Is this real? Fake Volumes on Hitbtc and Binance
TL;DR
This analysis reveals anomalies in distribution of trade frequencies on major exchanges — Hitbtc and Binance. Such anomalies may indicate that either the trading participants or the exchanges themselves are involved in wash trading (i.e., producing “fake volumes”).
Preface
The community is actively analyzing public trading data and tries to find shady practices and market manipulation on cryptocurrency exchanges. Fake volumes is the frequent topic of such analysis. Many suspect that reported and even adjusted volumes published by CoinMarketCap is far from being real.
The following articles are worth reading to become familiar with the topic:
- The analysis based on trade volume & order book liquidity comparison
https://medium.com/@sylvainartplayribes/chasing-fake-volume-a-crypto-plague-ea1a3c1e0b5e - Comparison of visitors tracked by SimilarWeb and reported trading volume
https://www.blockchaintransparency.org/exchangerankings - The analysis based on trade volume time series
https://blog.cryptoexchangeranks.com/industry-news/fake-volume-investigation-wash-trade-vol-2/
Analysis
Disclaimer: It’s not a comprehensive analysis across all the exchanges and all the available assets. It’s rather more like the hands-on analysis that could be reproduced on any PC with the help of Jupiter notebooks and some data analysis skills. We plan this article will be followed by further research papers.
Most exchanges don’t expose historical API or don’t provide millisecond timestamps that are necessary for this kind of analysis. Thus, we have decided to start with Binance and Hitbtc because they provide fast and supposedly reliable APIs.
The process:
- Extract 10 days of trades for the specific symbol
- Calculate a millisecond time interval between consequent trades
- Exclude outliers
- Plot a distribution (histogram)
Let’s start with BTC/USD symbol on Hitbtc as an example.
The time between trades tends to be concentrated near zero. It’s natural because the single order can cause several trades executed at the exact same moment.
For graphical analysis values near zero should be filtered out because they dominate the histogram.
After filtering out small values the distribution becomes close to log-normal. Everything seems organic and there are no visible anomalies.
Let’s continue will another high-volume symbol from Hitbtc — BCH/USDT
The same plot reveals some irregularities — the distribution has 2x density plateau between 5 and 15 sec that doesn’t match log-norm function at all.
But let’s take a look at much more suspicious examples from Hitbtc — LSK/BTC and XRP/USDT
On both plots there are extreme outliers near the round values:
- 3, 6, 7, 8, 9, 10, 11 seconds on XRP/USD
- 6, 7, 8, 9, 10, … seconds on LSK/BTC
It’s highly unlikely that the process of getting a trade exactly N seconds after the previous one is random. We would be surprised to learn that individual users are accurate enough to place a new order exactly N*1000 milliseconds after the previous one.
The credible explanation might be that sophisticated trading algorithms are configured this way and running there. Their aim is to produce inflated volumes in these symbols. It may be done via so-called self-trades, which are by the way allowed on Hitbtc, or as a concurrent trading activity from several controlled accounts.
Let’s turn to Binance and take a look at XRP/BTC and LSK/BTC there.
XRP/BTC distribution is an ideal example of log-normal distribution. May the reality be so ideal, God knows. In contrast, LSK/BTC has suspicious peaks at 1 second and 3 seconds. Although the peaks are much lower than on Hitbtc, they are still visible.
BCH/BTC symbol on Binance has also suspicious peak at ~1 second where the number of trades is much greater than at other frequencies.
Conclusion
On some high-volume symbols on major exchanges Hitbtc and Binance there are suspicious anomalies in the trade frequency distribution. The hypothesis is that trading algorithms are tuned in such a way that trade executes every N seconds ± random number. It may be an indication of wash trading on some symbols.
Given the fact that we have managed to detect anomalies in the trade distribution of LSK symbols on both exchanges, we draw a conclusion that the concerned parties such as major Lisk stakeholders might be the originators of inflated volumes. The rationale is simple — more volume indicates more interest in the coin and may attract more investors, which in turn results in the increasing demand and price.
On the contrary, anomalies in trade distribution of XRP and BCH have been found only on Hitbtc. It may support the hypothesis that the exchange itself is supposedly involved in wash trading. Inflated volumes help the exchange to be among the top exchanges by volume.
The article was written in October 2018. Its aim is bringing transparency and efficiency to cryptomarkets and cryptoeconomy.
