, synchronized with blocks | Private transaction channels, MEV-resistant protocols |

**Table 19-2.** The two-dimensional classification framework of the toxic-flow spectrum (Data source: constructed by the author)

This table characterizes the dominant features of the four types of toxicity under normal market conditions, not a mutually exclusive discrete classification. In extreme markets, the four types can superimpose and resonate, and individual types (such as arbitrage extraction) may even reverse direction during the price-recovery phase; the relevant dynamics are detailed in Section 19.4.5. Although the dominant regions of these four types of toxic flow each have their own emphasis, they often interweave and superimpose at the market's extreme moments, forming composite risk events that cause sharp liquidity contraction. Conflating them leads to a serious mismatch of response strategies: using a retreat strategy meant for informed trading to deal with liquidation flow would miss the rebound opportunity, whereas the reverse could incur major losses in a genuine price impact. Understanding this spectrum is the key to understanding the microdynamics of liquidity—and indeed the systemic risk—of perpetual-futures markets. The sections that follow analyze the mechanisms and characteristics of these four types of toxic flow in turn.

### 19.4.1 Informed-trading toxicity

Informed-trading toxicity is the most classic type in the toxic-flow spectrum—the naming, within this coordinate system, of the adverse-selection risk systematically analyzed in Section 19.2. As noted, incorporating it into the spectrum does not introduce a new risk but provides a common benchmark for the systematic comparison of the four types of toxicity. This section does not repeat its theoretical mechanism (the applicability of the Glosten-Milgrom model (Glosten and Milgrom, 1985 [4]), the four sources of informed trading, the amplifying effects of leverage and 24/7 trading, and quantitative tools such as VPIN have all been elaborated in detail in Section 19.2); instead, it focuses on placing its core characteristics into the comparative framework of the toxic-flow spectrum, so as to form a systematic contrast with the three types of toxicity that follow.

Informed-trading toxicity has three core defining characteristics. First, its price impact is permanent. The buying and selling of informed traders reflect fundamental information not yet priced by the market, so the price change they drive does not revert after the trade is completed, and the market maker's loss is deterministic and irreversible. Second, its temporal characteristic is an event-driven pulse distribution, erupting in the brief windows before and after the release of major information. Third, its optimal response strategy is "detect and avoid": widening spreads to raise the informed trader's cost, reducing depth to limit maximum loss, and skewing quotes to actively manage inventory direction. These three characteristics—permanent impact, pulse-like temporal distribution, and avoidance-type response—recur repeatedly in the contrast with the three types of toxicity that follow, and the differences among them are precisely the core value of the toxic-flow classification.

### 19.4.2 Arbitrage-extraction toxicity

If informed-trading toxicity is a game between a market maker and those with an information advantage, then arbitrage-extraction toxicity manifests more as a structural friction between different market mechanisms. This toxicity is especially pronounced in the automated market maker (AMM) or liquidity-vault models of DEXs; it stems from the inherent lag of DEX pricing mechanisms relative to CEXs. A CEX has a high-frequency order-book matching engine, whose internal processing can reach the microsecond level and whose external price updates are at the millisecond level. A DEX, by contrast—especially a protocol that relies on oracle price feeds—has an inherent lag in its price updates, which can range from seconds to minutes. This time difference creates a risk-free arbitrage opportunity.

When the market price swings sharply, the CEX price has already reflected the change quickly, while the oracle price on the DEX has not yet updated. At this point, arbitrageurs step in swiftly. They buy at the latest low price on the CEX and then sell to the liquidity provider (LP) on the DEX at the not-yet-updated, relatively high "stale price"; or they sell at the latest high price on the CEX and then buy from the DEX's LP at the lower stale price. For the DEX's LP, they are forced, in every round of price fluctuation, to trade at a belated, wrong price with the fastest-reacting arbitrageurs in the market. This systematic loss is termed loss-versus-rebalancing (LVR) in the academic literature (Milionis et al., 2022) [15]; it constitutes a structural cost facing DEX LPs and forms a continuous value-extraction effect.

Unlike informed-trading toxicity, the price impact caused by arbitrage-extraction toxicity is not necessarily permanent. The behavior of arbitrageurs is aimed precisely at eliminating the spread, and their trades push the DEX price toward the CEX's fair price. But for the LP, the harm is already done. Their loss comes from the moment of trading at an unfavorable price, and this loss is amplified as market volatility increases, because higher volatility means more frequent and more violent CEX-DEX spreads. Arbitrage-extraction toxicity is therefore not an event-driven pulse but a persistent, structural toxicity positively correlated with market volatility. As long as the informational and mechanistic lag between the CEX and the DEX exists, this value extraction will not stop.

Faced with this toxicity, the responses of on-chain market makers lie more at the level of protocol design. For example, a protocol can strive to shorten the oracle's update lag and adopt more manipulation-resistant price-feed mechanisms; it can introduce dynamic fees that automatically raise trading fees when market volatility intensifies, thereby increasing the arbitrageurs' cost; or it can weaken arbitrageurs' time advantage through mechanisms such as trading-speed limits and batch auctions. For on-chain market makers, LVR is a structural cost that must be paid to participate in DEX market making, and its return (trading-fee income) must adequately compensate for this foreseeable loss.

The concept of LVR strictly applies to liquidity providers based on AMMs or oracle pricing (such as GMX's GLP model), because its core mechanism is the systematic deviation between the "stale price" and the "latest price." For professional market makers operating on an on-chain CLOB (such as Hyperliquid or dYdX), the form of arbitrage extraction they face differs from LVR and is closer to cross-platform latency arbitrage between CEXs—essentially identical to the second "cross-market information-relay" type of informed trading discussed in Section 19.2.2, except that the higher latency of the on-chain environment amplifies this risk. This is precisely one piece of evidence for the earlier statement that "the boundaries of the four types of toxicity can overlap": the same harmful order flow can, depending on its on-chain or off-chain carrier, fall simultaneously at the boundary of the two dominant regions of "arbitrage-extraction toxicity" and "informed-trading toxicity." The "arbitrage-extraction toxicity" in Table 19-2 should therefore be understood as an umbrella concept encompassing two concrete manifestations: LVR in the AMM environment and latency arbitrage in the order-book environment.

### 19.4.3 Liquidation-cascade toxicity

Relative to the other three types of toxic flow, liquidation-cascade toxicity has higher instantaneous impact intensity and a stronger positive-feedback character in perpetual-futures markets. Its mechanistic root lies in the positive feedback loop between high leverage and forced liquidation. When a sharp price move triggers the forced liquidation of opposing high-leverage positions, these market-order closings further impact the price, thereby triggering a new round of forced liquidations and forming a dangerous positive feedback loop that generates a huge one-directional order flow in an extremely short time—the liquidation cascade. The empirical analysis by the OECD (2023) of three major DeFi lending protocols—Aave, Compound, and Maker—provides quantitative evidence for the positive feedback loop between liquidations and price volatility [16]; although its subjects are DeFi lending protocols rather than CEX perpetual futures, the liquidation positive-feedback mechanism it reveals is essentially common to both.

This toxicity differs fundamentally from the other three types. The initiators of liquidation orders (the traders being liquidated) are not "informed." Their motive for trading is not that they possess future information but that they are forcibly liquidated due to insufficient margin. However, liquidation flow has characteristics similar to informed-trading flow in its price-impact effect, and because of its price-insensitive market-order nature, its impact intensity per unit of time is often higher.

The key difference, however, is that the price impact driven by informed trading reflects a permanent change in fundamentals, whereas the price impact driven by a liquidation cascade contains a large "temporary" or "overshoot" component. Liquidation flow is price-insensitive; it merely aims to close out risk exposure as fast as possible, and therefore often pushes the price to an extreme level far beyond its fair value. After the cascade ends, this excessive price deviation is usually partly or even fully corrected, and the price shows a pronounced V-shaped reversal.

The market event of August 5, 2024, can serve as a single illustrative case of the practical value of the toxic-flow spectrum, rather than as a hypothesis test of the framework. A prior caveat is in order: the following quantitative description of the event is synthesized from public exchange historical candlesticks and third-party liquidation data (such as CoinGlass Liquidation Data), with sample time points noted for readers to verify, but a single case alone is insufficient to confirm or refute general claims such as "toxicity superposition." This event has pedagogical value because it clearly demonstrates the process by which multiple types of toxic flow are activated in causal sequence. The initial shock did not originate within the crypto market but was a typical macro-information-driven event: an unexpected rate hike by the Bank of Japan triggered a large-scale unwinding of the global yen carry trade, setting off a chain sell-off of risk assets from Japanese equities to U.S. equities and then to crypto markets. In the first phase of the event, the price decline in crypto markets was driven primarily by macro-information-driven informed-trading flow: traders who were the first to read the signal of a global repricing of risk assets built short positions in the perpetual-futures market—this belongs to the informed-trading toxicity in this section's coordinate system. In the second phase, the continued price decline began to trigger the forced liquidation of many high-leverage long positions, and liquidation-cascade toxicity took over: according to public historical data, the BTC perpetual-futures price fell sharply from about $62,000 to about $49,000 within a few hours (a decline of about 21%), and according to third-party liquidation data, over $1 billion of long positions were forcibly liquidated. During the roughly 30-minute window of most intense liquidation, the bid-side order-book depth of the Binance BTCUSDT perpetual futures was heavily consumed, and the price briefly touched a low of about $49,100. For market makers, identifying the transition point from the first phase (information-driven, with the price reflecting a permanent fundamental repricing) to the second phase (liquidation-driven, containing substantial temporary overshoot) is the key to signal classification. After the flood of liquidations subsided, the price rebounded to about $56,000 over the following roughly 24 hours, recovering about 54% of the decline as calculated by $(56{,}000-49{,}000)/(62{,}000-49{,}000)$. This V-shaped movement is consistent with the core difference between liquidation-cascade toxicity and informed-trading toxicity in the permanence of price impact. For market makers that accurately identified the "liquidation-dominant" signal in the second phase and engaged in controlled absorption, this constituted a conditional positive-expected-return opportunity (whose positive expectation depends on the real-time accuracy of estimating the upper bound of the cascade's scale, as discussed below).

