The Law of One Bitcoin Price?∗
Total Page:16
File Type:pdf, Size:1020Kb
The Law of One Bitcoin Price?∗ Alexander Kroeger† and Asani Sarkar‡ January 19, 2017 Abstract Bitcoin, a digital currency, constitutesa textbook example where the law of one price should be satisfied as, unlike asset pairs previously studied in the literature, it is a fully fungible asset with identical payoffs. Despite this, we show the existence of persistent, statistically signifi- cant differences between US dollar-denominated bitcoin prices in multiple bitcoin exchanges. Further, the absolute values of price differences are positively related to the bid-ask spread, or- der book depth and volatility and negatively related to volume. Price differences are also higher on exchanges with smaller trade sizes, consistent with clientele effects from greater institutional trading. Moreover, impulse responses indicate that shocks to illiquidityand volatilityhave more persistent effects on absolute price differences between exchange pairs with more retail trading and greater counterparty risk. Finally, the speed of arbitrage and the amount of price discovery is related to the arbitrage frictions. Thus, limits to arbitrage remain relevant even in the context of a homogeneous asset class where many of the frictions in more traditional asset markets are absent. Keywords: bitcoin exchanges, limits to arbitrage, digital currency JEL classification: G12, G14 ∗The views expressed in this article are those of the authors and do not necessarily reflect the views of the Federal Reserve Bank of New York or of the Federal Reserve System. Any errors or omissions are the responsibility of the authors. †Federal Reserve Bank of New York; e-mail: [email protected] ‡Federal Reserve Bank of New York; e-mail: [email protected] 1 Introduction The law of one price states that assets with identical payoffs should trade at the same price but, on occasion, this law appears to be violated. Examples of such violations are “Siamese-twin” stocks with almost-identical dividend streams that trade at different prices (Jong et al. (2009), Rosen- thal and Young (1990) and Debora and Froot (1999)), parent and subsidiary company stocks trad- ing at prices such that the parent company value is negative (Mitchell et al. (2002), Lamont and Thaler (2003)) and the off-the-run minus on-the-run Treasury bond spread(Amihud and Mendel- son (1991), Warga (1992) and Krishnamurthy (2009)). Nevertheless, this evidence exists for only a limited set of asset pairs since those with closely related payoffs are hard to come by (Gromb and Vayanos (2010)). Additional violations of the law of one price come from cross-listed stocks, such as American Depository Receipts (ADR), where the stock of the same company trading in geographically-dispersed exchanges trade at different prices (Gagnon and Karolyi (2010)). In all these cases, although the payoffs are close, they are not identical. For example, even ADRs and the foreign shares they represent are not fully fungible (Gagnon and Karolyi (2010)). In this paper, we study price differences of the virtual currency bitcoin trading on 6 exchanges constituting 15 exchange pairs where the price is quoted in US dollars (USD). Unlike asset pairs previously studied in the literature, bitcoin in different USD exchanges is by design the same, fully fungible, asset with identical payoffs and no currency risk. As such, these exchanges constitute ideal, textbook examples where the law of one price should be satisfied. Our results, however, show the existence of persistent, statistically significant differences between bitcoin prices quoted in multiple USD exchanges. These price differences constitute two groups. For exchange pairs involed BTC-e, bitcoin consistently trades at substantially lower prices on BTC-e compared to the other exchanges, and the average price difference is statistically significantly different from zero. For the remaining exchange pairs, price differences are typically more episodic and smaller in magnitude, and they switch signs. For example, higher transfer costs and lower liquidity are likely to result in larger trading fric- tions and thus higher price discrepancies. Consistent with this idea, we find that the absolute values of price differences are positively related to the bid-ask spread, order book depth and volatility and negatively related to volume. We also account for explicit fees, in particular bitcoin network trading fees that users specify in order to incentivize so-called “miners” who create bitcoin.1 However, even after accounting for the implicit and explicit trading costs, we find that the average absolute price difference remains statistically significant for exchange pairs involved BTC-e. 1See Section 2.1 for a more detailed description. 