Nonlinear Impact & Dynamic Arbitrage

• Nonlinear Impact and Dynamic Arbitrage is the study of how nonlinear price impacts influence market microstructure, optimal execution, and dynamic round-trip arbitrage.
• The analysis integrates stochastic control, BSDE techniques, and empirical calibration to capture state-dependent, concave, and transient market effects.
• Empirical findings in both centralized and on-chain markets validate these models and highlight the role of market frictions in preventing exploitable arbitrage.

Nonlinear Impact and Dynamic Arbitrage refers to the study of market microstructure, optimal execution, and valuation under price impact models where the impact exerted by trades on prices is nonlinear in trade size or history. The central concern is the interplay between the mathematical form of market impact, the potential for dynamic (round-trip) arbitrage, and the structure of feasible pricing, hedging, and execution strategies. This area integrates stochastic control, nonlinear analysis, and financial economics to address how micro-level trading generates emergent price dynamics and constraints for both individual strategies and markets as a whole.

1. Nonlinear Impact Models: Types and Mathematical Structure

Nonlinear impact models generalize linear frameworks by making the price impact—how quoted prices respond to incoming order flow—an explicit nonlinear function of the trade size, trading rate, inventory, or past transactions. Impact may be classified as:

• Permanent vs. Transient Impact: Permanent impact accumulates with cumulative trades and does not revert, while transient impact decays over time or volume.
• Instantaneous (Slippage) Impact: Nonlinearities may also manifest in the execution price at the instant of the trade.

For single-asset markets, a general nonlinear impact model assumes that after applying a trading strategy $v_t$ (or a discrete order $y$), the price evolves as

$dS_t = \sigma\,dW_t - f_t(v_t)\,dt,$

where $f_t(\cdot)$ is a possibly nonlinear impact function, often depending on cumulative quantity already traded. In nonlinear transient impact models, the impacted price is a functional of the past trading rate $v$ via a propagator kernel $G(t,s)$ and nonlinear transformation $h$: $$S^u_t = S_t + h \left( \int_0^t G(t,s) u_s \, ds \right).$$ Empirical evidence supports nonlinear, frequently concave (sublinear) power-law forms, such as $f(v) \propto |v|^\delta$ with $\delta < 1$, most notably the ubiquitous square-root law for market impact (Guéant, 2013, Donier et al., 2014).

In endogenous impact models, the price curve itself is a nonlinear function derived from utility indifference of liquidity suppliers, i.e., $$P_t(z, y) = \inf\{p \in \mathbb{R} : I_t(H^M + zS - yS + p) \ge I_t(H^M + zS)\}$$, with $I_t$ a time-consistent $g$-expectation (Fukasawa et al., 2017).

2. Dynamic Arbitrage: Definitions and Structural Constraints

Dynamic (round-trip) arbitrage refers to the theoretical possibility that a trader, through a sequence of self-financed trades that returns the inventory to its origin, can generate strictly positive expected profits. In the frictionless and linear-impact world, non-arbitrage is enforced through the linearity of impact or through classical measures (NFLVR); in nonlinear impact models, there are broader and more nuanced constraints.

• Gatheral’s Principle for Permanent Impact: In linear permanent impact models, absence of dynamic arbitrage is secured if and only if impact is linear in trading rate. Nonlinear $k(v)$ (impact as a function of rate) generically admits price manipulation (Guéant, 2013, Schneider et al., 2016).
• Nonlinear, State-Dependent Permanent Impact: Concave, nonlinear permanent market impact is consistent with no dynamic arbitrage, provided impact depends on cumulative traded quantity, not instantaneous rate (Guéant, 2013). For state-dependent $f(|q_0 - q_t|)v_t$, all round-trip strategies yield zero expected profit, since the cash gain from impacting the price cancels on reversal.
• Pathwise and Cross-Impact Extensions: In multidimensional settings, the need for linear, odd, and symmetric cross-impact functions is required to prevent multi-asset dynamic arbitrage unless kernels have unbounded tails or other mitigating frictions are present (Schneider et al., 2016).