![Figure 19-9](./images/fig-19-9-en.png)

**Figure 19-9.** Distribution of liquidation volumes for major liquidation events in 2024 (Data source: CoinGlass Liquidation Data, January–December 2024)

This poses a key signal-identification problem for market makers. In the extreme environment of an erupting liquidation cascade, a market maker must judge within seconds or minutes: is this enormous selling pressure an informed-trading flow reflecting some major bad news, or merely a temporary liquidation flow triggered by the implosion of the leverage system? If it is judged to be the former, the market maker should immediately pull its quotes and stay away from the market, or it will suffer a huge permanent loss. But if it is judged to be the latter, this may instead constitute a positive-expected-value trading opportunity: by actively absorbing these price-insensitive liquidation orders, the market maker can profit in the subsequent price rebound. The asymmetric payoff structure of this decision demands high-precision, real-time signal classification.

A market maker's response strategy therefore becomes highly dependent on the ability to parse real-time data. It closely monitors indicators such as the liquidation data, open-interest changes, and funding rates published by major exchanges, trying to identify the true source of the current order flow. If the data confirm that the selling pressure comes mainly from the liquidation engine, a mature market maker may adopt a "controlled absorption" strategy: it provides liquidity moderately, absorbing part of the liquidation orders, while strictly controlling its own risk exposure through algorithms to avoid exhausting all available capital as the cascade continues to expand. The positive expected value of this strategy is conditional, not unconditional: it holds only if the market maker can make a sufficiently accurate real-time estimate of the cascade's remaining scale; once that estimate fails, the act of absorption itself turns into a major loss. In other words, "controlled absorption" has both success stories and many failures, and one must beware of survivorship bias when describing it as a hallmark capability of top market makers. This strategy faces multiple constraints in practice. First, the total scale and duration of a liquidation cascade are unknowable in advance, and absorbing too early or on too large a scale may risk the price falling a further tens of percent. In the August 5, 2024, event, the time span from the initial decline to the final low exceeded 6 hours, with multiple false rebounds along the way. Second, the exchange's liquidation engine enjoys priority of execution, and a market maker's buy orders usually rank behind the liquidation sell orders in the matching queue, which means the market maker may not be filled at the expected price. Third, implementing an absorption strategy simultaneously on multiple exchanges may cause large amounts of capital to be frozen on exchanges experiencing extreme conditions, losing the flexibility to make necessary reallocations in other markets. Controlled absorption is therefore by no means simple contrarian buying but requires extremely high technical execution, precise risk-budget management, and the ability to assess the cascade scale in real time; whether it can produce a positive return depends heavily on whether the above conditions are met simultaneously.

### 19.4.4 MEV-attack toxicity

MEV-attack toxicity is a distinctive risk born of the blockchain environment, especially DEXs. Its root lies in the transparency of transactions on public chains. When a user or market maker submits a transaction, it is not executed immediately but enters a public "mempool," waiting for a miner or validator to package it into the next block. This waiting time provides an attack window for "MEV searchers" who monitor the mempool and perform precise arbitrage on high-value transactions. In their systematic study of front-running behavior in decentralized exchanges and miner extractable value, Daian et al. (2020) were the first to provide an ontological definition and empirical characterization of MEV and priority gas auctions (PGA) [17], laying the foundation for understanding the extractable value of on-chain transaction ordering.

For market makers providing liquidity on on-chain order-book DEXs, the MEV attacks they face take mainly two forms: front-running and sandwich attacks. In a front-running attack, an MEV searcher detects that a market maker has submitted a profitable limit order (for example, a buy order below the current market price) and immediately submits its own buy order, paying a higher gas fee so that its transaction is executed before the market maker's, thereby intercepting the fill opportunity that should have belonged to the market maker. In the more destructive sandwich attack, when a searcher observes a large transaction likely to have a significant price impact, it acts immediately: first, it pays a high gas fee to submit its own buy order, executing ahead of the large transaction to push the price up (front-running); then, after the large transaction is filled at the elevated price, it immediately submits a sell order to sell off the position it bought earlier for a profit (back-running). The market maker is sandwiched in the middle, completing two trades at unfavorable prices, and its value is systematically extracted by the MEV searcher.

The temporal characteristic of this toxicity is highly synchronized with block production; it is a high-frequency, structural toxicity. In every block, an on-chain arbitrage attack may occur. Every quote a market maker posts and every position adjustment it makes are exposed to the monitoring of all MEV searchers, and its strategic intent is publicly interpreted and exploited. This makes active, high-frequency market-making strategies extraordinarily difficult and dangerous to run on DEXs.

The strategies for countering MEV-attack toxicity are also highly technical. The most direct method is to bypass the public mempool and send transactions directly to block builders through private channels, avoiding having transactions "seen" in the public market. In addition, market makers can establish cooperative relationships with block builders, exchanging payment for priority processing of their transactions and protection from attack. From the perspective of protocol design, some new DEX architectures—such as exchanges that adopt batch auctions or commit-reveal schemes—aggregate orders within a specific time window and then clear them at a uniform price, eliminating the value of transaction ordering and thereby simultaneously weakening both LVR extraction and sandwich attacks at the mechanism level. The research of Canidio and Fritsch (2023) [18] provides theoretical and simulation evidence that a function-maximizing batch-auction AMM can make LP returns slightly higher than the empirical level of Uniswap v3 while sharply compressing arbitrageurs' profits and the space for sandwich attacks. For DEX market makers, choosing a trading platform with an MEV-resistant design is the first line of defense for survival.

### 19.4.5 Toxicity superposition and synergistic dynamics

This section does not introduce a fifth type of toxicity; it analyzes the cross-cutting superposition dynamics of the preceding four types within the same coordinate system. The previous four sections characterized the typical regions of the four types of toxicity by dominant cause; this section focuses on how they are activated simultaneously and reinforce one another at the market's extreme moments, thereby forming a composite risk that no single type of toxicity alone can explain. A market crash or violent swing is never a linear process triggered by a single factor but a complex-system phenomenon in which multiple risks resonate and positive feedback loops are activated.

We can envision a typical market-crisis scenario. First, an unexpected major piece of bad news (for example, the depegging of a large stablecoin) appears, triggering large-scale selling by informed traders—the manifestation of "informed-trading toxicity." The sharp price decline causes the CEX-DEX spread to widen far beyond normal levels, attracting a swarm of arbitrage bots to pour into the DEX and extract value—the further intensification of "arbitrage-extraction toxicity." Immediately after, the continued price decline triggers the forced liquidation of the first batch of high-leverage long positions, setting off "liquidation-cascade toxicity," as huge market sell orders flood into the market and push the price down further. And on-chain, as panic spreads, trading volume surges, gas fees soar, and the mempool becomes a battleground of fierce competition among MEV searchers, with "MEV-attack toxicity" reaching its peak; any attempt to conduct a rescue or adjust a position on-chain may be sandwiched or front-run.

At this moment, the four types of toxic flow reach their peaks simultaneously. A market maker finds that the total level of toxicity it faces may be far from the simple arithmetic sum of the four toxicity intensities, but instead exhibits a superlinear amplification. We call this proposition the "toxicity-resonance hypothesis." To make this hypothesis distinguishable by data rather than always true, we offer a falsifiable criterion: if, in an event where multiple types of toxicity occur concurrently, the total price impact the market bears does not exceed the sum of the impacts produced by each type of toxicity acting alone (at comparable scale), then "superlinear superposition" does not hold in that case—that is, the resonance hypothesis is supported only when the concurrent impact is systematically greater than the sum of the individual impacts. This criterion could, in principle, be tested by comparing the price-impact elasticity of a "multi-toxicity concurrent window" and a "single-toxicity-dominant window" at the same order size; this chapter does not undertake such a formal estimation, so it is proposed as a hypothesis awaiting testing.

The core transmission mechanism of this superlinear superposition (when it holds) lies in the positive-feedback path of "depth consumption → impact amplification": the first wave of toxic flow (informed trading) consumes the first layer of order-book depth, so that the subsequent arbitrage flow and liquidation flow face a thinner order book; an order of the same size produces a significantly larger price impact on a thinner order book, and this amplified impact in turn triggers more liquidations and more space for MEV extraction. Each wave of toxic flow creates more extreme impact conditions for the next. Every buy order a market maker posts may be attacked simultaneously from three directions—by the informed, by arbitrageurs, and by the liquidation engine—while an on-chain LP must also endure the fourth blow of MEV. In this extreme environment, the expected profit of market making quickly turns deeply negative, and any rational market-making algorithm points to the same conclusion: immediately withdraw from the market, or widen the spread to a level at which it is almost impossible to fill. This is not because market makers are "irresponsible" or "lacking in commitment," but is the inevitable choice driven by their economic rationality. When the cost of providing liquidity soars far beyond any possible return, the supply of liquidity stops.

As a contrast to the resonance scenario above, consider a "mild pullback in which toxicity does not resonate": when a price pullback is driven by a relatively single, mild factor (such as a neutral-to-bearish macroeconomic data release), and the first layer of order-book depth, once consumed, can be replenished in time by off-book market makers' supplementary quotes, then subsequent orders do not face a continuously thinning order book, no depth-consuming positive feedback forms among the various types of toxicity, and the total price impact roughly equals the linear superposition of the individual factors' impacts. The resonance hypothesis predicts that these two scenarios should differ systematically in the ratio of "concurrent impact to the sum of individual impacts"—in the mild-pullback scenario the ratio is close to or below 1, whereas in a genuine resonance scenario the ratio is significantly greater than 1. The significance of introducing this contrast scenario is to make "toxicity resonance" a conditionally holding claim distinguishable by data, rather than an ex post narrative that can justify itself however the market behaves.

This is the microlevel explanation of the core question of "why liquidity contracts sharply when market pressure is greatest." Liquidity is not an exogenous resource that can be supplied without limit but a commodity endogenously produced by market makers after precisely calculating costs and returns. When the supply cost rises sharply because of the resonance of toxic flow, the supply activity quickly stops. The four types of toxic flow do not superimpose in the same direction at all phases. In the price-recovery phase after a liquidation cascade ends, the direction of arbitrage-extraction toxicity may reverse: arbitrageurs now relay price information from the underpriced market toward fair value, objectively aiding the price's return. This phase difference means that the superposition effect of toxic flow is concentrated mainly in the initial phase of a crisis's eruption, while in the recovery phase a partial hedge may emerge. Understanding the toxic-flow spectrum and the dynamics of its superposition and hedging is not only a survival guide for market makers but also the key to understanding the fragility of the entire market ecosystem. It foreshadows the theme we will examine in depth in the next chapter: the endogeneity of liquidity and the roots of its fragility in a crisis.

