Price-impact Martingale optimal transport
summary
The gist
Martingale optimal transport (MOT) provides model-independent bounds for option prices calibrated to vanilla options…
In short
This work introduces Price-impact Martingale Optimal Transport (PIMOT) to find model-independent bounds for option prices when dynamic trading causes quadratic temporary price impact. It reformulates this problem as a weak optimal transport problem and shows it interpolates between standard Martingale Optimal Transport (MOT) and classical optimal transport, providing rigorous convergence rates.
Key concepts
- Martingale Optimal Transport (MOT)
- MOT provides model-independent bounds for option prices assuming frictionless trading in the underlying asset. It establishes a baseline price range based on the assumption that no temporary price impact occurs during trading.
- Price-impact Martingale Optimal Transport (PIMOT)
- PIMOT is a new framework designed to find option price bounds when dynamic trading incurs quadratic temporary price impact. It uses a dual formulation that penalizes deviations from a specific martingale, incorporating the cost of trading into the pricing bounds.
- Weak Optimal Transport (WOT) Characterization
- For single-asset, one-period problems, PIMOT is shown to be equivalent to a WOT problem with an added quadratic penalty on the conditional drift. This characterization allows researchers to use established WOT theory to analyze the pricing bounds.
- Convergence Interpolation
- PIMOT acts as an interpolation between MOT (when impact is small) and classical optimal transport (when impact is large). The paper derives convergence rates, showing how the price bounds shift smoothly between these two limits depending on the order size.
Terminology used across episodes
This episode discusses
- Price-impact Martingale optimal transport · Paper Radio
- Weak Transport for Non-Convex Costs and Model-independence in a Fixed-Income Market
- The Fundamental Theorem of Weak Optimal Transport
- Robust pricing and hedging of options on multiple assets and its numerics
- Bid--Ask Martingale Optimal Transport
The paper
Price-impact Martingale optimal transport · Read on arXiv
Erhan Bayraktar, Shuoqing Deng, Daxin Huang, Dominykas Norgilas
University of Michigan · Hong Kong University of Science and Technology · North Carolina State University
Transcript
Introduction to the show: ident: Quantitative Finance. The day in quantitative finance: models of markets, risk and prices.
Ruth: We're looking at the paper: "Price-impact Martingale optimal transport".
Warren: Here's the gist of the paper. Martingale optimal transport (MOT) provides model-independent bounds for option prices calibrated to vanilla options, assuming frictionless trading in the underlying Abstract Martingale optimal transport (MOT) provides model-independent bounds for option prices calibrated to vanilla options, assuming frictionless trading in the underlyingPrice-impact Martingale optimal transport abstract Problem Formulation and Core Result The paper introduces Price-impact Martingale Optimal Transport (PIMOT) to provide model-independent bounds for option prices when dynamic trading incurs quadratic temporary price impact In this work, we consider the case where dynamic trading incurs quadratic temporary price impact PIMOT dual problem asks for the cheapest semi-static superhedge, which combines vanilla options traded at time zero with dynamic trading in the underlying, and
8: show that the two values coincide under suitable conditionsdual problem asks for the cheapest semi-static superhedge, which combines vanilla options traded at time zero with dynamic trading in the underlying, and
8: show that the two values coincide under suitable conditions.
Ruth: The primal formulation of PIMOT is given by Dκ(Φ):= infX i,a Z φ a i dµ a i: X N i 0 X d a 1 φ a i (X a i) + G κ (H) ≥ Φ on ΩoThe superhedging price of a claim is Dκ(Φ):= infX i,a Z φ a i dµa i: X N i 0 X d a 1 φ a i (X a i) + G κ (H) ≥ Φ on Ωo.
Warren: Theorem 2.2 establishes the duality result, showing Dκ(Φ) = supQ Q∈Π(µ) EQ
Φ: − Iκ(Q), where Iκ(Q) is the largest expected trading gain under Q, net of impact costsTheorem 2.2, our first main result, gives PIMOT a transport formulation for upper semicontinuous claims of linear growth, with any number of assets and fixing dates.
Ruth: It shows that Dκ(Φ) = supQ EQ
Φ: − Iκ(Q), where Q ranges over the laws with marginals µ a i and the penalty Iκ(Q) is the largest expected trading gain under Q, net of impact costs.
Paper summary: Ruth: So, we're looking at this paper called "Price-impact Martingale optimal transport". It looks like they're trying to get model-independent bounds for option prices when there is some temporary price impact from trading in the underlying asset.
Warren: That seems like a pretty big deal. Because normally, when you do option pricing, you assume frictionless trading in the underlying market. This paper tackles what happens when that assumption breaks down and there's quadratic temporary price impact from dynamic trading.
Iris: What they claim is that they introduce Price-impact Martingale optimal transport, which is called PIMOT. It's a way to find these bounds under those conditions, and their main idea seems to be finding a superhedging price that combines options traded at time zero with dynamic trading in the underlying.
Ruth: Exactly. They prove a duality result for this superhedging price, which they call D kappa of phi. The primal problem relaxes the strict martingale constraint by penalizing how much you deviate from a specific martingale, which is pretty clever because it lets them handle those friction costs directly in the optimization.
Warren: And the dual problem shows that this superhedging price is equal to something else, which is they call sup Q of E phi minus I kappa of Q. That I kappa term represents the largest expected trading gain you can get under a specific law Q, net of those impact costs we're talking about.
