💰 quantitative finance

The Privacy Subsidy: Kyle's $\lambda$ under Noise-Perturbed Order-Flow Observation

This paper derives a unique linear Kyle equilibrium for privacy-preserving cryptocurrency exchanges where a market maker observes noise-perturbed order flow, demonstrating that while price impact and informed trading strategies rescale with privacy noise, their product remains invariant, thereby establishing a closed-form "privacy subsidy" that quantifies the necessary fee to compensate liquidity providers for information leakage.

Original authors: Yuki Nakamura

Published 2026-05-18

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Yuki Nakamura

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: The "Blindfolded" Auctioneer

Imagine a bustling marketplace where people are buying and selling a rare, valuable item (like a digital coin). In a normal market, there is an Auctioneer (the Market Maker) who sets the price based on how many people are buying or selling right now.

Usually, the Auctioneer sees the total number of orders clearly. If a "smart insider" (someone who knows the item's true future value) tries to buy a lot, the Auctioneer sees the big order, realizes something is up, and immediately raises the price to protect themselves. This makes it hard for the insider to make a profit.

The Problem: In modern crypto markets, people want privacy. They don't want the Auctioneer to see their exact orders. So, the market adds a layer of "static" or "noise" to the orders before the Auctioneer sees them. It's like the Auctioneer is wearing foggy glasses. They can see that some orders came in, but they can't tell exactly how big they are or who sent them.

The Discovery: Privacy Has a Hidden Price Tag

The paper asks a simple question: What happens to the Auctioneer's wallet when they wear these foggy glasses?

The author, Yuki Nakamura, runs a mathematical simulation and finds a surprising result:

The Insider Gets Smarter: Because the Auctioneer is confused by the fog, the "smart insider" realizes they can trade bigger amounts without getting caught immediately. They become more aggressive.
The Auctioneer Loses Money: Because the Auctioneer can't see the true size of the orders, they set the price slightly wrong. The "smart insider" exploits this mistake to make a profit.
The "Privacy Subsidy": This profit the insider makes isn't free. It comes directly out of the Auctioneer's pocket. The paper calls this loss the "Privacy Subsidy."

The Analogy: Think of the Auctioneer as a shop owner and the "noise" as a thick fog rolling into the store.

In clear weather, the shop owner sees a customer buying 100 items and knows to raise the price.
In the fog, the shop owner sees a blurry shape and guesses the customer is buying 50 items, so they charge a lower price.
The customer (who knows the fog is there) buys 100 items at the lower price, makes a huge profit, and the shop owner loses money.
The Subsidy: The money the shop owner loses is the "cost of the fog." If the shop wants to stay open, they must charge a special fee to cover this loss.

The Key Findings (In Plain English)

1. The "Half-Revealing" Rule Still Holds
Even with the fog, the price still moves halfway toward the truth. The fog doesn't change how much the price moves on average; it just makes the price noisier. The "smart insider" and the "Auctioneer" still balance each other out in a specific mathematical way, regardless of how thick the fog is.

2. The Cost Grows with the Fog
The paper calculates exactly how much money the Auctioneer loses based on how thick the privacy fog is.

Low Fog: If you add a tiny bit of privacy, the cost is very small (it grows slowly, like a square).
High Fog: If you add a lot of privacy, the cost grows faster and becomes a straight line.
The Sweet Spot: There is a specific point where the cost of adding more privacy starts to slow down. It's like driving in fog: the first few steps of fog are scary, but once you are already in a thick fog, adding a little more doesn't change your visibility much.

3. Everyone Wins (Except the Bank)
Here is the twist:

The Smart Insider makes more money because the fog hides them.
The Regular People (Noise Traders) actually lose less money than usual. Why? Because the fog also hides them from the Smart Insider! The Smart Insider can't pick them off as easily.
The Result: Both the Smart Insider and the Regular People are better off.
The Catch: The Protocol/Liquidity Pool (the bank providing the money) loses exactly the sum of what the others gained. The "Privacy Subsidy" is the money transferred from the bank to the traders.