## 19.5 Market making on CEX CLOBs

In perpetual-futures markets, the CLOB of centralized exchanges has always occupied a core position. Although the wave of decentralized finance has brought various innovative paradigms such as on-chain order books, in terms of the absolute scale and depth of liquidity, the CEX CLOB still dominates to this day. According to estimates in CoinGlass's industry annual report, the average daily volume of the CEX derivatives market in 2025 was on the order of $260 billion (institutional data source, sample period 2025) [19], of which perpetual futures accounted for the overwhelming majority. The liquidity of this market does not form naturally but is systematically built and maintained by highly specialized market-making institutions. Leveraging their enormous advantages in technology, capital, and information, these institutions have built the core skeleton of perpetual-futures market liquidity, while also forming a complex game structure around risk and profit. This section analyzes in depth the ecosystem, strategic framework, and risk management of CEX CLOB market making, as well as the symbiotic relationship between market makers and exchanges, and ultimately reveals, through comparison, the distinctiveness and extreme difficulty of the perpetual-futures market-making business.

A disclosure about evidence grade is in order at the outset: the specific figures involved—market-maker concentration, order-book depth share, and fee rebates—come mainly from the in-house metrics of data vendors, market makers, and exchanges (industry research reports and official announcements), and have not been independently verified by academia or regulators. In interpreting these figures, readers should treat them as order-of-magnitude references provided by interested parties rather than as precise measurements confirmed by peer review; where such figures are cited below, the nature of their source is noted at each point, and independent academic or regulatory cross-sources are added where possible.

### 19.5.1 The market-maker ecosystem

The market-making arena for CEX perpetual futures is a classic oligopolistic market. The participants are primarily a small number of well-capitalized, technologically leading professional institutions, which provide the ecosystem with its most fundamental and important element: liquidity. This ecosystem can be roughly divided into three layers.

At the top of the pyramid are the global Tier 1 crypto market makers, such as Wintermute, Jump Crypto, Amber Group, GSR Markets, B2C2, and Cumberland/DRW. These institutions typically originate from traditional high-frequency trading firms and possess deep quantitative-trading backgrounds and powerful technical infrastructure. Their business spans the major CEXs worldwide, and they provide liquidity for hundreds of crypto assets simultaneously. Their advantages are all-encompassing: ample capital allows them to withstand enormous inventory risk, cutting-edge low-latency systems let them capture arbitrage opportunities at the microsecond level, and large data teams support complex pricing models and toxic-flow-identification algorithms. These top institutions are the core providers of market liquidity. According to industry reports from firms such as Kaiko and Wintermute, the top three to five market makers together contribute on the order of 70% of the order-book depth in mainstream perpetual-futures markets such as those for Bitcoin and Ethereum. The evidence grade of this figure must be treated with caution: the relevant estimate comes from in-house metrics, its calculation method is not public, and Wintermute is itself a top market maker, so an interested party estimating the share of the very group to which it belongs entails an inherent conflict of interest; this figure is therefore used not as a single precise number bearing a systematic claim but only as an order-of-magnitude reference. As cross-corroboration, Aramonte et al. (2021), in a study for the Bank for International Settlements (BIS), also point out that liquidity provision in both decentralized and centralized crypto markets shows a pronounced "decentralization illusion" in which supply is highly concentrated among a few participants [20]; this independent observation from an authoritative institution supports, in direction, the judgment that "market makers are highly concentrated," although it does not provide a metric fully comparable to the "70% order of magnitude" above. Whatever the specific number, this high degree of concentration brings, on one hand, economies of scale that let the market enjoy narrower bid-ask spreads and greater depth; on the other hand, it plants the seeds of systemic risk: if these top institutions simultaneously withdraw because of an extreme market event, a technical failure, or regulatory pressure, market liquidity will face a sharp collapse.

The business model of top crypto market makers is far more diversified than "pure market making." They typically play multiple roles at once—liquidity provider, cross-market arbitrageur, and over-the-counter (OTC) counterparty—and market-making income is only part of their composite profit matrix. This business fusion has important analytical implications: it may explain why certain market makers are willing to keep quoting even when their spread revenue on a particular pair is negative—some top institutions may obtain compensating revenue from cross-market arbitrage opportunities associated with their market-making inventory. The scale of this compensation is hard to separate out in public data, so the explanation above should be regarded as an explanatory hypothesis that helps make sense of market makers' quoting and retreat behavior, rather than an established fact.

Below Tier 1 are numerous regional Tier 2 market makers or those focused on specific asset classes. They may have a competitive advantage on only one or two exchanges, or focus on market-making the tokens of a particular "ecosystem." Their strategies are more flexible, and they may offer more competitive quotes than Tier 1 institutions in certain niche markets. However, in terms of capital scale and technological coverage, they lag noticeably behind the top institutions.

The final layer consists of the proprietary market-making desks of exchanges themselves. Some CEXs run in-house market-making teams to provide basic liquidity for certain pairs on their platforms (especially newly listed or thinly traded pairs). Although this practice can ensure the basic functioning of the market to some degree, it also raises serious concerns about conflicts of interest. As the maker and enforcer of market rules, an exchange that simultaneously participates directly in market trading gains an unfair advantage over external market makers in information access (such as seeing all users' order flow) and rule enforcement (such as the liquidation mechanism). From a regulatory perspective, the severity of this problem is rising: in traditional finance, exchange proprietary market making is subject to strict information-barrier requirements and independent audit oversight, whereas the widespread absence of these institutional protections in crypto markets is not only a governance challenge but may also translate into direct legal risk as global crypto-regulatory frameworks tighten. We will explore this governance challenge in greater depth in Chapter 21.

Overall, the market-making ecosystem for CEX perpetual futures exhibits a clear pyramid structure: the three to five Tier 1 institutions at the top control on the order of 70% of the order-book depth on mainstream pairs (an institutional estimate, with the evidence grade as noted above); the dozens of Tier 2 market makers in the middle seek differentiated survival space in niche markets and emerging assets; and at the bottom, exchanges' own proprietary market-making desks fill the gaps in the thinnest corners of liquidity. This highly concentrated pattern is the product of both a technology race and capital barriers: in an industry whose profit space is continuously compressed, only top participants with economies of scale can sustain a positive profit equation. Yet this concentration also means that the entire market's liquidity supply depends heavily on the health and operational continuity of a few institutions—a structural fragility discussed further in Section 19.7.2.

### 19.5.2 The quoting strategy framework

A professional market maker's operation on a CEX CLOB is not simple quoting but a precise, dynamically adjusted strategic framework. The core of this framework can be reduced to the real-time optimization of three key parameters: the width, depth, and skew of quotes. Together, these three parameters determine a market maker's risk exposure and profit potential, and they are continuously adjusted dynamically in a fast-changing market environment. Their mathematical skeleton is precisely the optimal market-making framework of Avellaneda and Stoikov (2008) cited earlier in Section 19.3.1 [13]: a market maker sets bid and ask quotes symmetrically around the reservation price and adjusts the width and skew of quotes in real time according to the inventory level, price variance, and degree of risk aversion ($\delta \propto \gamma\sigma^2 Q$); the three-parameter optimization discussed in this section can be viewed as an engineering implementation of that framework in the perpetual-futures context.

Width (the bid-ask spread) is a market maker's core source of profit as well as its foremost tool for risk compensation. Setting the spread is, in essence, a market maker's judgment and pricing of the current market's "toxicity" level. When the market is calm, the market maker faces low adverse-selection risk and tends to narrow the spread to compete for order flow. Conversely, when the market swings sharply or toxic flow increases, the market maker rapidly widens the spread, raising the cost for traders to obtain liquidity so as to compensate for the losses it may suffer from information asymmetry. Observing the dynamic changes in a pair's bid-ask spread is therefore an effective indicator of its current liquidity quality and market risk level.

Depth reflects the total amount of inventory risk a market maker is willing to bear at a given moment. Providing greater market depth means the market maker is willing to absorb larger orders when the price moves slightly, which usually attracts large traders; but greater depth also means that, once the price moves unfavorably, the market maker accumulates more risk exposure. The setting of depth is therefore a trade-off between risk and return. The following figure shows a real-time depth profile of the BTC perpetual-futures order book—that is, how cumulative depth on the bid and ask sides varies with distance from the mid-price: cumulative depth climbs rapidly as price levels extend outward, and the distributions on the two sides are not perfectly symmetric (see the figure for specific values). This asymmetry reflects market makers' implicit view of market direction and their current inventory state.

![Figure 19-10](./images/fig-19-10-en.png)

**Figure 19-10.** Depth profile of the BTC perpetual-futures order book: cumulative depth distribution on the bid and ask sides (Data source: real-time order-book data for Binance BTCUSDT perpetual futures, March 2025; this figure is an instantaneous snapshot profile of the inner order book, covering roughly ±9 bps around the mid-price)

In a high-volatility or high-uncertainty environment, a market maker "withdraws depth"—reducing the resting order size at each price level to control potential losses. This sharp withdrawal of depth means that a market order of the same size will cause a price impact several times greater than in normal times, further intensifying market volatility. Depth withdrawal is the rational product of a market maker's profit equation under extreme conditions, but its collective effect—liquidity vanishing precisely when it is most needed—constitutes a core fragility of market microstructure.

Skew describes the asymmetry of a market maker's quotes on the bid and ask sides. A perfectly neutral market maker would post quotes symmetrically around the market mid-price. In reality, however, a market maker's quotes are almost always skewed. Skew serves two main purposes: inventory management and directional betting. When a market maker has accumulated net long (or net short) inventory through successive fills, it skews its quotes to incentivize the market to trade in the opposite direction, thereby returning inventory to neutral. For example, if it holds long inventory, the market maker lowers its bid (dampening buy-side fills so that it accumulates less) and lowers its ask (making its sell orders fill more readily so that it offloads more), attracting buyers to "take away" its inventory. In addition, some more aggressive "statistical market-making" strategies actively skew quotes based on predictions of short-term price movements, trying to capture a small directional gain while providing liquidity. For example, if a model predicts a high probability of a short-term price rise, the market maker will quote more aggressively on the bid side and relatively conservatively on the ask side.

