💰 quantitative finance

Optimal Portfolio Choice with Cross-Impact Propagators

This paper provides an explicit solution to continuous-time optimal portfolio choice problems involving transient cross-impact and temporary price impact by reducing the first-order conditions to a system of stochastic Fredholm equations, thereby deriving optimal strategies that balance revenue, risk, and alpha signals while ensuring the absence of price manipulation.

Original authors: Eduardo Abi Jaber, Eyal Neuman, Sturmius Tuschmann

Published 2026-02-20

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Eduardo Abi Jaber, Eyal Neuman, Sturmius Tuschmann

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

Imagine you are the captain of a massive ship (your investment portfolio) sailing through a busy, crowded harbor (the stock market). Your goal is to sell off your cargo (assets) or rearrange your cargo to maximize profit while minimizing the risk of a storm.

However, there's a catch: Your ship is so big that your movements change the water around you.

This paper, written by three financial mathematicians, tackles a very complex problem: How do you trade stocks when your own buying and selling messes up the prices, and when the actions you take on one stock affect the prices of other stocks?

Here is the breakdown using simple analogies:

1. The Problem: The "Wake" Effect

In the old days, financial models assumed that if you sold a stock, the price would drop a little bit and then immediately snap back to normal, like a rubber band. They also assumed that selling Apple stock had nothing to do with the price of Microsoft stock.

The authors say: "That's not how the real world works."

The Wake (Price Impact): When a giant ship moves, it leaves a wake that ripples out for a long time. Similarly, when you trade a lot of stock, the price doesn't just bounce back instantly; the "disturbance" lingers.
The Cross-Wake (Cross-Impact): If you steer your ship hard to the left, it creates waves that hit the ship next to you. In the market, if you dump a bunch of Tech stocks, it doesn't just lower Tech prices; it might also drag down the price of Energy stocks because they are connected.
The Memory (Volterra Propagators): The authors use a fancy math tool called a "Volterra propagator." Think of this as a memory bank. The market remembers your trades not just for a second, but for minutes, hours, or even days, and the "memory" fades away slowly (like a power-law decay) rather than instantly.

2. The Goal: The Perfect Dance

The investor wants to find the perfect "dance steps" (trading strategy).

Too fast: You splash too much water, creating a huge wake that hurts your own price.
Too slow: You miss the opportunity to sell before the market turns.
The Signal: You also have a crystal ball (a "predictive signal" or "alpha") telling you which way the wind is blowing. You need to use that crystal ball to time your moves perfectly.

The paper asks: Given that my moves affect the water, and the water affects my neighbors, and I have a crystal ball, what is the exact speed and direction I should move at every single second to make the most money?

3. The Solution: The "Resolvent" Map

The authors didn't just guess; they solved a massive, tangled knot of equations.

The Knot: Because the market has "memory" (it remembers past trades), the problem is "non-Markovian." This is a fancy way of saying: You can't just look at where you are right now to decide what to do next; you have to remember everything you did in the past.
The Map: They created a mathematical "map" (using something called Operator Resolvents) that tells the investor exactly how to trade. It's like having a GPS that doesn't just show you the road, but calculates how your car's weight will affect the road surface for the next 10 miles, and adjusts your steering accordingly.

4. The Safety Check: No Cheating Allowed

A major worry in these models is Price Manipulation.

The Scam: Could a trader trick the system? For example, could they buy a stock, sell it, buy it again, and somehow end up with more money than they started with just by creating artificial price movements?
The Guardrail: The authors proved that if their "Wake" model follows certain rules (specifically, if the "memory" of the market fades in a smooth, predictable way), then cheating is impossible. You cannot create free money just by shaking the water. This gives investors confidence that the model is fair and stable.

5. What They Found (The "Aha!" Moments)

The authors ran computer simulations to see how this works in real life:

The "Round Trip" Trick: If you are selling Stock A, and Stock A and Stock B are connected, your selling might crash Stock B's price. The smart strategy? Short-sell Stock B (bet against it) while you sell Stock A. This lets you profit from the crash you caused in Stock B, which helps you sell Stock A at a better price later. It's a complex, two-step dance.
The "Fast vs. Slow" Signal: If you have a signal that says "Stock A will go up soon, but only for a minute," and "Stock B will go up for a week," the cross-impact model tells you to sell Stock A immediately. Why? Because selling Stock A will temporarily crash Stock B's price, allowing you to buy Stock B cheaply before its long-term rise begins. You are using the "shock" of one trade to buy the other asset at a discount.

Summary

This paper is like a masterclass in navigating a crowded, slippery dance floor.

Old models said: "Just dance to the music."
This paper says: "The floor is sticky, your moves push your partner, and your partner's moves push you back. Here is the exact mathematical formula for how to dance so you don't fall, you don't hurt your partner, and you get the most points."

They provide the tools to handle complex, long-lasting market reactions and interconnected assets, ensuring that investors can trade efficiently without accidentally manipulating the market or getting stuck in a bad strategy.

1. Problem Statement

The paper addresses the optimal portfolio choice problem in continuous time for an investor trading $N$ risky assets. The core challenge is to maximize a revenue-risk functional while accounting for:

Transient Price Impact: The execution of trades causes temporary price distortions that decay over time.
Cross-Impact: Trades in one asset affect the prices of other assets.
Predictive Signals: The investor has access to progressively measurable signals (alphas) predicting future price movements.
General Decay Kernels: Unlike previous literature (e.g., Gârleanu and Pedersen) that assumes exponential decay, this model allows for general Volterra propagator matrices $G(t, s)$ , which can capture power-law decay and other non-Markovian behaviors observed empirically.

