Caution Advised When Using Bitfinex

Update August 30, 2015: After one year Bitfinex might not have defaulted, but the platform is struggling in some areas. The site’s continuous poor performance has caused numerous losses, while the exchange also got hacked in May 2015 losing 1,400 BTC. These events largely confirm the risk of trading at the platform as argued below. A full review on the exchange can be found here.



Bitfinex is a cryptocurrency exchange that markets itself as the most “advanced financial place in the Bitcoin world.” But a close look at one of the services offered by the exchange, the possibility for providing liquidity, reveals that users have a very good reason to be cautious or even avoid using the exchange operated from the British Virgin Islands (BVI).

Loans

On Bitfinex users can take on loans for a default period of 30 days, at a rate of currently 0.1078 percent per day. They can be used by traders to leverage trading positions in margin trading. The loans are offered by other users who want to earn money in a safe way, by providing liquidity. If a trader’s position loses money, then a margin call will trigger and eventually lead to a force-closure of the position. This happens well before the funds required to pay back loans are even touched. For the lender, this results in a risk free return. The lender has no other obligations towards the trader.

Credit spreads

As all offered loans and respective interest rates can be observed, it becomes possible to derive some information from the rates involved. To this purpose, the credits spreads have to be taken into consideration. These would be equal to the difference between the return on a 30 day loan via Bitfixed and the return on 1 month Treasury Bills. The latter return is considered the risk-free interest rate over this period, currently equal to 0.03 percent. Treasuries are considered risk-free because they are backed by the U.S. government.

Loans made on Bitfinex are also risk-free according to the platform. Even so, the interest rates are a lot higher than 0.03 percent for 30 days. In fact, the rate was 0.1078 percent per day. For 30 days, this comes down to 3.29 percent rather than 0.03 percent. Liquidity providers pay 15 percent of their income to Bitfinex, hence effectively interest rates are 2.79 percent. This might seem odd, as higher returns must be directly related to higher risk, but is mainly caused by the fact that loans made on Bitfinex are not as risk-free as the platform claims they are. After all, if the exchange suddenly defaults or disappears, then all money present on the platform might be gone as well. Since there is no counterparty risk on the traders due to the exchange’s system, the full credit spread is caused by counterparty risk on the exchange itself.

Implied Default Probability

To give further meaning to this, it is required to know that a credit spread is equal to the probability of default times the loss given default. In reality, only the credit spreads can be observed. By estimating the loss given default, it is possible to derive the implied default probability. In other words, the market derived probability that Bitfinex will default over a certain period of time.

For 30 days, the credit spread on Bitfinex loans is equal to 2.79 percent minus the 0.03 percent risk-free rate. To get the implied default probability, the remaining 2.76 percent has to be divided by the loss given default. This would be the percentage of money lost if a default actually occurs. This is another estimation, but given that Bitfinex is based in Hong Kong, retrieving any money is likely to be difficult in the event of a default. Because of this, it is a fair assumption to estimate losses at either 100 percent or 75 percent (the standard percentage applied by ISDA for subordinated debt) if Bitfinex defaults. The probability of default of Bitfinex over a period of 30 days can then be determined at 2.76 percent or 3.68 percent respectively.

Note that the probability of default increases if the loss given default decreases in this calculation. This is due to the fact that the credit spreads in reality follow from default probabilities, and not the other way around. If the probability of default would remain static, a decrease in the loss given default would result in a lower credit spread.

Using the same methodology, the annualized interest rates on Bitfinex loans would be equal to 39.70 percent. The risk-free rate for 1 year Treasuries is just 0.11 percent, hence the both the credit spread and implied probability of default would be 39.59 percent. If a loss given default of 75 percent is assumed, then the implied probability of default would be 52.79 percent over a one year period.

Caution Advised

Given the previous, the markets imply a substantial chance that Bitfinex will default over the coming year. Simply holding Bitcoins is actually a lot less risky. The current 1 day volatility of Bitcoin, based on the past 30 days, is just 1.15 percent. This can be annualized by multiplying with the square root of 365 (days), which results in nearly 22 percent. This figure implies a 16 percent chance that Bitcoin will lose more than 22 percent over a one year period. A loss of more than 66 percent will statistically only have a 0.135 percent probability. The conclusion is that providing a USD loan via Bitfinex is a lot more risky than holding Bitcoins, not to mention trading and holding Bitcoins via Bitfinex.

  • Yeah, there seems to be a growing concern among many that the Bitfinex “credit bubble” will not end well

    https://bitcointalk.org/index.php?topic=667105.0

  • bitcoinbravo

    Complete garbage analysis — how does a USD swap rate equal a credit default swap(CDS) ?also what do you have to say then to the BTC swap rate of 0.0055% ? You see how as a trader I completely curb stomped your model and threw it into the worthless scrap bin of heap of economic models by so called “economists”

    • You throw this under “garbage” even though you cannot even tell the difference between a loan and a swap. In a swap agreement, variable performance is swapped for a fixed-rate. The USD lender does not get any variable performance in these so-called “swaps” hence they are plain loans. In a BTC swap the lender does get exposed to BTC performance, so there is no way they can be compared to USD loans.