The dynamic adjustment of these three core parameters is driven by a series of complex triggers. These triggers include, but are not limited to: sharp changes in market volatility, jumps in toxic flow detected by indicators such as VPIN, the market maker's own inventory level exceeding a preset threshold, the approach of a funding-settlement window, abnormal spreads relative to other exchanges, and large liquidations about to be triggered as detected by on-chain data monitoring. The distinctive mechanisms of perpetual futures, such as the funding rate, add extra dimensions to this adjustment framework. For example, when a high positive funding rate is expected, the market maker has an incentive to reduce its net long position, or even build a net short position, before settlement, in order to collect the funding rate as extra income. This kind of strategic adjustment based on institutional rules is a complex game that does not exist in traditional market making.

### 19.5.3 Cross-exchange market making and position management

For top professional market makers, the scope of business is by no means confined to a single exchange. To maximize capital efficiency and capture broader order flow, cross-exchange market making is an inevitable choice. Yet this operation, while magnifying opportunities, also brings higher-dimensional complexity and challenges, especially in position management. Market makers must precisely manage risk and return across multiple platforms simultaneously.

First, multi-exchange market making faces a severe "margin fragmentation" problem, a concept in line with the capital-efficiency constraints discussed in Chapter 17. A market maker must deposit sufficient margin on every exchange where it is active to support its market-making activity and potential risk exposure. This means that an institution making markets on N exchanges has a total capital requirement far greater than N times that of making markets on a single exchange, because the margin pools of the various exchanges are mutually isolated and cannot be shared in real time. This fragmentation of capital severely tests a market maker's capital-management capabilities.

The more central challenge lies in the cross-exchange hedging of inventory. In theory, if a market maker accumulates a net long of 100 BTC on exchange A through fills while accumulating a net short of 100 BTC on exchange B, its total position appears neutral. In practice, however, this is a dangerous "pseudo-neutrality," because these two positions are each exposed to the counterparty risk, platform risk, and liquidation risk of two different exchanges. Once an extreme move on exchange A causes its long position to be forcibly liquidated while the price on exchange B does not move in sync, the market maker instantly shifts from a seemingly balanced state to bearing enormous one-sided risk. A market maker must therefore build a global risk-management system that monitors and aggregates net positions across all exchanges in real time and actively hedges, rather than relying merely on the "natural" offset of positions across different exchanges.

Cross-exchange market making also faces a risk dimension often overlooked in theoretical analysis but central in practice: counterparty risk. The margin a market maker deposits at each CEX essentially constitutes an unsecured credit exposure to that exchange. The collapse of FTX in November 2022 already proved, in the most brutal way, that even a top-three global exchange can go bankrupt within days, leaving a market maker unable to recover all its deposited funds. For a market maker operating on five to ten exchanges simultaneously, counterparty-risk management (including exposure caps for a single exchange, a dynamic capital-allocation strategy based on exchanges' financial-health indicators, and an early-warning capital-withdrawal mechanism triggered by abnormal withdrawal delays) constitutes an indispensable part of its risk-management system. A market maker must continuously seek a balance between "concentrating more capital on the most liquid exchanges to maximize capital efficiency" and "dispersing capital across multiple exchanges to reduce single-counterparty risk."

The active-hedging process itself is also fraught with risk. When a market maker's inventory on one exchange exceeds a threshold, it must execute an offsetting trade on another highly liquid exchange (or in the spot market) to balance risk. This process involves a critical "hedging latency window." Under normal market conditions, the round-trip cross-exchange, cross-region hedging latency (even for top market makers that have already deployed servers nearby) is usually on the order of 10–50 milliseconds, and this latency is itself manageable. The 10–50 milliseconds here refers to the end-to-end round-trip latency across exchanges and regions, not the internal latency of co-location in an exchange's data center—the latter is usually at the microsecond to sub-millisecond level. However, in extreme conditions, cross-exchange hedging latency may expand sharply to the seconds level because of the target exchange's application programming interface (API) rate limiting and system congestion. More critically, in extreme conditions the true bottleneck for hedging failure is often not network latency itself but the simultaneous drying-up of liquidity on the target exchange: even if the hedging order arrives instantly, it may incur slippage costs far beyond expectations because the counterparty market's order book has likewise thinned. The value of low-latency technology therefore lies not only in compressing the hedging window under normal conditions but, more importantly, in completing the hedge ahead of competitors in the first few seconds of an extreme move, when the counterparty market's depth has not yet been exhausted by other market makers' hedging demand.

In practice, the boundary between cross-exchange market making and the cross-exchange arbitrage analyzed in Chapter 16 is not clear. Many top institutions in fact play the dual roles of market maker and arbitrageur simultaneously. Their strategy can be described as "hedging the inventory generated by passive market making on one exchange through active trading on another exchange where the price is more favorable." This hedging trade, from the perspective of inventory management, is part of "market making," but from the perspective of exploiting the cross-exchange spread, it also constitutes a typical cross-exchange arbitrage. This fusion of market making and arbitrage allows large institutions to build a more complex profit matrix, and their integrated return and risk-management capabilities far exceed those of participants able to execute only a single strategy.

### 19.5.4 The symbiotic relationship between market makers and exchanges

Market makers and centralized exchanges share a deep and complex symbiotic relationship. Exchanges need market makers to provide liquidity, which is the core element for attracting and retaining traders; and market makers need the exchange platform to implement their strategies and earn profits. This interdependence gives rise to a complex set of mechanisms built around incentives and games, the most influential of which are the institutional arrangements exchanges provide to market makers, including fee rebates, API priority, information advantages, and margin benefits.

Among the various incentives, market-maker fee rebates have the broadest applicability and the most direct impact on the profit model. In most exchanges' fee structures, a taker who actively takes liquidity pays a fee, while a maker who provides liquidity pays a lower fee or even receives a negative fee, i.e., a rebate. This means that each time a market maker successfully provides liquidity, it not only pays nothing but also receives income from the exchange. This rebate is an important component of a market maker's profit, especially on mainstream pairs with extremely narrow spreads, where rebate income may even exceed the spread itself. Exchanges design tiered rebate structures to incentivize market makers to provide more liquidity. For example, Binance's USDⓈ-margined futures market maker program rates market makers according to indicators such as their weekly market-making volume share and quote quality; the higher the tier, the higher the rebate ratio the market maker enjoys. According to the fee arrangements of Binance's official market maker program, the maker rebate at the best tier is usually in the range of -0.5 to -2.5 bps (i.e., -0.005% to -0.025%), and during specific limited-time promotions the maker fee applicable to top market makers can go even lower (in some promotional periods the maker fee even drops to 0.0000% or offers a higher rebate) [21]. Even at the regular-tier rebate level, for top market makers with daily volume on the order of billions of dollars, this still constitutes an extremely substantial source of income. The rebate figures above come from the exchange's own announcements, and the specific values for promotional and regular tiers change as the exchange adjusts its policies, so readers should treat them as order-of-magnitude references.

![Figure 19-11](./images/fig-19-11-en.png)

**Figure 19-11.** Binance USDⓈ-margined futures maker/taker fee structure (Data source: Binance official market maker program announcement, March 2025; the figure shows the regular best tier, while during limited-time promotions the maker rebate can go deeper, on the order of -2.5 bps, which is not shown in the figure)

The figure above shows the tiered fee structure of Binance's market maker program: from the basic VIP0 tier to the highest MM top (invitation-only) tier, the maker fee decreases step by step to a negative value (i.e., a rebate), and the taker fee also declines as the tier rises. The economic meaning of this tiered design is that the exchange reserves the most favorable rebates for the market makers with the highest volume and quote quality, forming a positive-incentive "flywheel effect": higher rebates attract deeper liquidity, deeper liquidity attracts more traders, and greater trading volume in turn further consolidates the market maker's rating and rebate level. However, this also intensifies the competitive asymmetry between large market makers and small participants.

Broadening the view to the whole industry, the following figure compares the fee structures of six major perpetual-futures trading platforms (including CEXs and two on-chain DEXs). At the base tier (VIP0), fee differences across platforms are relatively small, with maker fees generally around 0.02% (i.e., 2 bps); but at the best tier of the market maker programs, the differences widen substantially: most mainstream platforms offer negative-fee (rebate) incentives to top market makers, with the most favorable maker rebate reaching the order of -2.5 bps (see the figure for each platform's base and best-tier fees). This fee competition directly affects market makers' platform selection and capital-allocation strategies.

![Figure 19-12](./images/fig-19-12-en.png)

**Figure 19-12.** Comparison of fee structures across major perpetual-futures trading platforms (including CEXs and on-chain DEXs): base tier vs. market-maker best tier (Data source: each exchange's official fee pages and market maker program announcements, March 2025; the six subjects in the figure include two on-chain DEXs, dYdX and Hyperliquid; the market-maker best maker rate shown in the figure stops at -0.55 bps, while the -2.5 bps order of magnitude mentioned in the text is the limited-time-promotion end and is not shown in the figure)

Beyond direct fee incentives, priority technical access is likewise an important source of competitive advantage. Exchanges provide top market makers with priority service on the API. This may include faster network connections, higher API request-rate limits, and lower data-push latency. In the world of high-frequency trading, where every fraction of a second counts, even a lead of a few milliseconds can determine the life or death of a strategy. Obtaining API priority means a market maker can receive market data faster than ordinary traders and submit and cancel orders faster, thereby managing risk and capturing opportunities more effectively.

Information advantage is a more hidden but potentially more valuable "privilege." Some exchanges may provide their core market makers with finer-grained order-flow data, such as anonymized statistical information about large orders or the behavior of specific types of traders. This information can help market makers judge market sentiment more accurately and identify potentially toxic order flow, thereby optimizing their pricing and risk models. However, this information advantage is also precisely the most controversial point in the symbiotic relationship between exchanges and market makers, because it not only touches the bottom line of market fairness but may also constitute a legal risk. In several jurisdictions, providing nonpublic order-flow data to specific market participants may be deemed market abuse. As the EU's Markets in Crypto-Assets (MiCA) Regulation and the crypto-asset regulatory frameworks of other major economies are gradually implemented, the compliance-scrutiny risk facing such information-sharing arrangements is rising systematically.

