University of Texas at Austin finance professor John Griffin and graduate student Amin Shams just posted a paper suggesting that cryptocurrency prices are manipulated. The paper has received a great deal of attention in the media, but there is a disconnect between the paper and the press coverage in terms of quantification.
The authors suggest the cryptocurrency exchange Bitfinex buys bitcoin with another cryptocurrency — tether — to push up Bitcoin prices. How much? Four basis points per 100 bitcoin. With Bitcoin at $10,000, for example, that means Bitfinex spends $1 million to push the price up to $10,004.
The authors assert that the purchases are not random in that they occur more often after Bitcoin prices have fallen. How much? After the biggest drops in Bitcoin price, 1 Bitfinex buys 72 extra bitcoin. More than 100,000 Bitcoin frequently trade in an hour, and often much more during periods of high volatility.
The authors examined the 87 hours with the largest flows of bitcoin and tether and found that the hours following them accounted for 50 percent of the “meteoric rise” in bitcoin. That sounds impressive, but the rise over the period discussed was from about $1,000 to $8,000. The authors used compounded returns, which mean returns in those 87 hours averaged 1.2 percent. For the S&P 500 Index, a 1.2 percent move is a big hour. For Bitcoin, not so much.
To an academic, it’s often more important that a result have “statistical significance,” which is to be clearly something other than random noise, than “practical significance,” which is to be of a size that would interest a trader or regulator. But, here too, the results are less than impressive.
The four basis-point increase in Bitcoin prices explains less than 1 percent of the variance. While it technically meets normal standards of statistical significance, given the guesswork the authors had to do to collect data and the complexity of the analysis, I consider it more likely to be noise than signal.
The authors have only 20 data points to support the claim of 72 extra Bitcoin, and they fit them using a model with six parameters. The usual rule of thumb is you want 30 observations per parameter to rely on the results, and that’s not taking into account the data and complexity issues mentioned above.
The 50 percent “meteoric rise” claim is declared significant because the authors picked 10,000 sets of 87 hours at random, and none of them averaged a 1.2 percent return for Bitcoin. That is indeed strong evidence that the 87 hours following the hours with the largest Bitcoin and tether flows are not random. But we already knew they weren’t random, they were times of high transaction volume, which would be expected to have more volatile prices than average, and trading volume was much higher on the upswings than the downswings.
I don’t mean to be completely negative on the paper. I’ve played with the data the authors use, and it is very frustrating. There is a lot of work — including a lot of guesswork, unfortunately — to compile numbers that can be used to test hypotheses. The tests themselves are complex. Moreover, it’s likely that the phenomena under study are evolving rapidly during the study period. As a result, any conclusions have to be taken as suggestive only. But in areas as little understood as cryptocurrency trading dynamics, even suggestions are valuable. (Full disclosure: I own Bitcoin and other cryptocurrencies.)
I’ve only touched on the headline claims of the article. There is a lot more substance in it. Some of the claims that got less attention have better support. Anyone seriously interested in these issues should read the paper, not as an authoritative last word, but as a plausible account backed by some data.
I also don’t mean to deny that cryptocurrency prices are manipulated. I’m a crypto-believer, and am impressed by the huge amount of honest developer talent and innovative vision improving the crypto code base every day, whether prices are soaring or crashing, more for love than money. That is where I put my faith. But there are plenty of crooks and profiteers as well. It would be extraordinary if some of them weren’t trying to manipulate prices — and worse.
Moreover there are obvious inconsistencies in the complex and novel cryptocurrency markets. That makes it unclear whether behavior is manipulation or rationalization. This paper is a small step in improving our understanding, not an indictment of tether or cryptocurrencies.