A precise mathematical statement is that for any round-trip schedule $q(0) = q(T)$ (or, in transient kernels, $\int_0^T m_s ds = 0$), the expected cost (or profit) is nonnegative.

Table: No-Dynamic-Arbitrage Criteria (Single-Asset Impact)

Impact Form	Arbitrage-Free if...	Reference
$f(v)$, time-homogeneous (permanent)	$f(v)$ linear in $v$ (classic)	(Guéant, 2013)
$f(|q_0 - q_t|) v_t$ (state-dependent)	$f(\cdot)$ positive, decreasing	(Guéant, 2013)
$h(D^u_t)$, $h$ concave (transient)	Monotonicity: $$\langle u-v, \mathcal{A}(u) - \mathcal{A}(v) \rangle \ge 0$$	(Jaber et al., 6 Mar 2025)

3. Impact on Pricing, Hedging, and BSDE Representation

In impact-aware models, contingent claim replication, pricing, and hedging must be reformulated. Nonlinear impact induces path-dependent, nonlinear wealth dynamics. Several frameworks have been developed:

• g-Expectations and Indifference Pricing: Modeling a market's quoted curve as the g-expectation (utility-indifferent) price of a liquidity supplier makes the price function nonlinear, inventory-dependent, and cash-invariant. The lack of dynamic arbitrage is an immediate consequence of the indifference structure: for any $z, y$, $P_t(z, y) + P_t(z-y, -y) = 0$ (Fukasawa et al., 2017).
• Nonlinear Black–Scholes Equations: For linear permanent impact, exact replication yields a PDE for the option price $u(t,s)$:

$$\partial_t u + \frac{1}{2} \sigma^2 \frac{\gamma}{1 - \lambda \gamma} = 0,\quad \gamma = s^2 \partial_{ss}u,$$

which is singular when $1 - \lambda \gamma \le 0$. The non-arbitrage region is characterized by $1 - \lambda \gamma > 0$ (Loeper, 2013).

• BSDE/Reflected BSDE Characterizations: Nonlinear market frictions (impact, funding spreads, etc.) are encoded in the driver $g$ of a backward stochastic differential equation. Under suitable (strict) comparison for the BSDE/Reflected BSDE, arbitrage is excluded and replication cost is unique (Bielecki et al., 2017, Kim et al., 2018, Nie et al., 2018).

4. Execution Costs and Optimal Trading under Nonlinearity

Optimal execution under nonlinear impact is analyzed via minimization of expected shortfall, with cost functionals of the form

$$J[x] = \mathbb{E}\left[\int_0^T v(t)\int_0^t f(v(s)) G(t-s)\,ds\,dt\right],$$

where $f$ models impact (often concave), and $G$ the decay kernel. Key findings are:

• Front-Loading and Buy–Sell Oscillations: With concave $f$, optimal strategies front-load trades; more aggressive trading is optimal early to exploit sublinear impact (Curato et al., 2014, Jaber et al., 6 Mar 2025).
• Pathologies and Arbitrage: For sufficiently strong concavity or power-law decay, optimal solutions can exhibit negative expected cost (i.e., profitable dynamic round-trips), indicating arbitrage (Curato et al., 2014).
• Restoring Regularity: Introducing execution spreads (L1 regularization) or enforcing convexity in $f$ for large $|v|$ mitigates arbitrage and yields monotone, non-manipulative strategies (Curato et al., 2014). In the Fredholm approach, monotonicity criteria on $h$ (impact nonlinearity) are necessary: for $h(x) \propto |x|^c$, must have $c \geq 1/2$ for absence of arbitrage (Jaber et al., 6 Mar 2025).
• Market Resistance: Inclusion of market resistance—via an adversarial trader or order book response—increases the range of conditions ensuring no-arbitrage and alters the optimal strategy profile. For linear resistance, the optimal strategy is unique; strict convexity ensures existence (Chahdi et al., 6 Jan 2026).