Finally, margin benefits are also a common incentive. Exchanges may offer reputable, well-capitalized market makers lower margin requirements or higher leverage caps. This allows market makers to operate at higher capital efficiency, supporting larger-scale market-making activity with the same amount of capital.

However, this symbiotic relationship is not always harmonious; it is full of ongoing games and tension. An exchange's goal is to maximize the platform's trading volume and liquidity, so it always wants market makers to provide the narrowest possible spreads and the deepest possible market depth. A market maker's goal is to maximize its own profit, so it must ensure that its risk is controllable and its business profitable while providing liquidity. There is an inherent contradiction between these two. An exchange's rebate and incentive policies are, in essence, a search for an equilibrium in this game: attracting market makers to provide sufficient liquidity without letting the exchange's own operating costs run too high. When the market environment deteriorates and market makers' risk rises sharply, they tend to widen spreads and withdraw depth to protect themselves—which is precisely what the exchange least wants to see. This fundamental conflict of interest is the key to understanding why liquidity contracts sharply in moments of crisis.

### 19.5.5 How perpetual-futures market making differs from traditional market making

To understand just how demanding perpetual-futures market making on a CEX CLOB is, the best approach is to compare it directly with more traditional spot market making and delivery-futures market making. Although all three follow the basic logic of "earning the spread by bearing risk," the distinctive institutional design of perpetual futures makes it fundamentally different from the other two in risk dimensions and operational complexity. We can dissect the distinctiveness of this business along six core dimensions. The following table first summarizes the differences among the three types of market-making businesses across these six dimensions, and each is then elaborated in turn.

| Dimension of comparison | Spot market making | Delivery-futures market making | Perpetual-futures market making |
| :--- | :--- | :--- | :--- |
| Inventory time horizon | Can be held indefinitely, but unleveraged with linear risk | Has an expiration date; the basis necessarily converges and inventory is cleared to zero at maturity | No expiration date; inventory risk is indefinite and must be actively hedged |
| Certainty of holding cost | Fiat funding interest, slow and predictable | Day-by-day basis decay, a largely determined path | Dominated by the randomly fluctuating funding rate; direction can reverse within hours |
| Liquidation risk | No leverage, no liquidation risk | Leveraged, but with an expiration constraint plus institutional dominance, relatively controllable | High leverage plus no expiration plus many retail traders; liquidation-cascade risk is extremely high |
| Source of extreme toxic flow | Primarily traditional informed trading | Primarily traditional informed trading | Beyond informed trading, uniquely subject to liquidation cascades—a positive-feedback toxicity |
| Operational continuity | Usually has market-closure windows for calibration/maintenance | Usually has market-closure windows for calibration/maintenance | 24/7 without interruption; systems and teams must operate around the clock |
| Complexity of the information environment | Stricter disclosure regulation in mature jurisdictions, relatively transparent | Stricter disclosure regulation in mature jurisdictions, relatively transparent | Inconsistent disclosure standards, higher manipulation/insider risk |

**Table 19-3.** A six-dimensional comparison of spot, delivery-futures, and perpetual-futures market making (Data source: compiled by the author)

First, the inventory time horizon. The inventory a spot market maker holds can, in theory, be held indefinitely, but because there is no leverage, its risk exposure is linear. A delivery-futures market maker enjoys an important institutional safeguard: the expiration date. However far the inventory deviates along the way, the basis necessarily converges to zero at expiration, all positions are forcibly delivered or settled, and risk is confined to a finite time window. A perpetual-futures market maker, by contrast, faces the most severe combination: it must, like a spot market maker, confront indefinite inventory-holding risk, while also, like a futures market maker, bear the risk-amplifying effect of high leverage. The absence of an expiration date as a natural settlement mechanism means that inventory risk can only be continuously managed through active hedging.

Second, the certainty of holding cost. The holding cost of spot market making is mainly fiat funding interest, which changes slowly and predictably. The holding cost of delivery-futures market making manifests as the day-by-day decay of the basis, whose path is also largely determined. However, the inventory holding cost of a perpetual-futures market maker is dominated by a highly random and volatile variable, the funding rate (whose randomness and magnitude were analyzed in detail in Section 19.3). This random "implicit tax" or "implicit subsidy" brings enormous uncertainty to a market maker's risk pricing and inventory management.

Third, liquidation risk. Spot trading, having no leverage, has no liquidation risk. Delivery futures, though leveraged, have relatively controllable large-scale liquidation risk because of the expiration-date constraint and the participation of relatively more mature institutions. The perpetual-futures market, with its high leverage, absence of an expiration date, and large numbers of retail traders with weaker risk-management ability, together constitutes a forced-liquidation environment extremely prone to triggering chain reactions, pushing liquidation risk to the extreme.

Fourth, the source of extreme toxic flow. The toxic flow faced by spot and delivery-futures markets comes mainly from traditional informed trading. The perpetual-futures market, on top of this, adds a uniquely destructive form of "extreme toxic flow": the liquidation cascade. This positive feedback loop, triggered by the market's own decline, can generate overwhelming one-directional order flow several times the normal level in a short time, posing a significant threat to the safety of a market maker's positions.

Fifth, operational continuity. Most traditional financial markets have closing and holiday hours, which provide market makers with necessary adjustment windows for recalibrating models, managing risk, and maintaining systems. The 24/7 uninterrupted trading feature of crypto markets requires that perpetual-futures market makers' systems and teams also operate around the clock, which imposes significant additional constraints on operating costs, technical reliability, and human-resource allocation.

Sixth, the complexity of the information environment. Spot and futures markets, especially in mature jurisdictions, are subject to relatively strict information-disclosure regulation, and their information environment is relatively transparent. Crypto markets, especially the perpetual-futures space, have inconsistent disclosure standards and higher risks of market manipulation and insider trading, which intensifies the adverse-selection problem market makers face.

In summary, if we rate the difficulty of these three types of market-making businesses, spot market making is at a higher difficulty level, delivery-futures market making is at an expert level, and perpetual-futures market making is at an extreme difficulty level. It pushes the risk of the market-making business to the limit along virtually every dimension. In theory, because of the additional dimension of institutional cost, the bid-ask spread of perpetual futures (measured in basis points) tends to be wider than that of spot and delivery futures under comparable liquidity conditions; this extra spread is precisely the risk premium the market pays for market makers bearing these three superimposed (adverse-selection, inventory, and institutional) ultra-high costs. Rigorously verifying the judgment that "perpetual spreads are systematically wider than spot/delivery spreads" would require paired typical-spread data for the same asset across the three markets, which this chapter does not provide; the judgment is therefore proposed as a directional theoretical prediction rather than an established conclusion.

## 19.6 Liquidity provision on DEXs

From centralized to decentralized exchanges, the paradigm of liquidity provision for perpetual futures shifts fundamentally. This book takes order-book-based perpetual futures as its core object of study, and the corresponding forms in the DEX domain are the on-chain central limit order book paradigm and the hybrid market-making vault paradigm derived from it. The market dynamics of decentralized exchange adoption itself have been studied specifically (Capponi & Jia, 2021 [22]); this section instead focuses on the different answers these two perpetual-market-making paradigms give to three core questions—"who bears the risk," "how liquidity is priced," and "how toxic order flow is handled"—and constructs a unified comparative framework.

### 19.6.1 The on-chain CLOB paradigm

The on-chain CLOB paradigm takes a direct path: replicating the mature central limit order book model of CEXs onto the blockchain. Protocols represented by dYdX v4 and Hyperliquid achieve high-performance on-chain order matching and settlement by building dedicated application chains, making it possible for professional market makers to operate on-chain.

On dYdX v4's Cosmos application chain or Hyperliquid's proprietary high-performance HyperBFT chain, market makers can, just as on Binance or Bybit, connect via API, deploy complex market-making strategies, submit and cancel limit orders in real time, and actively manage their inventory risk. This model returns the responsibility for liquidity provision to professionals, who provide dense liquidity on both sides of the order book and earn the bid-ask spread by virtue of their capital, technology, and strategy advantages. This fundamentally changes the risk-bearer: risk is no longer borne collectively by a passive group of LPs but by individual professional market makers pursuing excess returns.

However, deploying a CLOB on-chain is not a simple technical migration; the inherent properties of the blockchain bring distinctive challenges and opportunities. First, transparency has a dual effect. The full state of the on-chain order book is publicly visible to all, which significantly increases the market's information symmetry but also makes a market maker's strategy easier to "reverse-engineer" and raises the risk of MEV attacks. For example, an MEV bot can monitor a market maker's order-cancellation behavior and use the information advantage to front-run and extract value. Second, latency remains an issue. Although Hyperliquid's block time has been compressed to the 200-millisecond level (and its optimistic confirmation mechanism may make the user-experienced latency lower than the block finalization time), this still lags significantly behind a CEX matching engine's internal processing latency (microseconds) and external/end-to-end price-update latency (milliseconds). Longer latency means a market maker's quotes are exposed to adverse-selection risk for a longer time, placing higher demands on its risk-management capabilities.

Despite the above challenges, the on-chain CLOB paradigm still demonstrates strong competitiveness thanks to its friendliness to professional market makers. Hyperliquid is a case in point: by virtue of its efficient application chain and market-maker incentives, it rose rapidly in 2024. Here it is necessary to clarify several share metrics that are often confused. First, the single-protocol metric: during the peak of the fourth quarter of 2024, the single protocol Hyperliquid at one point held about 55% of the DEX perpetual-futures market, and over the same period its daily volume and open interest reached roughly 10% and 15–20%, respectively, of the Binance perpetual-futures market. During this period, Hyperliquid's points-incentive campaign may have positively distorted its volume figures to some degree, and the organic-volume share after excluding points incentives may be lower than the 55% peak figure above. Second, the paradigm-aggregate metric: as of March 2025, all protocols using the CLOB paradigm (including Hyperliquid, dYdX, and others) together accounted for about 93% of DEX perpetual-futures volume. Third, the absolute-volume metric: over the same period, Hyperliquid led DEX perpetual protocols by a wide margin with weekly volume of about $53 billion. The "55%" and "93%" above are not measurements of the same object at the same point in time—the former is the share of a single protocol at its Q4 2024 peak, distorted by points incentives, while the latter is the aggregate share of the CLOB paradigm in March 2025; the two differ in both metric and time point and cannot be directly compared or substituted for each other. Combining these three metrics yields a robust judgment: in the specific domain of perpetual futures, the CLOB paradigm dominated by professional market makers has taken a leading position in the competition on efficiency; but this judgment applies mainly to high-liquidity mainstream-asset scenarios, while in emerging markets or low-barrier scenarios, other paradigms such as market-making vaults still retain their niche (see Section 19.6.3).