The objective is to find a trading speed strategy $u_t$ that maximizes:
$J(u) = \mathbb{E}\left[ \int_0^T -u_t^\top (P_t + D_t + \frac{1}{2}\Lambda u_t) dt + (X_T)^\top P_T - \frac{\gamma}{2} \int_0^T (X_t)^\top \Sigma X_t dt \right]$
where $D_t = \int_0^t G(t, s)u_s ds$ represents the price distortion, $\Lambda$ is the temporary impact matrix, $\Sigma$ is the covariance matrix, and $\gamma$ is the risk aversion parameter.

2. Methodology

The authors tackle the problem using stochastic control theory adapted for non-Markovian settings.

Non-Markovian Framework: Since general Volterra kernels (like power laws) do not possess the Markov property, standard dynamic programming or Backward Stochastic Differential Equations (BSDEs) are not directly applicable.
First-Order Conditions: The authors derive the first-order optimality condition (Gâteaux derivative) for the objective functional. This leads to a coupled system of stochastic Fredholm integral equations of the second kind involving both forward and backward components.
Operator Theory: The solution is formulated using operator resolvents. The authors define linear integral operators on the space $L^2([0, T], \mathbb{R}^N)$ induced by the propagator kernels.
Solving the System: They reduce the first-order condition to a system solvable via the resolvent of a specific integral operator $B$ . The solution is expressed explicitly in terms of the resolvent $(I+B)^{-1}$ .
Price Manipulation Analysis: To ensure the model is financially sound, they investigate conditions under which the model precludes "round-trip" price manipulation (creating profit from zero net inventory). This requires the propagator kernel to be nonnegative definite.

3. Key Contributions

A. Explicit Solutions for General Propagators

The paper provides the first explicit solution for optimal portfolio choice with general Volterra propagator matrices in continuous time.

Theorem 2.8: Derives the unique optimal trading strategy $u^*_t$ $u_{t}^{*}$ in terms of operator resolvents. The solution handles:
- General nonnegative definite propagators (including power-law and exponential).
- General progressively measurable signals (stochastic alphas).
- Multi-asset cross-impact.
Corollary 2.9: Simplifies the solution for deterministic signals, yielding a deterministic linear operator equation.

B. Conditions for No Price Manipulation

The authors establish sufficient conditions to prevent price manipulation in the sense of Huberman and Stanzl.

Theorem 2.15: Proves that if a convolution kernel $H(t)$ is nonincreasing, convex, nonnegative, and symmetric, then the associated Volterra operator is nonnegative definite. This ensures that expected transient impact costs are non-negative for any round-trip strategy.
Corollary 2.18: Applies this to the popular factorized model $G(t, s) = C\phi(t-s)$ , proving that if $C$ is symmetric nonnegative definite and $\phi$ is nonnegative, nonincreasing, and convex, the model is manipulation-free. This validates the use of power-law kernels in multi-asset settings.

C. Numerical Implementation

The authors provide a practical numerical scheme (Section 3.4) based on the Nyström method to discretize the integral operators. This allows for the computation of optimal strategies for complex kernels (e.g., fractional power laws) which were previously intractable.

4. Key Results and Financial Insights

Influence of Cross-Impact on Strategy

Numerical experiments (Section 3) reveal how cross-impact alters trading behavior:

Transaction-Triggered Round Trips: In the absence of signals, cross-impact can induce a "round trip" in an asset with zero initial inventory. For example, liquidating Asset 1 may trigger a temporary price drop in Asset 2, prompting the trader to short Asset 2, and then buy it back later to profit from the cross-impact recovery.
Alpha Decay Interplay: When signals are present, the optimal strategy leverages the interaction between alpha decay rates and cross-impact.
- If Asset 1 has a fast-decaying alpha and Asset 2 has a slow-decaying alpha, the trader may aggressively sell Asset 1 early. This lowers the price of Asset 2 (via cross-impact), allowing the trader to buy Asset 2 at a discount and hold it to capture the persistent alpha.
Chain Effects: In multi-asset chains (Asset 1 $\to$ Asset 2 $\to$ Asset 3), indirect cross-impact can influence the liquidation of non-adjacent assets, creating complex hedging strategies.

Impact of Frictions

Smoothing: Temporary impact ( $\Lambda$ ) smooths the trading path, preventing infinite trading speeds.
Kernel Persistence: Strategies with power-law (fractional) kernels are smoother and more persistent than those with exponential kernels, reflecting the longer memory of market impact.

5. Significance

Theoretical Advancement: The paper bridges the gap between empirical observations (power-law decay, strong cross-impact) and theoretical optimal control. It moves beyond the restrictive exponential decay assumption of Gârleanu and Pedersen and the discrete-time limitations of Alfonsi et al.
Mathematical Innovation: It introduces a novel method for solving non-Markovian stochastic control problems by reducing them to systems of stochastic Fredholm equations and solving them via operator resolvents.
Practical Applicability: The provided numerical implementation allows practitioners to compute optimal execution strategies for realistic market models involving power-law decay and complex cross-asset correlations, which are critical for high-frequency trading and algorithmic execution.
Robustness: By rigorously proving conditions for the absence of price manipulation, the paper ensures that the derived strategies are economically viable and do not rely on pathological market behaviors.

In summary, this work provides a comprehensive framework for optimal trading in a multi-asset environment with realistic, non-Markovian price impact dynamics, offering both rigorous mathematical solutions and actionable financial insights.