Robust Decision Making Under Knightian Uncertainty

This direction is split into other directions: Knightian uncertainty. Its last version stays readable and citable here.

Version 1 of 1114 papers145 conceptsfrozen 2026-10-04

Standard probability theory is built on a comfortable assumption: we might not know the outcome of an event, but we know the rules of the game. We know a fair coin has a 50% chance of landing heads. In economics and mathematics, this is known as risk.

However, real-world systems—from climate-impacted aviation to emerging AI markets—operate under Knightian uncertainty (or radical uncertainty). Here, we don't even know the probabilities. We aren't just unsure if the coin will land heads; we aren't sure if the coin is fair, or if it's even a coin at all.

Research in this field develops mathematical frameworks to make decisions that remain effective even when our underlying models are wrong.

The Limits of Classical Probability

In classical models, such as those used in finance or routing, we typically assume a single, fixed probability measure. As Marcel Nutz and Ariel Neufeld point out in Nonlinear Lévy Processes and their Characteristics, this assumes we know the parameters of the process (like volatility) perfectly.

If those parameters are even slightly off, the "optimal" strategy can become fragile. This is the "gap" identified by R. Wilkinson and colleagues in Managing uncertainty – principles for improved decision-making: traditional risk quantification fails when variables are fundamentally unknowable, leading to catastrophic failures in complex systems like geopolitics or climate change.

Defining the Neighborhood of Uncertainty

To solve this, researchers shift from a single probability measure to an ambiguity set—a collection of many possible probability distributions. Instead of asking "What is the best move given this distribution?", we ask "What is the best move that performs well across all plausible distributions?"

One powerful way to define this "neighborhood" of models is through Wasserstein Ambiguity Sets. Haochen Wu and Max Z. Li apply this to the Airport Ground Holding Problem, creating schedules that are robust to weather patterns that no longer follow historical data. Similarly, Jan Obloj and Johannes Wiesel use Wasserstein balls in Robust Estimation of Superhedging Prices to ensure financial hedges remain stable even during market crises where standard models break down.

Nonlinear Expectations and G-Brownian Motion

When we don't know the "true" probability, we can no longer use standard linear expectations (the simple weighted average of outcomes). Instead, we use sublinear expectations (or nonlinear expectations).

Think of this as a "worst-case" expectation. If you are uncertain about the volatility of a stock, you don't just pick one volatility value; you calculate the expectation over a range of possible volatilities. Yan Dolinsky, Marcel Nutz, and H. Meté Soner provide a foundational bridge for this in Weak Approximation of G-Expectations. Their work extends the famous Donsker’s Theorem—which allows us to approximate complex continuous processes with simple discrete ones—to a world where volatility is uncertain. This framework, often involving G-Brownian motion, allows for pricing and hedging without assuming a single fixed path for market fluctuations.

Decision Making as a Game Against Nature

A recurring mechanism for finding these robust solutions is framing the problem as a zero-sum game against nature. In this framing:

  1. You choose a strategy to maximize your return.
  2. An adversarial "Nature" chooses the worst possible probability distribution from your ambiguity set to minimize your return.

Stefano Ermon and Carla P Gomes demonstrate this in Playing games against nature, showing that for renewable resource management, the optimal robust strategy is often a "pulsing" policy—periodically stopping harvests to allow for recovery—which guarantees higher returns than any strategy based on a single (potentially wrong) environmental model.

This minimax approach is also central to Optimal stopping under adverse nonlinear expectation by Marcel Nutz and Jianfeng Zhang, which provides the mathematical proof for finding equilibrium values when the "stopper" and the "controller" (Nature) have conflicting interests.

Ensuring Time Consistency

One of the hardest problems in this field is dynamic inconsistency. If you make a "robust" plan today, will it still seem like the best plan tomorrow?

Standard robust measures can lead to "irrational" behavior where a decision-maker changes their mind simply because time has passed, even if no new information has arrived. Jan Obloj, Sigrid Källblad, and Thaleia Zariphopoulou address this in Dynamically consistent investment under model uncertainty. They develop "forward performance criteria" that allow an investor to update their strategy over time while remaining robust to model ambiguity, ensuring that the strategy remains optimal as the market evolves.

Beyond Finance: AI and Social Networks

While much of this theory originated in mathematical finance, its applications are expanding. Jordi Grau-Moya explores how meta-learning can produce agents that are sensitive to "unknown unknowns" in Beyond Bayes-optimality, allowing AI to recognize when it is in a novel environment where its training no longer applies.

In social networks, Angelia Nedic shows in Online discrete optimization in social networks that decentralized agents can still reach optimal collective decisions even when they have zero prior probabilistic knowledge of their environment, effectively "learning" the structure of the world through local interactions despite Knightian uncertainty.

Papers in this version

Showing 60 of 114 papers.

History & citationVersion 1 · 2026-10-04

Split into other directions · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_robust_decision_making_under_knightian_u_29fc3427,
  title = {Robust Decision Making Under Knightian Uncertainty},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/robust-decision-making-under-knightian-uncertainty/rdr_29fc342722b2441ab97a17ff48c250b5},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_robust_decision_making_under_knightian_u_29fc3427_v1,
  title = {Robust Decision Making Under Knightian Uncertainty},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/robust-decision-making-under-knightian-uncertainty/rdr_29fc342722b2441ab97a17ff48c250b5/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}