Cascading Margin Calls and the Flash Crash

Hi Everyone,

This week markets have tumbled from $590/BTC to $500/BTC, seeing a low of $450/BTC.

Thursday at around 1am PST, bitfinex saw a flash crash taking the price to $450/BTC.  This flash crash was the result a series of very large market sell orders on bitfinex at around $525/BTC which further triggered cascading margin calls on the way down.  

Over 9000 BTC were sold in a matter of minutes, 650 BTC of those being liquidated leveraged long positions due to the cascade of margin calls.  While 650 BTC liquidated in margin calls is only about 7% of what was entirely sold, their presence is enough incentive for the initial dump.  In other words, given that you know that $5/BTC below the current price is a series of cascading margin calls worth about 650 BTC, how much would you be willing to sell ahead of it if you think there is a good chance for a slight drop in price or you had the long term intention to sell anyway?  And also, stop-losses on the way down would also cascade in the same way as liquidated margin positions.  Moreover, given that many traders have arbitrage algos running on multiple exchanges, stop-losses in non-bitfinex books could still end up impacting the bitfinex selloff once the arb bot would mirror some part of the foreign stop-loss sell as one leg of the arb. [Also, as a slight tangent, given that Ethereum will be liquidating some of their pre-sale bitcoins to pay back debts and use for development, getting in front of those coins is also worth some value.  4150 BTC will be withdrawn from the exodus address some of which will be used to pay off debts and prior expenses and establish development teams in Berlin and Amsterdam.  It is unclear how much of that 4150 BTC will be liquidated and how much will be kept in BTC].

Looking at bitfinex's swap market, active USD Swaps almost doubled between late May and early July.  This suggests that many traders entered into leveraged long positions somewhere between $560-$680.  Given the maintenance margin of bitfinex as 15%, margin calls should be spread between $476-$578.  While the slower selling pressure throughout the week above $525 did not trigger a cascading effect, the sharp sells around $525 did.

On the way down, bitfinex uses an algo to throttle the price (as an alternative to circuit breakers).  It works something like this: if an order would drop the price more than 10% in a minute, execute part of the order to take the price down 10%, then wait a few seconds, then execute another part of the order to take the price down 10% more, and wait a few seconds and repeat.  This is to give the market some breathing time to refill the bid wall before the large sell market order continues its way down the orderbook.

I suspect this also gives bitfinex some time to insert their margin call liquidation orders in front of individual chunks of the large market sell order.  If they are doing this, it would give added protection to swap underwriters since margin-called positions would be liquidated at higher prices and would also give traders who were forcibly closed by margin calls smaller loses due to a better price.  The person who loses out is the person executing the large market sell order which triggered the crash.

If bitfinex, instead, gives priority to the large market sell order over the margin call liquidation orders, then the large seller benefits from less slippage at the cost of the intermediate buyers on the way down who refill the bid walls during each "speed bump" (although there is less slippage, the overall price could be worse since market orders inserted between successively executed chunks of the initial order could depress the price more).

Bitfinex gives the second explanation for why they have this "price throttling".  They suggest that any market order of that size is likely to be manipulative since no rational person would prefer a worse execution price over breaking up the market order for a better price.  Therefore, they throttle it.

Personally, I don't buy that explanation.  Market manipulation is an ominous term which can mean anything from spreading false rumors to painting the tape to hunting for stop losses.  What some people call manipulation, others call good strategy.  In any case, "manipulation", real or not, is very difficult to pull off and usually the manipulator takes on exorbitant risk which results in huge losses if the market does not react how he wants it to react.  Regardless of which practices we term manipulation, there are good non-manipulative reasons for executing a large market sell in the face of cascading margin calls below you.

I think bitfinex is just looking out for its swap writers' solvencies and, thus, indirectly their own interests.  If there are full or partial defaults in bitfinex's swap market, people would be less willing to write them and those that do would require a higher premium.  This would make leverage more expensive and reduce margin trading volume altogether.

What this whole episode shows us is that the bitcoin markets are becoming more automated and more intricately interwoven.  Already fundamental news has stopped affecting the price and now we are starting to see flash crashes (much like the real world markets).  Maybe bots are getting progressively more tangled up with other bots and there will be more flash crashes to come.

Other news:
  • Bank of Canada came out with a paper which looks at arbitrage opportunities in the bitcoin markets and finds arbs no longer exist:
    • Unfortunately, the paper's methodology was questionable.  They use daily closing prices for their historic data which makes no sense since arbitrages don't last 24 hours barring special events like the China ban.
    • As a side note, if a market seems to have no arbs possible, it's because arbitrageurs are doing those arbs.  So tautologically, the lack of apparent arbs is evidence of the presence of arbs being done, so the question itself is rather pointless.  A better question would be, how fast do you have to be to capture the arbs from the other arbitrageurs?
  • Jed and Ripple strike a deal to avoid immediate selloff of Jed's $9bn XRP:
    • I wonder what Jed got out of that deal.  Maybe Ripple agreed not to sue Stellar.  Not that there is a strong basis to sue over open-sourced tech.
  • Viacoin unveils ClearingHouse, decentralized p2p trading on the blockchain:

Kevin & Team Buttercoin
Bitcoin Trading Made Easy | 

Buttercoin currently in Private Beta, if you'd like access right now apply here: