When a cryptocurrency futures exchange faces losses exceeding its insurance fund, it must reduce positions and distribute losses among solvent participants. This auto-deleveraging mechanism is deployed rarely but matters enormously — it determines who bears losses when the system breaks.
Note (2026-11-12; read against the paper's full text): the paper's objective is the exchange's risk, and it says it focuses on risk minimization; it cites another paper's impossibility results showing that no deleveraging rule can satisfy a full set of desiderata including fairness, and says its own work is not directly comparable. I found no claim in it that the optimal policy is the most equitable one, so the essay's closing statement that fairness and optimality coincide is the essay's. The optimality result is stated for a single asset with isolated margin (for all monotone risk measures; the unique minimizer for expected loss), the shadow-price separation is under the expected-loss objective, and the paper discusses a possible criticism that the criterion focuses on the exchange's exposure to the side being deleveraged.
arXiv:2603.15963 formulates auto-deleveraging as an optimization problem minimizing the exchange's risk from future shortfalls. The optimal single-asset policy has a clean form: minimize the maximum leverage among participants, reducing the most leveraged positions first.
The solution is a water-filling algorithm. Imagine leverage as the height of water in columns of different widths. The optimal policy pours water out until all columns reach the same level — the most leveraged accounts are reduced first, and the process stops when risk is equalized. This is not a heuristic. It is the provably optimal solution to the risk minimization problem.
The design resists wash-trading and Sybil attacks because splitting an account into two accounts does not change the water-filling outcome — each fragment gets the same treatment the whole would have received. The solution is also distribution-free: it does not depend on assumptions about the distribution of future price movements.
In multi-asset cross-margin settings, the problem remains separable using asset-level shadow prices. When risk is driven by a dominant factor, the optimal policy takes the same water-filling form but in a factor-adjusted notion of leverage. Hedged portfolios are treated less aggressively than gross leverage alone would suggest — the optimization distinguishes between leverage that creates risk and leverage that manages it.
The elegance comes from the fact that fairness and optimality coincide: the policy that minimizes systemic risk is also the one that treats participants most equitably.