BitMex Funding Rate Moving The Bitcoin Exchange Rate
Quick introduction
There is a vast number of academic papers dedicated to the day-of-week, month-of-year and other similar effects (anomalies). In general, the anomaly is proven to exist when there is an excess return or significantly higher return for the given day in comparison to the rest of the days in the week. However, there are also publications concerned about the validity of such effects and according to some, those positive findings are caused, to a certain extent, by data snooping biases which was shown by Sullivan, Timmermann and White (2001).
Let’s focus on cryptocurrencies now. As an example, there was a research published by Caporale and Plastun (2018) who has found Monday effect on BTCUSD. More specifically, there were tendencies for BTCUSD pair to grow on this day. This can be of a use as an inspiration in future studies. Whether this holds until today is questionable, nevertheless this is not a purpose of this article.
This article and its research aims at what influence can the BitMex funding tate have on XBTUSD and ETHUSD perpetual contracts at which this rate is paid/charged.
The rationale behind
BitMex pays/charges futures trading accounts with funding rates based on the current demand/supply of XBTUSD and ETHUSD contracts. For each contract independent funding rates are set. When the funding rate is positive, the buyers (clients in long position) are charged this funding rate and sellers (clients in short position) gets paid this rate and vice versa for negative rate. This funding takes place every 8 hour period.
The initial hypothesis to test was: Traders who are long and the funding rate is positive, will rather close their position to avoid the additional payments. This should raise the contract supply and the price is assumed to drop from here.
Research question
Does an 8 hour effect (anomaly) exists on XBTUSD and ETHUSD contracts?
Used tests and hypothesis definition
Two approaches to test above stated research question were chosen. These are also used by researchers and can be found to be used in Zhang et al. (2016), Caporale (2018) and many many others.
- Regression model
where,
R_t . . . is return at time t
D_it . . . is dummy variable which mark certain time period
By rejecting:
we conclude there is an anomaly on the given period.
2. Welch’s t-test
is conducted with returns on the t-th time period against all returns of the periods ≠ t, e.g. returns in the first 15 minutes against returns of the rest of 15 minutes periods which fit into the 8 hour funding rate period.
Data and workflow description
Author’s note: This research was conducted some time ago, thus accept our apologies that it’s not examined on the most recent data, we also kept test data separated. However, if anyone wanted the Jupyter notebook with code to replicate this with the most recent data, feel free to pm me.
For XBTUSD data since 2018–01–01 until 2018–09–01 (720 fundings)
For ETHUSD data since 2018–08–28 (first quotation) until 2019–03–20 (688 fundings)
There is a problem with the sign of the funding rate, therefore the data were split into two groups, one group is a time series where only positive funding dates occurred and the other one is a time series where only negative funding rates occurred. Thus, the testing runs for negative and positive effects separately.
From other point of view this anomaly existence is also dependent on the timeframe we use (candle size), simply because those assumed selling offs can occur an hour before the funding but also just 5 minutes or a few seconds before the Funding Rate is being paid/charged, from this reason all of the hypotheses were tested on 5 minute and 15 minute timeframe.
Results
Initial research question was confirmed and there was an anomaly found. However, this anomaly does not happen before the funding rate is paid/charged. More oppositely, it happens right after the new funding rate is announced.
a) Testing 5 minute time frame
b) Testing 15 minute time frame
XBTUSD
The anomaly exists in the first 5 minute of the every 8 hour period, no matter the sign of the Funding Rate. As a practical result for traders, this can be a quick rise/drop right after the funding rate is announced and thus impossible or difficult to trade. The justification why this might be a quick rise/drop in price is supported by the fact that we haven’t confirmed this anomaly on 15 minute data where this effect is weak with p-value ~ 0.048 (for dummy regressor) and ~ 0.051 (for Welch’s t-test).
ETHUSD
On ETHUSD an anomaly exists but after 100 minutes after every 8 hour period. Which does not make any fundamental sense. However, as noted in table above, this anomaly holds on 15 minutes candles as well.
Conclusion
It was found that anomaly on XBTUSD for the time of first 5 minute candle after funding rate announcement does exist.
Furthermore, from yet unknown reason there is strong return anomaly at approx 100 Minutes after rate announcement on ETHUSD, if anyone could come up with any reasonable justification or fundamental reasoning, please let us know in the chat below.
Last word on the sustainability of this effect
It might be the size of the funding rate, as the crypto market becomes less volatile and more stable, this could result in the lower the funding rate, because there would be no need by BitMex to equalize demand/supply.
Following charts tells the story.
