ak0@annanay
ADL (Auto-Deleveraging) [the fun part]
When designing our ADL system, we noticed a critical flaw in current implementations (it has everything to do with the chart attached).
No matter what others tell you, ADL is necessary - even in regulated, tradfi markets (eg ‘tear-ups’ in futures markets). Without it, the exchange itself risks bankruptcy and unlimited liability to all users.
Researchers like @tarunchitra and @danrobinson have written about how subtle and complex these systems can become. ADL is strictly a last-resort mechanism, but when designed well, it can aid market design and price discovery, not hinder it.
At a high level, ADL socializes losses across profitable traders when the exchange can no longer safely close losing positions.
Instead of closing the losing trader’s position in the market, the exchange forcibly matches it with a profitable trader on the other side.
This means the winning trader:
has their position partially or fully closed, and
and may lose part of their unrealized profit (aka variation margin gains haircutting).
For clarity, we refer to traders who absorb these positions as ADL Liquidity Providers (ALPs).
Designing a fair ADL system raises several difficult questions:
>How should the ADL trade price be determined?
>Which traders should be selected as ALPs?
>How should liquidation quantities be distributed across ALPs?
>How do we guarantee the system converges to a stable state with no remaining positions?
A natural first idea would be to close all ADL trades at the current mark price.
If we settle positions at the current mark price, we are simply converting unrealized PnL into realized PnL. The trader’s equity does not change, which means a trader who has already gone negative would remain negative.
Therefore, the ADL mechanism must close some positions at a better price for the liquidated user so that their equity can be restored to a non-negative level.
One way to achieve this is to track the last mark price at which the trader still had positive equity during the previous liquidation cycle. If we close the position at that price, we effectively restore their account equity to what it was at that point.
This has problems. Consider a trader who profited on AAPL, but lost heavily on TSLA.
If we revert both positions to the previous mark price:
>the TSLA loss gets reduced, which is necessary to restore equity,
>but the AAPL profit also gets removed.
Now remember that every long has a short on the other side.
If we revert the AAPL profit, the short counterparty would regain the loss they previously incurred, effectively receiving free money, even though their position had nothing to do with the liquidation problem.
This would create unnecessary wealth transfers between unrelated traders.
The correct approach is therefore asymmetric:
>losing positions are settled at the previous mark price to cover the loss.
>winning positions are settled at the current mark price, leaving those profits untouched.
This guarantees that only the losses required to stabilize the system are socialized, while profitable trades that are unrelated to the liquidation remain unaffected.
In other words, ADL redistributes only the losses, never the profits.
ALP Selection
Many exchanges determine ADL priority using the metric
profit × leverage
The intuition is that traders with the highest profits and the highest leverage should be the first to absorb ADL trades.
At first glance, this seems reasonable. However, this rule is fundamentally a heuristic with no clear mathematical justification, and it leads to unintuitive outcomes.
In fact, we can formally show that this metric can mis-rank traders in the ADL priority queue.
Specifically, there exist cases where an ALP with percentage return < R is ranked ahead of another ALP whose percentage return > R.
In other words, a trader who has achieved higher returns on their capital can still receive lower ADL priority than a trader with worse performance, purely because of the way the profit × leverage heuristic behaves.
See the graph attached.
This is tends to affect new traders with small positions. A user who has just started trading and makes a modest profit can suddenly appear near the top of the ADL queue and get force-closed early.
From a UX perspective, this is extremely undesirable: you start trading, make a profit, and immediately get ADL’d.
That is not the behavior we want from a fair liquidation system.
At QFEX, we rank ALPs purely by effective leverage, removing the profit component entirely. Users with the highest leverage across their positions are prioritized for ADL.
This aligns the ADL mechanism with the actual objective of the risk engine:
system-wide de-risking.
Traders with the highest leverage contribute the most to systemic risk, so reducing their exposure first naturally stabilizes the system.
You might now wonder:
What happens if the selected ALP does not have enough profit to absorb the losses?
We address that tomorrow.