1 Since trading frictions do not fully account for price differences in the exchange pairs involving BTC-e, we next examine market segmentation and institutional factors as determinants of bitcoin arbitrage profits. Segmentation may arise from clientele effects with the more active exchanges hosting more institutional traders, as inferred from differential trade sizes. We define dummy vari- ables bothretail and oneretail that indicate whether both or one of the exchange pairs, respectively, have small trade sizes and find that absolute price differences are positively related to the incidence of retail trading.2 We account for institutional factors through exchange-pair fixed effects. These fixed effects are generally statistically significant, implying that institutional features matter. Other factors, such as anonymity and exchange failure risk are also important, although we cannot quan- tify them. In particular, BTC-e offers enhanced anonymity relative to other exchanges, making it potentially attractive to users wishing to convert illicitly-obtained bitcoin into fiat currency, thus creating increased selling pressure (and hence lower prices) compared to other exchanges. Traders may also percieve BTC-e as carrying a greater risk of loss of customer accounts due to the opacity of their operations, entailing a risk premium on an arbitrage trade. Do trading frictions account for the persistence of price differences? We estimate impulse re- sponses from a Vector Autoregression (VAR) of price differences and search frictions. We find that shocks to illiquidity and volatility have persistent effects on price differences between BTC-e re- lated exchange pairs, lasting 10 days or more. In contrast, the effect of shocks on price differences are more attenuated, lasting 2 to 3 days, for exchange pairs that do not involve BTC-e. Overall, both the magnitude and persistence of price differences are related to search frictions. How do arbitrage frictions affect price efficiency dynamics and price discovery? We estimate a Vector Error Correction Model (VECM) and find that prices of all exchange pairs are cointegrated of order one, indicating the existence of a long-run equilibrium relation between prices of every exchange pair. However, there are substantial differences between exchanges in the speed of ad- justment of prices to deviations from the long-run equilibrium relation. Specifically, the half-life of a price deviation (i.e. the number of days required to eliminate half of the price deviation) is two to three times higher for BTC-e, as compared to other exchanges. Moreover, the information share of BTC-e is small compared to other exchanges, indicating that little price discovery occurs on this exchange for the period analyzed. By contrast, a substantial part of price discovery occurs on Bitfinex, relative to other exchanges. These results imply that arbitrage frictions impede the speed with which prices revert to equilibrium and the amount of price discovery. 2 “Retail exchanges” are defined as having a 75th percentile of trade size that is below the 75th percentile of all trade sizes for the past 30 days. 2 We contribute to the literature by showing that search frictions, market segmentation and insti- tutional features all contribute to limiting arbitrage between bitcoin exchanges in the cross-section and in the time series. We demonstrate how frictions affect arbitrage profits across exchange pairs. Further, we relate the persistence of arbitrage profits to search frictions, and show how the dynam- ics also differ across exchange pairs. Finally, we relate arbitrage frictions to price discovery. Thus, limits to arbitrage remain relevant for price efficiency even in the context of a homogeneous asset class where many of the frictions in more traditional asset markets are absent. The article is organized as follows. Section 2 provides an overview of Bitcoin and the exchange market. Section 3 provides more information on the sources of data used to carry out the analysis. Section 4 documents the price relationships between the major USD–BTC exchanges and describes the various arbitrage frictions that could lead to price discrepancies. Section 5 presents regression results to quantify the effects of various frictions on price disparities. Section 6 analyzes some of the dynamic considerations for arbitrage. Section 7 concludes. 2 How bitcoin is owned, used, priced and regulated Bitcoin is a digital asset and payment system introduced in 2009 that has attracted a great deal of attention due to its innovative design. Over the last seven years, bitcoin has grown to become a major medium of exchange with an estimated 100,000 merchants worldwide accepting bitcoin.3 In this section, we describe the use of bitcoin as a decentralized payments system (section 2.1) and how bitcoins are traded over online exchanges (section 2.2). Regulatory issues related to exchanges are discussed in section 2.3. 2.1 The bitcoin payment system Ownership of the units of bitcoin is established through claims to the current state of the public transaction history known as the “blockchain.” Users of bitcoin are identified by one or more bitcoin “addresses.” These addresses are known to the public and are often managed by client software known as a “wallet.” In order to claim ownership of bitcoin, that is, to send bitcoin from a particular address, one must possess the “private key” associated with that address. As the name suggests, the private key is not known to the public.