A Comparison of Perp Funding Rates
UXD Protocol’s underlying stability mechanism for UXD Stablecoin is intimately tied to the concept of funding rates in perpetual futures markets. In order to further its understanding of funding rates on different platforms, UXD Protocol has completed an internal study of funding rates across 5 platforms: Binance, BitMEX, FTX, Mango, and dYdX.
Here is the full report. For the summary, scroll down below.
Remind me, what are funding rates?
As discussed previously in our educational material on perpetual futures, funding rates are the mechanism by which perpetual protocol tether the “real” price of an asset like BTC to the “virtual” price of BTC on said protocol.
For example, if the “real” price of BTC is $100k, and the “virtual” price on the protocol is $90k, then there needs to be some way to incentivize buy orders on the protocol until the price moves closer to $100k.
Without some way to converge prices, traders are unfairly penalized or rewarded. For example, suppose you and I enter into the short and long sides of a perp, respectively, for BTC at an initial price of $100k. If BTC’s “real” market price (i.e. the one Chainlink’s/Pyth’s Oracle would report) remains $100k but the “virtual” price of our perp goes to $90k, then I will show a PnL (Profit net Loss) of $10k, and you will see a PnL of -$10k. A consequence of this is that I will want to exit this position immediately, and you will not, at least until the “virtual” price returns to $100k.
Certainly this feels intuitively unfair- why should we have PnL generated if the “real” price of BTC has not changed? After all, we wanted to make our trade based on how much the asset BTC changes in price, not based on our single perp market. (for some nice examples of this logic, see Paradigm’s Perp Explainer)
Therefore, perpetual protocols need a mechanism to bring the prices back in line. They do this in the form of a “funding rate”, a periodic payment between those who are benefitting from the imbalance between “real” and “virtual” prices (in the above example, me), and those who are being hurt by the imbalance. Generally, the size of this payment is a function of the difference between the “real” and “virtual” prices (often referred to as “mark” and “index” prices), with some additional nuances.
In the study conducted by UXD Protocol (pdf link at top and bottom), funding rates on Binance, BitMEX, FTX, Mango, and dYdX are classified according to characteristics, and volatility, distribution and correlations of funding rates are investigated. These results strengthen UXD’s understanding of funding rates on these markets, informing the design of UXD Stablecoin. This medium article presents some of the results as they relate to SOL. For a more complete analysis, as well as results for BTC, ETH and SOL, see the pdf report.
A Classification for Funding Rates
Exact funding rate calculations and parameters vary from exchange to exchange, and so it is useful to have a classification system for discussing funding rates.
In particular, UXD identifies the key dimensions below:
Namely, funding rates can vary in regards to (i) the frequency of the funding payment (ii) how real and virtual price inputs are taken into account (iii) how “interest rate differentials” between quote and payment currencies factor in (iv) whether or not the funding rates are bounded (v) how “virtual” price is calculated (impact of a trade of a certain size vs best pricing). Although not comprehensive, these are some of the main characteristics along which different funding rates vary.
As an example, let’s look at BitMEX. BitMEX (like FTX), has a funding rate calculation consisting of two components: (i) a Premium component tracking the difference between “Real” and “Virtual” prices, and (ii) and interest component. The Premium component is defined as:
Where “Impact Bid Price” and “Impact Ask Price” are defined as the average execution price for 10000 USD notional on their respective sides. Therefore the “virtual” price on the BitMEX market takes into account the depth of this market as well. The “fair basis” is a decaying term corresponding to the last funding rate value.
The interest component measures the difference in interest rates between quote currency (generally USD in a TOKEN-USD pair), and the base currency (generally TOKEN, in a TOKEN-USD pair):
In theory this is supposed to help capture the relative desirability of borrowing/lending the base/quote currencies, which is effectively what happens when the “real” and “virtual” prices of a perpetual market get out of whack, because there is a “loan” being made between the two parties. Again, see Paradigm’s Perp Explainer.
Finally, there is a clamping component, which acts to (i) set a max and min on the funding rate (ii) reduces the overall volatility of funding. Rarely are these clamps actually met, but funding rate calculations do not always include them. Our final funding rate, r , is then:
Although the purpose of this study was not to explain the fine mechanics of funding rate calculations, it’s quite useful to understand how they are calculated in practice. Note for example, that the I term in the funding rate calculation introduces a positive bias from the interest component, meaning all else equal BitMex rates may tend to be more positive than those from other exchanges.
Volatility, Distribution & Correlation
Moving to the results of the study, SOL-Perps are used as an example to demonstrate the variance in volatility between the different calculations. Note, for example, the multiple extreme swings from Mango and dYdX funding rates. However, funding rate volatility will be a function of liquidity depth in addition to the particular funding rate calculation used, so it is not always possible to compare from an immediate chart.
However, some characteristics do reveal themselves quite clearly when looking at a distribution of funding rates:
For example, it is clear that BitMEX’s funding rate calculation often biases it towards positive funding, whereas dYdX’s funding rates are nearly perfectly symmetric around zero.
It’s also instructive to understand how correlated the funding rates are between exchanges. On a weekly basis (cumulative funding) we can see that BitMEX seems to be almost entirely uncorrelated with the other markets. This is due to BitMEX settling PnL in BTC, which tends to affect the funding rate positively.
Funding Rates Over Time
Notice that on a smoother, weekly basis, differences in funding rates and biases in calculations become quite pronounced. For example, it is clear that the interest component in BitMEX’s calculation consistently biases rates upwards.
Finally, note the high variability of funding rates on decentralized exchanges relative to centralized ones, though this was likely due to low liquidity during these spikes.
Funding Rate Autocorrelations
In general, UXD finds that funding rates tend to be quite autocorrelated (meaning the funding rate in period n+1 is strongly correlated to the funding rate in period n), as well as mean reverting (generally towards zero). For example, for FTX BTC-Perp, there is significant autocorrelation even several days later (much of this is likely intentional, due to the nature of the FTX funding rate calculation).
Notably, funding rate returns on decentralized exchanges such as Mango seem to exhibit slightly more volatile and weaker autocorrelations when compared to FTX:
UXD completed this study to further inform its strategies related to mitigating the impacts of funding rates on the stability of UXD Stablecoin. The goal broadly was to understand (i) how to classify and understand different types of funding rates (ii) a brief analysis of the data related to these funding rate types to look for biases, etc. Overall, funding rates appear to have become more stable over time, although they still exhibit sharp behavior during times of market volatility. As of the time of writing, of the exchanges considered rates are tightest (least volatile) on FTX and Binance and widest (most volatile) on Mango Markets and dYdX. Rates on decentralized exchanges are more volatile than those on centralized exchanges and contain substantially more outliers. However, this may be due in part to the particulars of the decentralized exchanges chosen.
As usual, please reach out on discord if you have any comments/questions. Thanks for reading!