Trading and Arbitrage in Cryptocurrency Markets (Summary)

Most cryptocurrency exchanges are non-integrated, independently owned, and country-specific operations. The order books are separate, and the exchanges lack mechanisms to ensure the best price. Further, the bitcoin settlements (registration on the bitcoin blockchain) take almost an hour, and payments can take from a few hours to several days. In most cases, investors are allowed to trade only on local exchanges, and these trades are settled through fiat currency (except for small but growing cross-cryptocurrency trades settled through tether). This dynamic, coupled with a lack of regulatory oversight, leads to governance risk, tight capital controls in many regions outside the United States and Europe, and inefficient market segmentation.

In this context, the authors analyze price formation and arbitrage opportunities in the evolving cryptocurrency markets. The findings, based on the transaction-level data, are also relevant for potential real effects of cryptocurrencies as payment and transaction mechanisms. In addition, this research adds to the study of the financial markets, especially arbitrage and price impacts of order flow.

The authors use tick-level trading data for bitcoin, Ethereum, and Ripple, the three largest and most liquid cryptocurrencies. The analysis focuses on the 15 most liquid exchanges but covers 34 exchanges in 19 locations. The raw data are primarily sourced from Kaiko and are supplemented with data from Bitcoincharts.com. January 2017 to February 2018, the period analyzed, is marked by a significant rise in prices as bitcoin prices grew to an all-time high of almost US$20,000 in December 2017 and, with some sharp corrections, finally ended near US$10,000. The period is also noted for high volatility with an annualized standard deviation of 107% and significant volume growth: The average daily trading volume of bitcoin in US dollars increased tenfold on the largest exchanges during the second sub-period of the study.

Price deviations are evaluated through an “arbitrage index” that is defined as the daily average of minutewise max/min of the volume-weighted average price (VWAP) across exchanges and that should be close to 1 for an arbitrage-free market. The actual arbitrage spreads are as high as 1.5 times that average. Even within a geographical region, the arbitrage indexes, although smaller, are still quite high compared with traditional markets. The plot of the “price ratio” for countries vis-à-vis the United States confirms that the persistent price deviations drive high overall arbitrage spreads.

To capture the quantum of “arbitrage profit,” the authors take the minimum of low-priced sell-initiated volume and high-priced buy-initiated volume for every second, and they aggregate for days. The correlation in these price deviations across geographies is done through a minute-level ratio of the VWAP price in each of the markets to the VWAP price in the United States, a proxy for world bitcoin prices. The Hodrick–Prescott filter is used to analyze the correlation. Residuals between the actual and the smoothed log price are used as a measure for bitcoin “buying pressure” in the United States. The regression of the price deviation with buying pressure gives an “arbitrage beta,” indicating the sensitivity of prices in a country to the buying pressure.

The capital requirement to close the arbitrage spreads is assessed using the impact of net order flow by regressing price differences or returns over a particular period on the signed volume of trades, divided into an exchange-specific component and a common component.

The authors demonstrate large and recurring arbitrage spreads in coin prices across exchanges. These price deviations can persist from several hours to days and weeks, and they are normally larger across countries than within the same country (even in the regions with highly liquid exchanges). For example, the daily average price ratio between the United States and South Korea for the period of the study was less than 15% and reached 40% for several days (the “kimchi premium”). Similarly, the ratio was approximately 10% and 3% for Japan and Europe, respectively. Between November 2017 and February 2018, the total “arbitrage profit” opportunity between the United States, South Korea, Japan, and Europe was, respectively, US$1.275 billion, US$675 million, and US$25 million.

The prices in other countries are typically higher compared with those in the United States, and these arbitrage spreads co-move at the same time across regions. Indonesia, Australia, Singapore, Japan, and South Korea show a correlation in arbitrage spreads of more than 75%, and almost half of the areas show a correlation of more than 50%.

Buying pressure on bitcoin often leads to opening-up of the arbitrage spreads. The regions with higher price deviations also see greater increase in arbitrage spreads. This result can be attributed to tighter capital controls or weaker financial institutions in locations outside the United States and Europe. This explanation is further supported by much smaller arbitrage spreads between cryptocurrencies relative to fiat currencies on these exchanges.

The arbitrage spreads cannot be explained by transaction costs, which are relatively low. In addition, the exchanges studied in each market have high liquidity. Ex ante safety concerns resulting from governance risk, however, may be keeping some arbitrageurs out of these trades and exchanges. Innovation and market development over time may reduce some of these constraints.

The price pressure of net orders is mostly permanent at the daily level, with approximately a 4% return for buying 10,000 bitcoins. The component of signed volume of the net order flow that is common across exchanges, however, explains around 50% of the returns at the five-minute and hourly level and almost 85% at daily level. The rest is because of exchange-specific idiosyncratic factors that are not arbitraged away immediately but do predict subsequent relative returns. These results explain the arbitrage spreads persistent over days and weeks, as well as the eventual closing of these opportunities.