Price Discovery on Bitcoin Exchanges (Digest Summary)

In the developing market of bitcoin, it is not surprising that price differences exist on various exchanges. But given that these differences have been temporarily as high as 143% and then 7%–14% for sustained periods, the authors explore the question of exactly where the value of bitcoin is decided. The larger exchanges obviously contribute a greater share of information to the price discovery, but the data suggest that some exchanges, including some of the smaller ones, have a greater influence than others.

Although still a minor market and perhaps even just a passing fad, crypto currencies are a point of interest to many economic thinkers. In a world of experimental central bank policies, interest in the decentralized, peer-to-peer currency recently drove the bitcoin market cap to more than $10 billion. The authors give a very brief history of bitcoin but focus on the role each exchange plays in the price discovery of the currency.

Market size, trading volume, spreads, and currency of trade are examples of variables that may lead to price differences observed on each exchange. In a price discovery model, the authors assess 11 months of trade data from seven bitcoin exchanges for their information contribution. Although all exchanges contribute information to the market, data from price-leading and price-following analysis reveal that Mt.Gox and BTC-e have the highest correlation with future market moves and, therefore, a larger “information share.”

Given that the bitcoin market is relatively immature, the authors also plot the data for each month and, not surprisingly, find that each exchange’s information share is dynamic and still evolving. Mt.Gox dominated in the beginning of the dataset but declined as it neared bankruptcy. Interestingly, BTC-e maintained a uniquely stable position, even with the shock from the shutdown of the Silk Road.

Because it is logical that a higher fraction of the price discovery can be attributed to exchanges with higher trading volume, the authors consider the ratio of information share to activity share (i.e., volume) to normalize this size difference. Although it still holds that smaller exchanges provide less information and follow the larger exchanges with a lag, the authors are able to identify a spike in BTC China’s information share as interest in bitcoin from the Chinese market increased.

The authors use free, publicly available, high-frequency trade data from seven exchanges that traded approximately 90% of the bitcoin volume over their sample period from 1 April 2013 to 25 February 2014. The raw tick data from five-minute intervals include timestamp, price, and volume for each exchange. Four of the exchanges trade in US dollars, whereas the remaining trade in Chinese renminbi, Polish zloty, and Canadian dollars.

The authors use a multivariate time-series model proposed by de Jong, Mahieu, Schotman, and van Leeuwen (Working paper 2001) to assess the contributions of a random walk (fundamental price) component and an idiosyncratic (noise) component. The covariance between the fundamental and noise component is an important parameter that indicates a price change is informative. The higher the value of this parameter, the higher the information share relative to its activity share (or volume). The authors also study the development of the ratio of the information share to the activity share to isolate informative changes on the smaller exchanges. Data and results are presented in tables and graphs.

This article is well written and on an interesting topic. Although the authors address the potential for (and various obstacles to) arbitrage opportunities across various exchanges, I believe the subject is dismissed too quickly given the important ramifications for price discovery.