![Figure 19-13](./images/fig-19-13-en.png)

**Figure 19-13.** Comparison of DEX perpetual-futures protocol trading volumes and paradigm shares (Data source: DeFiLlama Derivatives Protocol Data, March 2025)

### 19.6.2 The market-making vault paradigm

As a point of comparison with the on-chain CLOB, the market-making vault paradigm represents a different path for liquidity provision. Represented by Hyperliquid's HLP vault, the core feature of this paradigm is "the protocol is the market maker": ordinary users deposit funds into a vault managed by the protocol, and the funds no longer passively serve as counterparty (as in GMX's GLP model) but are used by the protocol's algorithm to execute active market-making strategies on the CLOB, including dynamically adjusting quotes, managing positions, and widening spreads as volatility intensifies. This grants the LP community the risk-management initiative that previously only professional market makers possessed.

The core trade-off of this model is that it offers a lower barrier to entry at the cost of limited LP autonomy and a reliance on trust in the algorithm and governance. An LP's returns depend heavily on the quality of the protocol's market-making algorithm: once the algorithm has a flaw or is exploited by a large holder, the vault may suffer major losses. For example, in March 2025, the HLP vault recorded a significant drawdown due to a squeeze-style operation via a single large position (the so-called JELLY incident), highlighting the fragility of algorithmic market making in the face of targeted manipulation. In addition, the ownership of the decision-making authority over the market-making strategy's key parameters (spread, depth, and risk-control thresholds) also constitutes a new governance challenge.

### 19.6.3 Comparing the two paradigms

Placing the two DEX liquidity-provision paradigms above within a unified framework, we can clearly see their different trade-offs across core dimensions. This is not a simple judgment of which is superior but a design choice about how decentralized markets construct their liquidity foundation. The following table first summarizes the trade-offs of the two paradigms across five core dimensions, and the discussion then unfolds.

| Dimension of comparison | On-chain CLOB paradigm | Market-making vault paradigm |
| :--- | :--- | :--- |
| Barrier to entry | High (requires technology, capital, and strategy) | Low (ordinary users need only deposit funds) |
| LP autonomy | Fully active (independent quoting and risk control) | Limited (executed on their behalf by the protocol's algorithm) |
| Toxicity exposure | Borne and managed directly by individual professional market makers | Centrally managed by the protocol's algorithm; individual LPs do not respond directly |
| Capital efficiency | Highest | Medium, dependent on algorithm quality |
| Trust dependency | Mainly on one's own strategy and the exchange's matching engine | Additionally on the protocol's algorithm and governance (algorithm risk plus governance risk) |

**Table 19-4.** A comparison of the on-chain CLOB paradigm and the market-making vault paradigm across core dimensions (Data source: compiled by the author)

From the core trade-off dimensions—"barrier to entry," "LP autonomy," and "toxicity exposure"—the two paradigms occupy different niches. The on-chain CLOB paradigm offers professional market makers with technology, capital, and strategy full autonomy and the highest capital efficiency, but its barrier to entry is also the highest. The market-making vault paradigm is a compromise: it provides LPs with a degree of risk-management capability through a protocol-managed algorithm, and its barrier to entry is lower, but LP autonomy is limited, and it introduces a trust dependency on the algorithm and governance.

This trade-off is also directly reflected in the cost structures of the different paradigms. For professional market makers on an on-chain CLOB, the cost structure is highly similar to that of CEX market makers: adverse-selection cost and inventory-risk cost dominate, MEV-attack toxicity constitutes an additional on-chain-specific cost item, and they must also bear costly technical infrastructure (low-latency nodes, strategy engines) and ongoing operating costs. This cost structure means that only professional institutions with ample capital and technical capability can profit sustainably. For the market-making vault paradigm, by contrast, the cost structure LPs face is more balanced: adverse-selection cost and inventory risk are centrally managed by the protocol's algorithm, and individual LPs need not respond directly; but the core risk LPs bear shifts to algorithm risk (systematic losses caused by flawed strategy design or improper parameter settings) and governance risk (the opacity of the decision-making authority over strategy adjustments). An LP's returns therefore depend heavily on a variable it cannot directly control: the quality of the algorithm.

Ultimately, the DEX perpetual-futures market will settle into an ecosystem in which the two paradigms—on-chain CLOB and market-making vault—coexist and evolve dynamically. High-liquidity mainstream-asset markets will be dominated by the most capital-efficient on-chain CLOB professional market makers, while emerging markets may rely more on market-making vaults to bootstrap initial liquidity. This still-ongoing paradigm contest will determine the trading experience and costs of future DEXs and leave a lasting mark on the form of on-chain financial infrastructure.

## 19.7 Competitive landscape and evolution

Liquidity provision in perpetual-futures markets is not a static picture but a dynamic ecosystem of competition and iteration. The preceding sections analyzed the cost structure, risk sources, and core strategies market makers face on CEXs and DEXs, but this concerns only the behavioral logic of individual market participants. This section extends the analytical dimension to the interactions among market participants and between participants and the institutional environment, examining how these interactions jointly shape the macro landscape of liquidity provision and its direction of evolution. This evolution unfolds around a complex game of technology, capital, information, incentives, and risk, exhibiting three major trends—concentration, algorithmic intelligence, and protocolization—which, while improving market efficiency, also breed new systemic risks.

### 19.7.1 The three-dimensional competition in technology, capital, and information

Competition among perpetual-futures market makers is a continually escalating technology race. The core driver of this competition stems from the fundamental source of market-making profit: capturing tiny spreads in the crevice between information asymmetry and inventory risk. Any advantage that can reduce latency, optimize models, or expand capital scale translates directly into higher profit margins and stronger survivability. This competition unfolds mainly along three dimensions: technology, capital, and information.

The technology race is most directly expressed in the extreme pursuit of "speed." In market microstructure, a speed advantage means being able to react earlier to market signals—whether updating quotes ahead of competitors or being the first to complete an arbitrage when a cross-market spread appears. This gives rise to the classic "latency race" of high-frequency trading. Market makers invest heavily in server colocation, physically deploying their trading servers in the same data center as the exchange's matching engine to reduce network latency to the microsecond level. However, as the technology of top market makers converges, the excess profit obtained from a pure speed advantage is quickly compressed. The focus of competition then shifts to a more complex arena: algorithms and models. A superior risk model can quantify adverse selection more precisely, a smarter inventory-management algorithm can minimize risk exposure while maximizing volume, and a more sophisticated trading-signal-generation system tries to mine faint predictive information from massive data.

The capital race is expressed as the powerful barrier of economies of scale. A market maker with more ample capital can withstand greater inventory risk, thereby providing deeper liquidity on the order book and capturing more trading flow. At the same time, it can make markets on dozens or even hundreds of pairs across multiple exchanges simultaneously, smoothing overall P&L through diversification. More importantly, ample capital provides the necessary risk buffer for a market maker to keep operating in extreme conditions. When the market swings violently and a liquidation cascade erupts, undercapitalized market makers are forced to liquidate their positions and exit the market, while well-capitalized market makers may find opportunities for high returns in this "liquidity vacuum." This cumulative-advantage effect makes the market-making industry structurally prone to concentration.

The information race is harder to observe directly than the previous two dimensions, but its impact on long-term competitive advantage is equally significant. Beyond public market data, market makers seek a unique "information advantage." This can come from a deep parsing of on-chain data—for example, anticipating market sentiment by monitoring anomalies in whale addresses or the fund flows of DeFi protocols. It can also come from the symbiotic relationship established with exchanges, obtaining more favorable trading fees, higher API rate limits, and, in some cases, finer-grained order-flow data through a market maker program. The accumulation of this information advantage further consolidates the competitive barriers of top institutions.

Competition along these three dimensions leads to one inevitable result: profit compression and market concentration. As competition intensifies, the bid-ask spreads of mainstream pairs are continuously pushed to the limit, and the profit space of market making becomes ever thinner. Market makers slightly behind in technology, capital, or information see their profits quickly eroded and are ultimately forced to exit the market. This causes the share of liquidity provision to concentrate increasingly in a few global top market-making institutions. Meanwhile, a new competitive dimension is emerging: as crypto ETFs are approved in several major markets and regulatory frameworks gradually clarify, traditional-finance market-making giants such as Citadel Securities and Jane Street are progressively expanding their footprint in crypto market making. The capital scale, cross-market experience, and compliance capabilities these institutions bring far exceed those of crypto-native market makers. The scale and pace of traditional giants' entry into crypto market making fluctuate with the regulatory environment (for example, some institutions' crypto exposure has at times contracted amid regulatory uncertainty), but the overall direction is that their entry may fundamentally change the competitive landscape: future competition will no longer be merely a contest among crypto-native institutions but a new-dimensional game between crypto-native market makers and traditional-finance giants.

By virtue of their technology and capital advantages, top market makers can continuously offer narrower bid-ask spreads than other competitors, and this gap widens over time. At the same time, as profit space is compressed, large numbers of small and mid-sized market makers are squeezed out of the market, the total number of active market makers declines year by year, and the market share occupied by top institutions correspondingly climbs steadily. This concentration trend constitutes the root of the systemic risk discussed in the next section.

### 19.7.2 Concentration and systemic risk

The concentration of the market-making industry has a dual effect. On the positive side, it brings significant efficiency gains. Top market makers, leveraging their economies of scale, can provide the market with more stable, deeper, and lower-cost liquidity. For ordinary traders, this means smaller trading slippage and better execution prices. From this angle, concentration is an optimization result under market competition and natural selection, handing the specialized work of liquidity provision to the most efficient producers.