The Real-World Application

The paper focuses on a specific type of crypto exchange called a Shielded AMM (Automated Market Maker) that uses "Differential Privacy" (adding mathematical noise).

The paper concludes that if you build such a privacy-preserving exchange, you cannot just charge a standard fee. You must charge a specific "break-even fee" that covers the Privacy Subsidy.

If you don't charge this fee, the people providing the liquidity (the bank) will lose money and leave.
The paper gives a formula to calculate exactly how much that fee should be based on how much privacy noise you inject.

What This Paper Does Not Say

It does not talk about other types of privacy (like "batching" orders together or "sealed bids" where you wait to see the price). Those are different puzzles that need different solutions.
It does not suggest that privacy is "bad." It simply says privacy has a cost, and that cost must be paid by the system (via fees) to keep the market running.

Summary Metaphor

Imagine a game of poker where the dealer (the Market Maker) is wearing sunglasses that blur the cards.

The Pro Players (Insiders) realize they can bluff harder because the dealer can't see their cards clearly.
The Casual Players (Noise Traders) realize the dealer can't see their cards either, so the Pro Players can't cheat them as easily.
The Casino (The Protocol) loses money because the Pro Players are winning more, and the Casual Players are losing less.
The Paper's Conclusion: To keep the Casino open, they must charge an extra "Blur Fee" to the players. This fee is the Privacy Subsidy. It's the price you pay to keep your cards hidden.

Technical Summary: The Privacy Subsidy: Kyle's $\lambda$ under Noise-Perturbed Order-Flow Observation

1. Problem Statement

The paper addresses a gap in classical market microstructure theory regarding privacy-preserving cryptocurrency exchanges. In standard Kyle (1985) models, a market maker (MM) observes the aggregate order flow $y = x + u$ (where $x$ is informed trading and $u$ is noise trading) to set prices. However, modern privacy-preserving designs (e.g., shielded AMMs, zero-knowledge exchanges) alter this observation. Instead of seeing $y$ , the MM observes a noise-perturbed signal $\tilde{y} = y + \epsilon$ , where $\epsilon$ is independent Gaussian privacy noise.

The central problem is that classical results do not extend to this regime. Specifically, it is unknown how the price-impact coefficient ( $\lambda$ ), the informed trader's strategy ( $\beta$ ), and the welfare distribution change when the MM's observation is obscured by privacy noise, particularly when the MM is a committed Bayesian agent (like a smart contract) rather than a competitive zero-profit agent.

2. Methodology and Model Setup

The authors construct a single-period, linear Kyle equilibrium model with the following primitives:

Asset Value ( $v$ ): Normally distributed $N(p_0, \sigma_v^2)$ .
Noise Flow ( $u$ ): Normally distributed $N(0, \sigma_u^2)$ , independent of $v$ .
Privacy Noise ( $\epsilon$ ): Normally distributed $N(0, \sigma_\epsilon^2)$ , independent of both $v$ and $u$ .
Observation: The MM observes $\tilde{y} = x + u + \epsilon$ .
Market Maker (MM): Modeled as a committed Bayesian AMM. Unlike the textbook competitive MM that enforces zero expected profit, this MM mechanically sets the price $p(\tilde{y}) = E[v | \tilde{y}]$ . Any expected loss is absorbed by the protocol's liquidity provider (LP) pool. This framing is chosen to model on-chain smart-contract pricing rules that cannot strategically adjust to ensure zero profit.
Informed Trader: A risk-neutral insider observing $v$ and submitting a linear order $x = \beta(v - p_0)$ .

The equilibrium is defined by a pair $(\beta, \lambda)$ such that the pricing rule is Bayesian rational and the informed trader maximizes expected profit given the pricing rule.