5. Market Microstructure and Empirical Implementation

Recent advances include the empirical study and calibration of nonlinear impact models in real-world financial and algorithmic environments:

• Dynamic-Weight AMMs and On-Chain Markets: In temporal function AMMs, convex invariant functions induce nonlinear price impact, leading to a natural sawtooth in mispricing and a dynamic arbitrage regime rapidly compressed by arbitrageurs through "Dutch reverse auctions." On-chain implementation demonstrates that such nonlinear impact pools can match or exceed centralized exchange execution efficiency when frictions (gas, MEV) are low (Willetts et al., 25 Feb 2026).
• Estimation and Model Selection: Impact estimation must account for concavity; naive linear estimation systematically under- or overestimates true impact coefficients. Power-law estimates with exponents $a \approx 1/2$ (square-root law) fit observed metaorder data (Guéant, 2013).
• Robustness and Regulatory Metrics: Frictions (spreads, finite speed) can mollify theoretical arbitrages in practice. Regulatory proposals suggest monitoring latent liquidity as a fragility metric, as low liquidity directly amplifies nonlinear impact (Donier et al., 2014).

6. Arbitrage Bounds, Game Options, and Nonlinear Market Completeness

Complex claims—American options, game options, high-dimensional functionally generated portfolios—require further refinement of arbitrage and pricing notions in a nonlinear impact context:

• Nonlinear No-Arbitrage Intervals: In inventory-based impact models, three notions of arbitrage-free price diverge: strong (at all sizes), level-$u$ (fixed size), and utility-demand (maximal position) arbitrage-freeness (Anthropelos et al., 2018). This leads to intervals (not points) for fair pricing, widening as risk aversion $\gamma$ or trade size increases; arbitrage, if present, is local in size, not scalable (Anthropelos et al., 2018).
• BSDE/Reflected BSDE Characterizations: In nonlinear settings including impact, funding, and default risk, existence, uniqueness, and admissibility of “fair” value processes are guaranteed by comparison theorems for (doubly-)reflected BSDEs. For American and game options, one obtains explicit delta-hedging strategies and stopping rules from the RBSDE/DRBSDE solutions (Bielecki et al., 2017, Kim et al., 2018, Nie et al., 2018).
• Portfolio Theory with Nonlinear Impact: Master formulas and relative arbitrage results extend to high-dimensional markets with nonlinear, decaying price impact. Provided trading speed is controlled, classic SPT results—such as diversity-driven relative arbitrage—remain robust to the presence of convex nonlinear impact costs (Itkin, 9 Jun 2025).

7. Conclusions and Implications

The study of nonlinear impact and dynamic arbitrage uncovers subtle structural relationships: concavity in impact is both empirically essential and mathematically hazardous, requiring either suitable model structure (state-dependent, monotone mapping, endogenous utility models) or market frictions (spread, resistance, slippage) for dynamic no-arbitrage to obtain. Practical execution and pricing in such settings demand recourse to nonlinear stochastic control and BSDE methods, with empirical verification essential. Modern on-chain markets and algorithmic venues exhibit the predicted nonlinearities and arbitrage compression in real time, validating the theoretical predictions of nonlinear impact models.

Fundamentally, the field demonstrates that microstructural regularities and macro-level arbitrage constraints dictate not only price dynamics but also the feasibility and efficiency of optimal strategies in real, frictional markets (Guéant, 2013, Donier et al., 2014, Fukasawa et al., 2017, Jaber et al., 6 Mar 2025, Curato et al., 2014, Anthropelos et al., 2018, Willetts et al., 25 Feb 2026, Chahdi et al., 6 Jan 2026, Itkin, 9 Jun 2025, Schneider et al., 2016, Loeper, 2013, Kim et al., 2018, Nie et al., 2018, Bielecki et al., 2017).