Liquidation Matters

Why TradeLayer's Design Is Anti-Fragile

In traditional derivatives clearing, the discrete problem of liquidation orders being left standing, and potentially causing a systemic shortfall, is not considered a problem because your broker, who is a clearing member, will sue you for the negative balance. Those are also more liquid markets, though BTC swaps & futures are now rivaling gold in terms of daily volume, but you couldn’t dump 100M of BTC contracts in a liquidation without dramatically impacting price. The nightmare of every market structure professional, is the *cascading* liquidation. This is where the quantity of firesales jumping to the front of the queue, N, is greater than the amount of bids on the book, B, that are above a price P, where the avg. trigger price of the next cluster of liquidations waits for the moment to strike. If your liquidations are too big, they trigger more liquidations, the whole thing behaves like an avalanche.

The long wicked candlesticks in mid-May, and in particular that short-covering blow-out that only occurred on Binance in late July, are examples of what happens when liquidations cascade. The phenomenon is so reflexive, mechanical and dramatic that these wicks are almost always good trading opportunities for those with the liquidity to take the other side.

When it comes to systemic risk for an exchange, or better yet a decentralized system in which there are retail savers at the end of the flow, we want to be as careful as possible. We can’t go around suing people, we need a system that works with available collateral and retains solvency in the face of extreme volatility. For example, a -50% daily move in TOTAL or ALL vs. BTC or LTC (respectively) would put a tremendous stress on the contracts we’ve built, what’s the best way to handle that?

1) Our default clearing model clears every single block, so there is no delay in realized or unrealized PNL becoming available balance. This is great for spread-trading.

Fast-clearing is a bottom-up approach to portfolio margining, as opposed to a top-down approach like Deribit’s where they analyze options based on Kurtosis and like-kind instruments blend down to a projected PNL at +/- 15%. That’s quite sophisticated, it does a decent job of protecting Deribit’s balance sheet while allowing option MMs to be net short lots of far OTM inventory, while dynamically delta hedging. Our options are going to end up being slightly more prosaic, the far OTM wings will be 100% cash margined (which isn’t *so* bad for a 70% OTM put), because there’s no elaborate on-chain orderbooks mechanism for myriad option contracts in that design, so no way to auto-liquidate naked-short contracts, they will all trade in channels only. But the real-time PNLs *will* blend with whatever swap or future you’re using to delta-hedge the options, and thus, it’s pretty good for being DeFi!

2) Individual liquidations, if unfilled, will group together.

Imagine the Wall of Doom, millions of contracts in bold red font, grouped together in a single order. BitMex invented this, it keeps re-rating the collective liquidation order to a lower (or higher, if we’re talking about blown-out shorts) price where the average break-even would be. Eventually the orders are filled, or at the 8hr. mark, BitMex would pay out the Insurance Fund.

If one doesn’t do this, two bad things happen. First, the earliest liquidations that happened with very leveraged traders early in a drop, will maybe never retrace to those prices, not anytime soon. Second, it’s possible to have a cluster of long liquidations from a drop and then another cluster of short liquidaitons on a very violent bounce, as late-shorters get rightfully punished. But not-rightfully, the insurance fund would be on the hook for shortfalls from both sides. That’s fragile. The Wall of Doom is the most anti-fragile way for the system to manage the hung-inventory.

3) Make liquidated orders counter-party to short-contract inventory in the insurance fund.

This one is for Native TOTAL/BTC and ALL/LTC perpetual swaps, we will have the more primal insurance fund in the protocol have a 50/50 inventory (yes, like an AMM) and so as the BTC price of TOTAL or the LTC price of ALL increases, it can steadily place at-the-money limit orders to get filled as a maker and get more synthetic BTC or LTC. As liquidations happen, the fund gets >50% long ALL or TOTAL, but then as the Wall of Doom eventually gets punched through, suddenly the fundamentals of the protocol are more engaging, the market pushes up through the level, possibly driving a very reflexive surging rally. Frankly, that’s fun, and it’s desired behavior, especially after a brutal day in recent memory where the Wall of Doom order originated.

4) Priority Clearing

Liquidations then Stop-loss orders get priority in matching with any on-chain orders. People who throw out a trade with a 0 price will be included ahead of the Insurance Fund to act as liquidity-provider-of-last-resort. Quants can model the likely liquidations and stop triggers based on analyzing the chain (though caching trades inside channels and not publishing limits this information leak). Thus trading systems can be built around probabilistically bidding for firesales and selling borrowed spot inventory on another venue (or visa versa if the liquidations are of shorts).

5) Dynamic Liquidation Penalty

Liquidations that can clear in the orderbook should do so at a price where the insurance fund can pocket some % of the notional, e.g. 0.65% (Deribit, seems to be not quite enough) or 1% or all-remaining-maint.-margin (variable) which is what made the BitMex fund large and hated. We’ll adopt a variable approach based on the Average True Range of mark prices in some time horizon, perhaps weekly. It would generally fluctuate between 0.5% and 1.5% depending on volatility, and would grandfather in as a property of any trade that occurred on the given day (e.g. open a long one week with a 0.5% penalty, it’s locked in, get liquidated 3 months later after missing out on profits in the dramatic sell-off). That detail is a little in-elegant but it prevents anyone from manipulating ALL or TOTAL spot to the point where volatility loosens up average market depth and combining that with a wash-trade attack, where two high leverage positions trade out. You get liquidated on one side and try to get a profit from the resulting chaos on the other side. Insurance Fund Liquidation Penalties act as a binary option premium to price out that quantum-level market structure abuse.