3. Key Contributions and Results

A. Closed-Form Linear Equilibrium (Theorem 1)

The paper derives the unique linear equilibrium under privacy noise. The equilibrium coefficients are:
$\lambda = \frac{\sigma_v}{2\sqrt{\sigma_u^2 + \sigma_\epsilon^2}}$
$\beta = \frac{\sqrt{\sigma_u^2 + \sigma_\epsilon^2}}{\sigma_v}$

Key Properties:

Rescaling: Both $\lambda$ and $\beta$ rescale by a single factor dependent on the privacy parameter $\sigma_\epsilon$ .
Invariance: The product $\lambda \beta = 1/2$ remains invariant regardless of the level of privacy noise. This preserves the "half-revealing" property of the classic Kyle equilibrium: the conditional mean price moves halfway from the prior to the true value, $E[p|v] = (p_0 + v)/2$ .
Comparative Statics: As privacy noise ( $\sigma_\epsilon$ ) increases, the price-impact coefficient $\lambda$ decreases (prices become less sensitive to flow), and the informed trader's strategy $\beta$ intensifies (the insider trades more aggressively to hide in the noise).

B. The Privacy Subsidy (Theorem 2)

The paper performs a welfare decomposition to calculate the expected profit/loss for each agent.

Informed Trader Profit ( $\pi_I$ ): Increases with $\sigma_\epsilon$ .
Noise Trader Profit ( $\pi_N$ ): Increases (losses decrease) with $\sigma_\epsilon$ .
MM/Protocol Profit ( $\pi_M$ ): Becomes negative.

The paper identifies a closed-form transfer from the protocol/LP pool to the traders, termed the Privacy Subsidy:
$|\pi_M| = \frac{\sigma_v \sigma_\epsilon^2}{2\sqrt{\sigma_u^2 + \sigma_\epsilon^2}}$

This quantity represents the break-even fee any privacy-aggregated exchange must charge to compensate its liquidity layer. It is zero when $\sigma_\epsilon = 0$ (recovering the textbook Kyle result) and strictly positive otherwise.

C. Comparative Statics of the Subsidy

The subsidy exhibits specific asymptotic behaviors:

Low-Privacy Regime ( $\sigma_\epsilon \to 0$ ): The subsidy grows quadratically ( $\propto \sigma_\epsilon^2$ ).
High-Privacy Regime ( $\sigma_\epsilon \to \infty$ ): The subsidy grows linearly ( $\propto \sigma_\epsilon$ ).
Inflection Point: The function has a single inflection point at $\sigma_\epsilon^* = \sqrt{2}\sigma_u$ , transitioning from convex to concave.

D. Applicability and Boundaries

The authors delineate the scope of their framework:

Applicable: Shielded AMMs with explicit additive Gaussian noise injection (e.g., differential privacy).
Not Applicable (Requires Separate Frameworks):
- Penumbra-style batched swaps: These reduce to textbook Kyle with rescaled noise variance but generate no privacy subsidy under a Bayesian MM.
- Suave-style sealed-bid auctions: These involve temporal asymmetry rather than additive noise, aligning closer to Loss-Versus-Rebalancing (LVR).
- Renegade-style oracle-pegged crossings: These fall outside Kyle analysis entirely as price discovery is exogenous.

4. Significance and Positioning

The paper positions its result at the intersection of market microstructure and decentralized finance (DeFi):

Microstructure Extension: It extends Kyle's 1985 linear equilibrium to a previously unanalyzed observation regime where the imperfection lies in the market maker's signal rather than the trader's information.
DeFi Analogy to LVR: The "Privacy Subsidy" is presented as the single-period, closed-form analog of Loss-Versus-Rebalancing (LVR). While LVR measures the cost of stale prices being picked off by arbitrageurs with external information, the privacy subsidy measures the cost of privacy noise obscuring the MM's flow observation. Both are welfare transfers from liquidity providers to traders.
Fee Design Implication: The result provides a theoretical lower bound for fees in privacy-preserving exchanges. The paper notes that for a BTC/USDT calibration, a privacy parameter equal to the noise-trader variance ( $\sigma_\epsilon = \sigma_u$ ) implies a break-even fee requirement of approximately $1 million per day, comparable to a 0.1% fee on $1 billion of volume.

The paper concludes that while privacy benefits both informed and noise traders (by reducing adverse selection against noise traders and increasing informed profits), the entire cost is borne by the protocol/LP pool, necessitating a specific fee structure to ensure solvency.