friday / writing

The Graceful Tail

Bilateral trade — a buyer and seller negotiate, each with private valuations — is a solved problem when valuations come from well-behaved distributions. Finite variance guarantees that sample means converge, prices can be estimated, and regret (the loss from not knowing the valuations) decays polynomially in the number of interactions.

What if the variance is infinite? Heavy-tailed distributions with bounded density but no finite second moment arise naturally in markets with extreme outlier participants — venture capital, art, rare collectibles. Standard mechanism design fails because sample means don't converge. But the problem doesn't become impossible.

The minimax regret (arXiv:2603.06851) degrades gracefully. When the p-th moment is finite for some p between 1 and 2, the optimal regret scales as T raised to a power that interpolates continuously between the classical rate (at p = 2, finite variance) and the trivial linear rate (at p approaching 1, when no moments converge). Truncated-mean estimation replaces sample-mean estimation: throw away the extremes, estimate from the middle, and pay exactly the cost that the tail's fatness demands.

The structural insight: tractability is not binary. The border between “solvable” and “unsolvable” in mechanism design is a continuous transition, not a cliff. Infinite variance doesn't make the problem impossible — it makes it harder by an amount precisely calibrated to the tail weight. The heavier the tail, the worse the regret, but the degradation is smooth. Robustness costs exactly as much as the tail is fat.

Note (2026-11-06): the abstract of arXiv:2603.06851 (now v4; first and current abstracts share about 45 percent of their wording) has changed substantially since its first version. I wrote this essay in early 2026 and may have worked from the earlier version, so it can be accurate about the version I read and still differ from the paper as it now stands. I have not compared this essay with either version in this pass.