However, this efficiency gain comes at a steep cost: the accumulation of systemic risk. When the overwhelming majority of a market's liquidity depends on a few highly homogeneous participants, the entire market becomes extraordinarily fragile. This concern is not unique to crypto markets; traditional market-microstructure research long ago provided an empirical foundation for it: Comerton-Forde et al. (2010), using inventory and P&L data of specialist market makers on the New York Stock Exchange, demonstrate that market makers' inventory levels and cumulative returns can significantly explain the time variation in market liquidity—when market makers' inventory is under stress or they suffer losses, bid-ask spreads widen nonlinearly and liquidity contracts accordingly [23]. This transmission mechanism from "market makers' financial condition to liquidity provision" provides a direct theoretical and empirical anchor for this section's analysis of the fragility of a concentrated market. In the crypto perpetual-futures market specifically, this risk is manifested mainly at two levels: single-point-of-failure risk and behavioral-correlation risk.

Single-point-of-failure risk is the danger that the sudden "going offline" of any top market maker can inflict a disproportionately large shock on the market. Such an offline event may stem from various factors: a technical failure, an operational error (such as a mistaken risk-parameter setting), a financial crisis (such as huge losses on its own assets), or even a regulatory raid. If a market maker holding a substantial share of liquidity (such as one of the top three to five institutions mentioned earlier) suddenly withdraws all its quotes, the order book will thin instantly, bid-ask spreads will widen sharply, and the market's price-discovery capacity will be severely impaired. In extreme cases, this could trigger a chain reaction, leading to broader liquidity evaporation and a price flash crash. An even more extreme scenario is that the market maker itself may be forcibly liquidated in a violent move: under a unified margin system, a huge loss on one pair may consume the market maker's available margin on other pairs, triggering a chain of forced liquidations across pairs. This passive, involuntary "going offline" is more destructive than an active retreat, because it both removes liquidity supply and directly injects additional selling pressure into the market.

Behavioral-correlation risk is more hidden and dangerous. Although top market makers are competitors in day-to-day business, their technology stacks, risk models, and decision logic are largely convergent. They all rely on similar data sources, follow similar microstructure theories, and are constrained by the same market institutions (such as the funding rate and liquidation mechanism). This means that when the market encounters extreme stress, their reactions are likely to be highly correlated. For example, in a sharp price decline, several market makers' risk models may simultaneously hit their maximum inventory limits or maximum loss thresholds, so they collectively decide to contract quotes or withdraw from the market. This "collective retreat" behavior is precisely the core microlevel mechanism of the "sharp liquidity contraction" phenomenon we will examine in depth in the next chapter. Rational decisions at the individual level produce a systemic negative externality at the collective level: at the very moments when the market's demand for liquidity is highest, the supply of liquidity contracts in sync.

However, the "collective retreat" narrative above needs an important qualification: market makers are not entirely homogeneous actors, and different types of participants behave differently in a crisis. This requires an observable demarcation criterion for the two scenarios—"collective-retreat-dominant" and "competitive-backfill-dominant"—so as to avoid ex post self-justifying explanations. We propose the number of heterogeneous participants and their capital buffers as the demarcation variables: when the number of heterogeneous participants "willing and able to provide liquidity" is ample and their risk-control models and data sources are mutually independent, the backfill effect dominates, and liquidity can be partly replenished during a crisis; conversely, when such heterogeneous participants are scarce, or their risk-control models trigger in sync because of high homogeneity, the retreat effect dominates, and liquidity dries up in unison. The specific thresholds corresponding to this criterion await empirical calibration, so this section proposes it as a theoretical inference. Under this criterion, several structural regularities can be observed: Tier 1 institutions, because of their thicker capital buffers and more mature risk-control systems, usually trigger their retreat thresholds later than Tier 2 market makers, but once they retreat, their impact on market depth is also greater. Strategy heterogeneity matters equally: institutions running pure market-making strategies tend to contract comprehensively in extreme conditions, whereas institutions running a mix of market-making and arbitrage strategies may increase cross-market arbitrage even as they withdraw from market-making activity, and the latter's behavior objectively provides the market with some resilience. In mild volatility, the number of heterogeneous participants is usually ample, and the backfill effect can effectively mitigate concentration risk; but in extreme tail events, when all institutions' risk-control models trigger in sync because they use similar risk measures and the same data sources, heterogeneity plunges, the backfill effect fails sharply, and the destructive power of behavioral correlation is fully revealed. The true critical point of systemic risk therefore lies not in the number of market makers in daily operation but in whether heterogeneous participants "willing and able to provide liquidity" still exist in extreme conditions.

The systemic risk brought by this concentration has precedents in traditional financial markets and has drawn close attention from regulators. But in the cryptocurrency market, this problem is especially acute, because the regulatory framework here is still immature and lacks a crisis-response mechanism analogous to a "market maker of last resort" or central-bank liquidity support. Understanding and monitoring the concentration of liquidity provision, assessing the health of top market makers, and designing mechanisms that incentivize market makers to remain in a crisis therefore become one of the core challenges of maintaining market stability.

### 19.7.3 Incentive design and market-maker behavior

Market makers and exchanges are locked in an ongoing game. Exchanges want market makers to provide the narrowest possible spreads and the deepest possible liquidity to attract more traders and improve market quality. Market makers want the highest possible return while bearing the lowest possible risk. The equilibrium of this game is determined by the incentive mechanisms exchanges provide. These incentive designs directly affect market-maker behavior and are ultimately transmitted to the liquidity condition of the entire market.

On centralized exchanges, incentive mechanisms revolve mainly around the "market maker program." The basic components of this program—fee rebates, API priority, and margin benefits—were analyzed in detail in Section 19.5.4 and are not repeated here in structure; instead, we focus on how these incentives shape market makers' behavioral patterns. A core behavioral effect is that higher rebates incentivize market makers to quote narrower spreads, because even if the spread itself is unprofitable, they can still profit from the rebate. However, this incentive can also lead to the emergence of "phantom liquidity," in which market makers post large numbers of orders to obtain rebates but quickly cancel them when the market truly needs liquidity. This "phantom liquidity" problem reveals the inherent limitation of a static rebate mechanism: it incentivizes depth supply under normal conditions but cannot constrain the collective retreat in extreme conditions.

On decentralized exchanges, incentive design becomes more varied and complex. For protocols adopting the on-chain CLOB or market-making vault model, incentives for liquidity providers come mainly from a share of trading fees and protocol-token incentives, with the design goal of attracting professional market makers to provide deep liquidity continuously.

For DEXs adopting an on-chain order book, incentive design is closer to that of CEXs, likewise including trading-fee rebates and volume-based rewards. But because of the transparency of the on-chain environment, incentive design faces new challenges. For example, a protocol must design sophisticated algorithms to identify "effective" market-making behavior and prevent market makers from cheating for incentives through fake fills. Hyperliquid's points-reward system is one example; it tries to reward market makers that genuinely contribute liquidity more fairly by comprehensively assessing multiple dimensions such as resting-order time, distance from the mid-price, and volume.

Furthermore, some protocols are beginning to explore "protocol-controlled liquidity" or keeping ownership of liquidity in their own hands through instruments such as bonds. This model attempts to fundamentally solve the problem of market makers withdrawing in a crisis. However, it also fully transfers the complexity of risk and incentive design to the protocol itself. The protocol must play the role of a "super market maker," whose performance depends entirely on the algorithm design and risk-management capabilities of its internal team.

From a game-theoretic perspective, the optimal incentive design should achieve "incentive compatibility"—aligning market makers' pursuit of their own maximum self-interest with the protocol's goal of improving market liquidity and robustness. This requires protocol designers to deeply understand market makers' cost structure (the three costs) and the risks they face (the four types of toxic flow). A good incentive mechanism should encourage competition and compress spreads when the market is calm, while providing sufficient risk compensation for those market makers that choose to stay and continue providing liquidity when the market is turbulent. For example, one can design a dynamic rebate mechanism that automatically raises the rebate ratio for makers when market volatility rises sharply, thereby partly offsetting market makers' increased inventory risk and adverse-selection cost. Implementing such a dynamic incentive mechanism is a potential advantage of DeFi protocols relative to CEXs and a key direction for future liquidity-governance exploration.

### 19.7.4 Future directions of evolution

The evolution of perpetual-futures liquidity provision does not stop at the current competitive landscape. Starting from the analytical framework constructed in this chapter—the three-pillar cost structure, the toxic-flow spectrum, and the market-making competitive landscape—we can identify two evolutionary directions directly relevant to the core argument. This is only a forward-looking, outline-level discussion, meant to point out these directions' connection to this chapter's framework rather than to develop them into independent new frameworks.

First, artificial intelligence is reshaping the risk-management and toxicity-detection capabilities of market making. Traditional market-making algorithms are based on a parameterized rule set—for example, "when inventory exceeds X, skew the quotes by Y basis points"—and their ability to adapt to changes in market state is limited. Reinforcement-learning-based market-making agents, by contrast, can explore optimal strategies in a multidimensional market-state space (volatility, order-flow imbalance, funding rate, etc.). This technological advance intersects with this chapter's framework at two direct points: first, at the level of toxic-flow identification, AI systems have the potential to distinguish in real time, more precisely, the four types of toxic flow defined in Section 19.4 (traditional indicators such as VPIN mainly capture informed-trading toxicity, whereas AI models have the potential to simultaneously identify liquidation-cascade signals, arbitrage-extraction patterns, and MEV-attack characteristics); second, at the level of inventory management, AI can dynamically optimize the parameters in the extended decision function $\delta^* = f(Q, \sigma, \gamma, FR_t, T_{\text{eff}}, \kappa)$ discussed in Section 19.3.1. A caution, however: when many market makers adopt similar AI architectures and training data, their strategies may produce highly correlated response patterns, withdrawing liquidity in sync during extreme moves and thereby intensifying the behavioral-correlation risk analyzed in Section 19.7.2. In addition, AI market making introduces an operational-risk dimension that traditional algorithmic market making does not have: a model may produce unpredictable anomalous behavior in market states not covered by its training data (the 2012 case in which Knight Capital lost about $460 million in 45 minutes due to an out-of-control algorithm is a cautionary precedent from traditional finance), and the black-box nature of AI makes both post hoc diagnosis and real-time intervention more difficult.