Iris: And then for the one asset and one trading period case, they frame this PIMOT problem as a weak optimal transport problem, which gives them a characterization in the spirit of Wasserstein projection. They show that all optimal couplings share the same drift, and the objective gap controls how far you are from that drift using an L squared distance to the initial law.
Ruth: That's really interesting because it connects this trading problem to a known structure in optimal transport theory. It suggests that even with these specific costs, there's a fundamental structure to how these option prices are bounded.
Warren: And what they do next is show that PIMOT sits right between two other well-known concepts: Martingale optimal transport, or MOT, and classical optimal transport, or OT. They derive convergence rates to both limits depending on how strong the price impact coefficients are.
Iris: That convergence rate analysis is important because it tells us how quickly these new bounds approach the existing ones as you change those impact parameters epsilon. Specifically, they find that the gap between PIMOT and MOT is of order epsilon when you have a bounded frictionless superhedge, but it grows as one over epsilon at the OT limit.
Paper summary: Ruth: And numerically, they use SPX option data to show how the price bounds per claim change depending on how big your order size is. They find that as the order size grows, the interval between the seller's upper bound and the buyer's lower bound widens from what you get with MOT towards what you get with OT.
Warren: And they give a concrete example there, saying for k equals one, increasing the order from one claim to ten claims raises that seller’s upper bound by about sixty-one normalized basis points. That shows how sensitive these bounds are to the scale of your trading activity.
Iris: They also looked at cross-maturity results for basket options and showed that interchanging impact coefficients actually affects the upper bound, demonstrating that putting larger impact coefficients on later assets raises the price bound.
Ruth: It seems like they've built a solid framework here by connecting these concepts and giving concrete convergence rates. But they did flag a limitation: their analysis relies on assumptions about the marginal laws being square-integrable for some of those key results to hold, which is something to keep in mind for real markets.
Warren: Right, so the paper establishes this PIMOT framework by linking it to weak optimal transport and showing how it interpolates between MOT and OT, with numerical examples illustrating the dependence on order size. It sets up a way to get model-independent bounds even when dynamic trading involves quadratic price impact.
Iris: And if you look at the conclusion, they confirm that the duality between their primal and dual problems holds through a minimax argument, and they also show that there's no semi-static arbitrage under their assumptions because of some condition on those marginals.
Ruth: So what this means for us is that we now have a more detailed way to think about option pricing bounds in markets where transaction costs from dynamic trading are significant, moving beyond the purely frictionless assumptions of MOT.
Warren: It moves the discussion from just finding static hedges to dealing with dynamic execution costs. The PIMOT framework gives a structured way to quantify those costs and see how they affect the price estimates we can get.
Iris: And for future work, they're pointing towards looking at how these bounds behave in more complex settings, like basket options, which we saw affecting the results in their analysis of cross-maturity effects.
Ruth: It sounds like a solid piece of work that bridges theoretical transport theory with practical option pricing problems under realistic trading frictions. We need to keep an eye on how these convergence rates hold up when we move beyond the simple one asset and one period case they focused on.
Warren: Yeah, it's definitely a step in understanding the real-world implications of dynamic trading costs in financial modeling. The way they handle those quadratic impacts is what makes this paper worth paying attention to right now.
Conclusion: Ruth: So, this paper is about Price-impact Martingale optimal transport, and its authors are Beiglb¨ock, Henry Labord'ere, and Penkner.
Warren: Right, they’re looking at option prices under quadratic temporary price impact from dynamic trading.
Iris: It sets up a way to get model independent bounds for these prices when you assume that friction is present.
Ruth: Basically, they’re moving beyond the standard frictionless assumption of martingale optimal transport to include real-world trading costs.
Warren: They show how this new method, PIMOT, connects two other major theories: martingale optimal transport and classical optimal transport.
Iris: The key result is that PIMOT interpolates between those two limits depending on how strong the impact coefficients are.
Ruth: That interpolation means you can see how the bounds change as you dial up the impact costs or dial them down.
Warren: And numerically, they’ve shown that this whole framework depends on your order size when looking at SPX option data.
Iris: So, it tells us that even with these specific trading frictions, there's a structured way to estimate prices without needing perfect market friction-free assumptions.
Ruth: It suggests that for real markets, just using the frictionless model isn't enough; you need to account for how your own trades move the price around.
Warren: And if you can get these bounds, it gives traders a more realistic picture of what they might actually be paying or receiving on their trades.
Iris: The authors confirm that the duality holds for this new formulation, which means the math is sound and there’s no arbitrage under their assumptions.
Ruth: It's a lot of work to prove this structure, but it gives us a way to bound option prices in markets with real-world costs baked in.
Warren: The next thing we want to look at is how these bounds behave when you change the setup, like looking at basket options or changing asset order.
More episodes
- The day in conversation for 2026-11-19
- Daily Summary for 2026-11-19
- The day in conversation for 2026-11-18
- Daily Summary for 2026-11-18
- 2601.05274-On the use of case estimate and transactional payment data in neural networks for individual loss reserving
- 2009.00868-Loss-Given-Default Modeling by Post-Last Passage Time Process
- 2606.30070-Financial Resilience Evaluation: From Conditional Expectations to Dynamic Convex Risk Measures
- 2212.12687-Measuring price impact and information content of trades in a time-varying setting
- Daily Summary for 2026-11-17
- 2401.08323-Dynamic portfolio selection under generalized disappointment aversion