      • Maximo Joshua Rossi

        I think YOU should have looked at any of the easily available data that could have tested this hypothesis. It is as if you had the hypothesis it is raining, checked the news and they said it was, wrote about it without even looking out the window. You make a claim, you should substantiate it. Efficient markets is not generally viewed as valid, and the examples that debunk that theory are so many as to be impossible to fit in this forum.

        • I’ve checked my story, but I’m not going to re-invent the wheel every time I write about a car. The argument that efficient markets are generally not seen as valid is only a half-truth. Maybe in the most strict form, but otherwise it allows for deviations around the true value. If you are going to argue a significant structural excess return on interest rates due to demand, then you really need to bring the big guns. A free lunch still does not exist in economics – no return without risk and definitely not for long.

  • Maximo Joshua Rossi

    Also, and I don’t mean to belabor the point, but this is not the first year Bitfinex has been around. These aren’t even great interest rates to point out over its history. There were times where it was over 400%, and yet that was over a year ago. I am pretty sure that your model would have had them fail within that year, if that was compensation for the risk they were undertaking. However, not one bitfinex lender has ever lost money, and it has been around for multiple years.

    Again, I only point this out to you so that you can adjust your model based on real world data.

  • Maximo Joshua Rossi

    Not sure why my main comment is not being shown, but it basically said that since more than one currency is offered on bitfinex, if truly the only risk that is being paid for was exchange risk, any asset on that exchange should have the same rate. They don’t. The 30 day rate on BTC is around .14%, or roughly 4X that of the 30 day treasury note. So, if BTC is .14% and USD is over 3%, how can both assets reflect the risk of the exchange defaulting? Per the author, if it defaults lenders lose all the money, not just one type of asset. It seems that the market is what is setting the rates, and not a 50 year old debunked theory of efficient markets.

    • It was pending approval because it contained a URL (anti-spam setting). 🙂 I’ve replied to that comment (now approved).

  • First I would like to note that comparing BTC to USD is not just different in terms of exposure, but also the BTC swap market on BFX is completely illiquid. The demand for BTC swaps is literally zero, and you only point out the best offer rate. The worst offer rate is 185 times higher (1%). The USD market is a lot more liquid, and the worst offer is no more than 1.07 times the best offer.

    We can argue market efficiency, but let’s not that due that based on liquid markets versus no market. Also, we don’t know where supply comes from. Anyone who tries to take advantage of excessive rates will also be on the supply side and cause pressure on the offered rates (until only a fair return remains). In order to make any serious statement on the market efficiency, we would need something like the day-on-day changes in credit spreads. Otherwise we will end up in making subjective statements. Why would Bitcoin markets be terrible inefficient when even Bitcoin hedge funds are entering the market (GABI), and there are no restrictions on money?

    You might have a good point, but it requires more data to be examined. We cannot just randomly assume extreme market inefficiency. Perhaps we could use this: http://www.bfxdata.com/swaphistory/usd.php but is there some downloadable price history?

    • Maximo Joshua Rossi

      That is exactly my point, you say that it is illiquid. That is because there is no demand for people to short. You can see a very full supply of offers to lend, but there are no takers. Simply put,

      Platform risk + demand = rate of return

      I point this out because that is data you can use to test your hypothesis without even leaving Bitfinex. I hadn’t even mentioned LTC as yet another example of a differing rate.

      I mean, any analysis should at least attempt to test itself. For example, not sure if you are aware of the future’s market, ICBit. Why don’t you take a look at the difference in the monthly future contract vs the rate over 30 days? They match up pretty closely. It seems that most people feel that bitcoin will be 3-4% higher in 30 days (I am not saying they are correct, rather that this seems to be a marketwide trend).

      tl;dr: there is no way that liquidity providers need to be compensated for risk in one area, but not in another. You only need look at the offer side of the book to see what rate of return, stripped of demand, people actually require.

      Also, I am still curious about your response to the fact that if we backtest this hypothesis it becomes even more easily disproven. The rates were 10X as high in the past, and yet this did not correspond with any rate of failure, as no lender has ever lost money.

      I personally think that any cursory review of the actual data available should show that this hypothesis is very weak.

      • As I tried to point out, you are neglecting pressures on the supply side. An argument such as demand is high so prices must be high will at best only apply for a short period of time. If you look at the graph, you can actually observe this easily. You can see some peaks in volume where rates also rise quickly, but as soon as they stabilize rates revert back. That’s exactly what you would expect, because there is no reason to assume credit risk would suddenly change. The additional demand results in an excess return for the lenders, but when demand stabilizes lenders will keep taking advantage of this until the excess is gone. At that point, only the risk compensation remains.

        Also, there is no data to do any backtest. To get any indicative result for that, we should be comparing 10 indentical exchanges over a full year. According to this model, about 4 would have defaulted. The scientific minimum would be 50 observations…

    • Maximo Joshua Rossi

      I don’t think we can randomly assume that there is only one factor at play in setting rates, and that it can predict a failure of a business which has never, even in times of much more volatility, failed to protect its liquidity providers.