Second, dynamic incentive mechanisms may change market makers' behavioral patterns in a crisis. The core contradiction repeatedly argued in this chapter—that liquidity provision is scarcest precisely when demand is greatest—is rooted in the fact that a market maker's profit equation points to withdrawal in extreme conditions. Dynamic incentive mechanisms attempt to modify the parameters of this equation. For example, one can design a volatility-responsive rebate mechanism that automatically raises the rebate ratio for makers when market volatility exceeds a specific threshold, directly increasing the "spread revenue" term in a market maker's profit equation and partly offsetting surging institutional cost and adverse-selection cost. Such a mechanism essentially transforms crisis-time liquidity provision from a purely market behavior into a behavior backed by institutional incentives, changing a market maker's optimal strategic choice under extreme conditions. On-chain programmable contracts make the implementation of such dynamic mechanisms possible, which is also a potential institutional advantage of the on-chain CLOB paradigm relative to traditional CEXs.

In addition, intent-based architectures and cross-chain liquidity integration are also gradually changing the infrastructural form of liquidity provision: the former shifts execution complexity from the user side to the professional-market-maker side, while the latter works to solve the margin-fragmentation problem. These two directions intersect only indirectly with this chapter's core analytical framework and belong more to the evolution of market infrastructure, so they are not elaborated here.

Overall, AI-driven intelligence and the dynamization of incentive design are the two evolutionary directions most directly relevant to this chapter's three-pillar cost framework and the toxic-flow spectrum. They have the potential to mitigate the inherent fragility of liquidity provision at the technical and institutional levels, respectively, but they also introduce new risk dimensions such as strategy resonance and incentive arbitrage, which market participants and protocol designers must continue to monitor.

## 19.8 Chapter summary

This chapter analyzed a central question: in perpetual-futures markets, where does liquidity come from? The analysis shows that liquidity is not a naturally existing resource but a commodity produced by participants in different roles at the cost of bearing specific risks and costs. It offered three interrelated frameworks for understanding this production process.

At the level of cost structure, this chapter argued that the twin pillars of classic market-making theory—adverse-selection cost and inventory-risk cost—are insufficient to explain market-making behavior in perpetual-futures markets, and that the institutional design of perpetual futures (the funding rate, the liquidation mechanism, no expiration date, and 24/7 trading) introduces a "third pillar": institutional cost. This "third pillar" is an integrative consolidation and unified naming of several already-known cost sources in the perpetual-futures context, not a newly discovered cost type. It extends a market maker's profit equation to $\text{Total profit} = \text{spread revenue} - (\text{adverse-selection cost} + \text{inventory-risk cost} + \text{institutional cost})$. Institutional cost structurally raises the threshold of perpetual-futures market making and explains why its liquidity cost is usually higher than that of spot or traditional futures markets under comparable conditions. From BTC to altcoins, the sharp rise in the share of adverse-selection cost further reveals the microeconomic root of the liquidity inefficiency of small-cap tokens. Building on this, this chapter constructed the "toxic-flow spectrum" classification framework, placing the toxicity risk market makers face into a unified coordinate system composed of "information content" and "permanence of price impact," and accordingly identifying four toxicity regions with different dominant mechanisms but overlapping (rather than mutually exclusive and collectively exhaustive) boundaries: informed-trading toxicity, arbitrage-extraction toxicity, liquidation-cascade toxicity, and MEV-attack toxicity. These four types of toxicity differ systematically in information content, permanence of price impact, temporal characteristics, and response strategy, and confusing them leads to a strategy mismatch: using a retreat strategy meant for informed trading to deal with liquidation flow would miss the rebound opportunity, whereas the reverse could incur major losses in a genuine price impact. At the level of market structure, this chapter proposed a unified framework for comparing CEX and DEX liquidity-provision paradigms. The on-chain CLOB paradigm and the market-making vault paradigm make different trade-offs across dimensions such as the risk-bearing party, pricing mechanism, LP autonomy, capital efficiency, and barrier to entry; there is no all-around dominant solution, and the two represent different positions in the trade-off among decentralization, efficiency, and risk control.

Together, these three frameworks answer this chapter's core question: the liquidity of perpetual futures is dynamically produced by professional market makers on CEXs and by liquidity providers in on-chain CLOBs and market-making vaults on DEXs, within the game of coping with the three costs and the four types of toxic flow. The cost structure and risk characteristics of this production process directly determine the quantity, quality, and stability of the liquidity observable in the market. However, this production system contains a structural contradiction: the quantity of liquidity supplied is negatively correlated with market demand. When the market is calm, market-making risk is low and liquidity supply is ample, manifested as narrow spreads and a deep order book; when the market swings violently, the four types of toxic flow erupt simultaneously, inventory risk and liquidation risk rise sharply, and a market maker's profit equation points to the only rational choice: widen spreads, cut depth, and even withdraw from the market. Liquidity supply is most ample when demand is smallest and scarcest when demand is greatest. This paradox does not stem from a behavioral flaw of market makers but is the inevitable result of the profit equation under extreme conditions—a fragility endogenous to market microstructure. The next chapter focuses on the dynamic unfolding of this core paradox: the endogeneity and reflexivity of liquidity, and how it leads to the recurring market phenomenon of liquidity contracting sharply precisely when it is most needed.

## References

[1] Makarov, I., & Schoar, A. (2020). Trading and arbitrage in cryptocurrency markets. *Journal of Financial Economics*, *135*(2), 293–319. https://doi.org/10.1016/j.jfineco.2019.07.001

[2] Ackerer, D., Hugonnier, J., & Jermann, U. (2025). Perpetual futures pricing. *Mathematical Finance*, *35*(1), 3–31. https://doi.org/10.1111/mafi.70018

[3] Akerlof, G. A. (1970). The market for "lemons": Quality uncertainty and the market mechanism. *The Quarterly Journal of Economics*, *84*(3), 488–500. https://doi.org/10.2307/1879431

[4] Glosten, L. R., & Milgrom, P. R. (1985). Bid, ask and transaction prices in a specialist market with heterogeneously informed traders. *Journal of Financial Economics*, *14*(1), 71–100. https://doi.org/10.1016/0304-405X(85)90044-3

[5] Kyle, A. S. (1985). Continuous auctions and insider trading. *Econometrica*, *53*(6), 1315–1335. https://doi.org/10.2307/1913210

[6] Ho, T., & Stoll, H. R. (1981). Optimal dealer pricing under transactions and return uncertainty. *Journal of Financial Economics*, *9*(1), 47–73. https://doi.org/10.1016/0304-405X(81)90020-9

[7] Amihud, Y., & Mendelson, H. (1980). Dealership market: Market-making with inventory. *Journal of Financial Economics*, *8*(1), 31–53. https://doi.org/10.1016/0304-405X(80)90020-3

[8] He, S., Manela, A., Ross, O., & von Wachter, V. (2022). Fundamentals of perpetual futures. *arXiv preprint arXiv:2212.06888*. https://doi.org/10.48550/arXiv.2212.06888

[9] Schmeling, M., Schrimpf, A., & Todorov, K. (2023). Crypto carry. *BIS Working Papers*, No. 1087. https://doi.org/10.2139/ssrn.4268371

[10] Galati, L. (2024). Exchange market share, market makers, and murky behavior: The impact of no-fee trading on cryptocurrency market quality. *Journal of Banking & Finance*, *165*, 107222. https://doi.org/10.1016/j.jbankfin.2024.107222

[11] Easley, D., O'Hara, M., Yang, S., & Zhang, Z. (2024). Microstructure and market dynamics in crypto markets. *SSRN Working Paper*. https://doi.org/10.2139/ssrn.4814346

[12] Easley, D., López de Prado, M. M., & O'Hara, M. (2012). Flow toxicity and liquidity in a high-frequency world. *The Review of Financial Studies*, *25*(5), 1457–1493. https://doi.org/10.1093/rfs/hhs053

[13] Avellaneda, M., & Stoikov, S. (2008). High-frequency trading in a limit order book. *Quantitative Finance*, *8*(3), 217–224. https://doi.org/10.1080/14697680701381228

[14] Brunnermeier, M. K., & Pedersen, L. H. (2009). Market liquidity and funding liquidity. *Review of Financial Studies*, *22*(6), 2201–2238. https://doi.org/10.1093/rfs/hhn098

[15] Milionis, J., Moallemi, C. C., Roughgarden, T., & Zhang, A. L. (2022). Automated market making and loss-versus-rebalancing. *arXiv preprint arXiv:2208.06046*. https://doi.org/10.48550/arXiv.2208.06046

[16] OECD. (2023). DeFi liquidations: Volatility and liquidity. *OECD Working Papers on Finance, Insurance and Private Pensions*, No. 48. https://doi.org/10.1787/0524faaf-en

[17] Daian, P., Goldfeder, S., Kell, T., Li, Y., Zhao, X., Bentov, I., Breidenbach, L., & Juels, A. (2020). Flash Boys 2.0: Frontrunning in decentralized exchanges, miner extractable value, and consensus instability. In *2020 IEEE Symposium on Security and Privacy (SP)* (pp. 910–927). IEEE. https://doi.org/10.1109/SP40000.2020.00040

[18] Canidio, A., & Fritsch, R. (2023). Arbitrageurs' profits, LVR, and sandwich attacks: Batch trading as an AMM design response. In *Proceedings of the 5th Conference on Advances in Financial Technologies (AFT 2023)*, LIPIcs Vol. 282, pp. 24:1–24:17. https://doi.org/10.4230/LIPIcs.AFT.2023.24

[19] CoinGlass. (2025). *2025 cryptocurrency derivatives market report*. https://www.coinglass.com/learn/2025-annual-report-en

[20] Aramonte, S., Huang, W., & Schrimpf, A. (2021). DeFi risks and the decentralisation illusion. *BIS Quarterly Review*, December 2021, 21–36. https://www.bis.org/publ/qtrpdf/r_qt2112b.htm

[21] Binance. (2025). *USDⓈ-margined futures market maker program: Fee schedule and program rules*. https://www.binance.com/en/support/faq/futures-market-maker-program

[22] Capponi, A., & Jia, R. (2021). The adoption of blockchain-based decentralized exchanges. *arXiv preprint arXiv:2103.08842*. https://doi.org/10.48550/arXiv.2103.08842

[23] Comerton-Forde, C., Hendershott, T., Jones, C. M., Moulton, P. C., & Seasholes, M. S. (2010). Time variation in liquidity: The role of market-maker inventories and revenues. *The Journal of Finance*, *65*(1), 295–331. https://doi.org/10.1111/j.1540-6261.2009.01530.x