BitMex Regains Price Leadership Before DDoS Attack
On BitMEX, in the early hours of Friday 13 March, a whale moved on board with $200 million worth of BTC and a few minutes later a DDoS attack drove the price of bitcoin down to 3,600 USD.(see our article on this) But unusual events were unfolding already, the day before…
Since the study of Alexander et al.[1] the price leadership of the exchange’s perpetual swap has been diminishing, in favour of OKEx and Huobi.[2] Here we use the same econometric models with ultra-high-frequency price and volume transaction data to show that BitMEX regained its dominance during the crash on March 12, the day before the DDoS attack.
Figure 1 depicts the price of the four most heavily traded bitcoin USD derivatives — the quarterly futures on Huobi and OKEx, and the perpetual swaps on OKEx and BitMEX. Between 10:30 and 11:00 on 12 March the dollar price dropped from around $7,250 to $5,500, briefly falling even further before recovering. Overall, this sharp price decline reduced bitcoin’s market cap by more than $20 billion in just 2.5 hours and led to record highs in the bitcoin volatility index.
During the same interval we witnessed record-breaking trading volumes: a total of more than $10 billion on all four products over the period.[3] See Figure 2. Huobi futures were the most-traded product, reaching a peak of $120 million in just one minute, at 10:38. But the BitMEX perpetual traded $80 million notional during the preceding minute. One can see from the figure how the volumes really start to rack up on these two products as selling pressure mounted — starting at about 10:15, well before the price actually fell.
Price Leadership
A four-dimensional price discovery analysis of the instruments’ influence, based on minute-level transaction data for the period from 09:30 to 12:00, using a vector error correction model (VECM) with price discovery metrics.[4] This shows that the BitMEX perpetual was responsible for 64% of the total price leadership, followed by Huobi and OKEx futures, each with around 15%. The OKEx perpetual was almost entirely a price follower.
Next, we limit our analysis to the interval from 10:00 to 11:10 — the period of the main price drop — and at the same time, we increase our frequency to second prices. However, we have to exclude the OKEx perpetual due to low trading volume. This UHF analysis confirms the leading role of the BitMEX perpetual (49%) over Huobi (27%) and OKEx (24%) futures.
Impulse Response
High-frequency traders are interested in the speed of information transmission between the two leading instruments during the price drop. Given the unusually strong price leadership role of the BitMEX perpetual on March 12, jumps up unexpectedly by 1%, how long would traders on Huobi have to follow this move and still make a profit?
To answer this question we use impulse-response analysis in a 2-dimensional VECM assuming a price jump of 1% on the BitMEX perpetual. The output is the expected percentage spread between the BitMEX perpetual and the Huobi futures, immediately after the shock and during the following minutes. The spread is shown in Figure 3 below.
Within the first 5 seconds after the shock, the Huobi futures make a large adjustment of 40bps. Subsequently, the spread continuously narrows and converges to -7bps after 10 minutes.[5]
Clearly, during the huge price drop in the morning of 12 March the BitMEX perpetual took control of the driving seat. But what was the reason? Was it just caused by panic about the COVID-19 virus (and, more importantly, the response to this shock in the US markets) or was it manipulation? Could be it associated with the DDoS attack on BitMEX on 13 March?
Order Book Imbalance
An examination of the BitMEX perpetual order book allows us to dig deeper into traders’ operations during the morning of March 12. Given that we want to examine selling pressure, we focus on the ask side here, although the order book skew indices that we have developed are applicable to either side. These are called λ.OBSA for the ask side, and λ.OBSB for the bid side.
Each index is calculated every minute as an exponentially weighted median — mean difference where the mean (c) and median (d) are calculated as follows:
(a) Record the 20 best ask price levels, including the corresponding volume on each level;
(b) Weight the volume on each of the 20 levels exponentially, depending on their distance from the best ask price;[6]
(c) Calculate the average of these 20 (weighted) volume figures; and
(d) Calculate their median also.
The rationale behind these calculations is that the difference between the mean and median captures the skew in this side of the order book. Normally, the ask side of the order book has positive skew (median < mean) but if the skew increases disproportionately (median << mean) some unusually large orders can be appearing just above the ask price.
The mean and median results as well as their difference — i.e. the OBSA index (with λ = 0.95) are shown in Figure 4. Until 10:30, mean and median ask volume were quite large and close together. This indicates continuous selling pressure on the book, but no unusually large sell orders. However, for several minutes around 10:38 and 11:40 the mean and median differ widely. That is, there were some extremely large sell orders inside the book, and it is these that finally brought down the price.
Manipulation
To see whether these unusually large order sizes were aimed at manipulating the BitMEX perpetual price, we extract (for closer examination) all minutes with a sell-order volume greater than $5 million on any of the 20 best ask price levels. That is, we look for a very large sell order inside the book but not too far from the best ask price.[7] For each of these very large orders Figure 5 depicts the percentage difference (in basis points) between the order’s ask price and the mid price at the time. We also include the respective time.
The five very large volumes at (or very near) the best ask price were probably placed by traders (two on BitMEX, two on Huobi and one on OKEx) who wanted to close their long positions as fast as possible after the price plummeted between 10:30 and 10:48. However, after this time we also see several very large offers on BitMEX, all quite deep in the book and most with a size exactly equal or very close to the maximum order quantity of $10 million.
This type of racking up of artificial liquidity is par for the course, in all the less-regulated bitcoin markets. But it doesn’t often happen on such a large scale as it did on the BitMEX perpetual on 12 March 2020.
Carol Alexander & Daniel Heck
@CryptoMarketRisk, QFIN, University of Sussex
[1] Alexander C., Choi, J., Park, H., and S. Sohn (2020) ‘BitMEX Bitcoin Derivatives: Price Discovery, Informational Efficiency and Hedging Effectiveness.’ Journal of Futures Markets. 40 (1). pp. 23–43. ISSN 0270–7314 .
[2] https://www.linkedin.com/feed/update/urn:li:activity:6643866814826721280/
[3] For comparison, Coinbase — the most liquid spot exchange — had a maximum trading volume of only $5 million during this interval.
[4]The full econometric methodology is described in Alexander et. al. (2020), and references therein.
[5] Seven basis points is a typical spread between these products, because of the basis. It will be positive or negative depending on the contango or backwardation in the futures term structure, which changes over time.
[6] That is, we choose some parameter λ between 0 and 1 and multiply the volume on the second best ask price by λ, the volume on the third best ask price by λ2, and so forth. The exponential-weighting parameter λ is used to turn up or down the effect on the index of large volumes that are deep in the order book — they have less effect on the index as λ decreases.
[7] If some trader wanted to manipulate the price by placing a very large sell order without the intention to execute — a method of manipulation called spoofing — he wouldn’t place it aggressively near to the current best ask price, but rather a few dollars above.
