Stackelberg Game for DP Data Pricing

• The paper develops a Stackelberg game framework that models the sequential interaction between a market-maker and a data buyer, balancing differential privacy and data utility.
• It introduces a balanced pricing mechanism that ensures incentive compatibility and arbitrage-freeness by linking noise variance selection to privacy compensation.
• The framework’s nonlinear power pricing extension demonstrates how adjusting price elasticity can tune the market outcome and influence precision in data responses.

A Stackelberg game framework for pricing differentially private (DP) data formalizes the sequential interaction between a market-maker, who sets a pricing rule, and a data buyer, who selects the noise variance of her query response subject to differential privacy constraints. This framework accommodates the privacy-utility trade-off inherent in DP mechanisms, allowing incentive-compatible, privacy-conscious, and arbitrage-free pricing for statistical queries in modern data markets (Bo et al., 20 Dec 2025).

1. Model Structure and Key Actors

The Stackelberg game is comprised of two players:

• Leader (Market-Maker): Commits to a pricing mechanism parameterized by $k > 0$.
• Follower (Data Buyer): Observes the posted price schedule and chooses the noise variance $\sigma^2 \ge \sigma^2_{\min}$ to maximize her net utility.

Let $q \in \mathbb{R}^n$ denote the linear query, $f(q) \ge 0$ a semi-norm (often $\|q\|_2$), and $A(q) > 0$ the buyer's value-intensity. The buyer's valuation for a noisy answer with standard deviation $\sigma$ is $V(q, \sigma) = A(q)/\sigma$.

To enforce $\epsilon$-differential privacy, Laplace noise $\xi \sim \mathrm{Lap}(0, b)$ with $\sigma^2 \ge \sigma^2_{\min}$0 is added, ensuring

$\sigma^2 \ge \sigma^2_{\min}$1

where $\sigma^2 \ge \sigma^2_{\min}$2 is the query's $\sigma^2 \ge \sigma^2_{\min}$3-sensitivity.

Each data owner $\sigma^2 \ge \sigma^2_{\min}$4 experiences privacy loss $\sigma^2 \ge \sigma^2_{\min}$5, and is compensated $\sigma^2 \ge \sigma^2_{\min}$6. The aggregate privacy-cost threshold is

$\sigma^2 \ge \sigma^2_{\min}$7

2. Balanced Pricing Mechanism and Utility

The market-maker posts a balanced (arbitrage-free) pricing function:

$\sigma^2 \ge \sigma^2_{\min}$8

ensuring payments always cover privacy costs. The data buyer's net utility is

$\sigma^2 \ge \sigma^2_{\min}$9

while the market-maker's profit is

$q \in \mathbb{R}^n$0

where $q \in \mathbb{R}^n$1.

3. Equilibrium Analysis and Market Regimes

The game is solved via backward induction:

• Follower’s Stage: For a fixed $q \in \mathbb{R}^n$2, the buyer maximizes $q \in \mathbb{R}^n$3 over $q \in \mathbb{R}^n$4. The switch point between pricing branches is

$q \in \mathbb{R}^n$5

yielding a piecewise utility:

$q \in \mathbb{R}^n$6

• Leader’s Stage: Anticipating the buyer's best response $q \in \mathbb{R}^n$7, the market-maker chooses $q \in \mathbb{R}^n$8 to maximize profit:

$q \in \mathbb{R}^n$9

Three regimes emerge based on $f(q) \ge 0$0 versus $f(q) \ge 0$1 (for $f(q) \ge 0$2 subject to $f(q) \ge 0$3):

Regime	Condition	Buyer Optimal $f(q) \ge 0$4	Market-Maker Profit
Profitable	$f(q) \ge 0$5	$f(q) \ge 0$6, $f(q) \ge 0$7	$f(q) \ge 0$8
Break-even	$f(q) \ge 0$9	$\|q\|_2$0	$\|q\|_2$1
No-trade	$\|q\|_2$2	$\|q\|_2$3 (buyer declines)	$\|q\|_2$4

Buyer participation requires $\|q\|_2$5. Profitable trade requires $\|q\|_2$6. In the break-even region, trade occurs with zero market-maker profit.

4. Differential Privacy Constraints and Micro-Payments

The pricing strategy is tightly linked to DP guarantees. For a target $\|q\|_2$7, the minimum achievable noise variance for query $\|q\|_2$8 is $\|q\|_2$9. Each data owner is compensated to reflect individual privacy loss, with aggregate micro-payments always covered by the balanced pricing function:

$A(q) > 0$0

This design ensures incentive compatibility for both market participants and data owners, aligning economic incentives with privacy constraints.

5. Nonlinear Power Pricing Extension

The framework generalizes to nonlinear pricing functions:

$A(q) > 0$1

for $A(q) > 0$2. The branch-switch (generalized threshold variance) is now

$A(q) > 0$3

In the profitable regime for $A(q) > 0$4, the equilibrium is

$A(q) > 0$5

This extension allows the market-maker to alter price elasticity with respect to accuracy, further tuning profit and query precision outcomes.

6. Significance, Interpretations, and Market Implications

The Stackelberg game framework for DP data pricing encapsulates several critical properties:

• Sequential Price Setting: The leader adjusts per-unit accuracy pricing to influence the trade-off faced by the follower between data utility (accuracy) and privacy.
• Privacy-Utility Encoding: The follower’s choice of noise variance $A(q) > 0$6 directly maps to the classical privacy-utility trade-off, with higher $A(q) > 0$7 implying better privacy (lower $A(q) > 0$8) at lower data value.
• Incentive Compatibility and Arbitrage-Freeness: The balanced price function enforces that total payment to data owners is never undercut, establishing arbitrage-freeness.
• Sharp Regime Division: The model divides the market into strictly profitable, break-even, and non-participation regimes based on $A(q) > 0$9 and $\sigma$0, simplifying strategic analysis and contract design.
• Mechanism Design Link: The approach bridges differential privacy with economic mechanism design, supporting unified, market-based yet privacy-conscious data exchange.

A plausible implication is that by adjusting the pricing function's structure (e.g., power pricing), data markets can flexibly regulate the sensitivity of pricing to buyer precision demands, tailoring market outcomes to policy or competitive considerations (Bo et al., 20 Dec 2025).