Trading and Arbitrage in Cryptocurrency Markets

Transcription

1 Trading and Arbitrage in Cryptocurrency Markets Igor Makarov LSE Antoinette Schoar MIT Sloan Gerzensee, October 10, 2018

2 Motivation Cryptocurrencies like bitcoin are built on the blockchain technology Allows verification of payments in the absence of a centralized custodian Verification is done by decentralized "miners" Bitcoin was originally introduced in a paper by Nakamoto (2008) and came into existence in 2009 First payment transaction on May 22, 2010: A bitcoin enthusiast (Laszlo Hanyecz) bought 2 pizzas for 10,000 bitcoins Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 2

3 Motivation (cont.) Since then the market for cryptocurrencies has evolved dramatically More then 1000 altcoins traded on more than 100 exchanges Market cap: $500B at the peak Academic research on bitcoin trading is in its beginnings Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 3

4 This paper A systematic analysis of the trading and efficiency of crypto markets Crypto markets are ideal settings for studying price arbitrage: Markets are segmented across many countries and jurisdictions Many naive investors and few large sophisticated investors (e.g., DRW, Jump Trading, or Hehmeyer Trading) Blockchain technology alleviates some constraints (e.g., capital mobility) but introduces others (the transfer of value between exchanges is subject to a delay) Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 4

5 Main results History of bitcoin exchanges marked by recurring episodes of arbitrage opportunities opening and closing again The total size of arbitrage profits from December 2017 to February 2018 is well above $1 billion Arbitrage opportunities persist for several hours or even days and weeks Market segmentation matters Arbitrage opportunities are larger across countries (or regions) than within the same country Arbitrage spreads are much smaller for exchange rates between different cryptocurrencies compared to exchange rates between cryptocurrencies and fiat currencies Arbitrage spreads across countries show strong co-movement Countries with higher average bitcoin premium also respond more strongly to periods of buying pressure Price deviations are asymmetric: Bitcoin price in rest of world is above US Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 5

6 Main results (cont.) Bitcoin returns and arbitrage spreads vary with net order flows We decompose signed volume on each exchange into a common component and an idiosyncratic, exchange-specific component The common component explains 80 percent of the variation in bitcoin returns Buying 10,000 bitcoins raises returns by 4% at the daily frequency The idiosyncratic components of order flow play an important role in explaining the size of the arbitrage spreads between exchanges Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 6

7 Overview Bitcoin price: January 1st, 2016 February 28, Volume-weighted daily price (using trades on Bitstamp and Coinbase) Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 7

8 Overview Daily bitcoin volume to fiat currencies in 2017 Exchanges: US: Makarov Coinbase, and Schoar, Bitstamp, Trading and Gemini, ArbitrageKraken in Cryptocurrency Markets 8

9 Data Tick level trading data from Kaiko, a private firm that has been collecting trading information about crypto currencies since 2014 The Kaiko data cover the 15 largest and most liquid exchanges: Bitstamp, Kraken, BTCC, Bittrex, Coinbase, OkCoin, Bitfinex, Poloniex, Bithumb, Gemini, Quoine, bitflyer, Huobi, Binance, and Zaif The 15 exchanges account for 85% of total bitcoin volume to fiat curencies Expanded sample of 30 exchanges across 16 countries from additional sources such as bitcoincharts.com and individual exchanges themselves Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 9

10 Summary statistics: returns Return frequency Std. Dev Skewness Kurtosis ρ 1 ρ 2 ρ 3 cross correlation 5 - Minute Hour Daily Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 10

11 Arbitrage index (all exchanges). Arbitrage index rbitrage index is calculated at minute-level, and then averaged by day. xchanges: S: Coinbase, Makarov and Bitstamp, Schoar, Trading Gemini, and Arbitrage Kraken in Cryptocurrency Markets 11

12 5. Arbitrage index within the region Arbitrage US index within the region index (within regions) 5.2 Japan 5.1 US 5.4 Europe 5.2 Japan 5.2 Japan US 5.3 Korea 6. Price ratios between regions: second level Europe price ratios, and then averaged by day 6.1 US and Japan Korea Korea Japan Korea 6.2 US and Korea Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 12

13 Hong Kong: Bitfinex Europe: Kraken, Coinbase, Bitstamp Arbitrage 5.4 Europe index (between regions) 4. Rank of regional prices (Minute-level data) 6.2 US and Korea 6. Price ratios between regions: second level price ratios, and then averaged by day 6.1 US and Japan All 6.3 US and Europe US vs. Korea 6.2 US and Korea US vs. Japan 7. Arbitrage profit across regions US vs. Europe - Arbitrage profit at second level, then aggregate by day; Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets - Price difference > 2%; 13

14 Arbitrage profit (between regions) 7.2 US and Korea Price difference taking into account bid-ask spread 7.2 US and Korea 7.3 US and Europe 7.2 US and Korea Japan: total profit $116M 7.3 US and Europe Korea: total profit $747M 7.3 US and Europe Europe: total profit $23M Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 14

15 Arbitrage index: ethereum and ripple 8. Arbitrage index: ETH Exchanges US: Coinbase, Bitstamp, Gemini, Kraken Japan: Bitflyer (ETHBTC BTCJPY), Quoine Korea: Bithumb Hong Kong: Bitfinex Europe: Kraken, Coinbase, Bitstamp 8.2 XRP Fiat currency Exchanges ethereum US: Coinbase, Bitstamp, Gemini, Kraken Japan: Bitflyer (ETHBTC BTCJPY), Quoine Korea: Bithumb Hong Kong: Bitfinex Europe: Kraken, Coinbase, Bitstamp 9. Price ratios between regions: price of BTC in number of ETH Volume-weighted price calculated at 1-minute level (then averaged by day, for the price ratios) Exchanges ripple US: Bitstamp, Kraken Korea: Bithumb Makarov and Schoar, Trading Hong Kong: andbitfinex Arbitrage in Cryptocurrency Markets 15

16 Ethereum-bitcoin exchange rate across regions US US and and Japan Korea 9.2 US and Korea 9.3 US and Europe 9.2 US and Korea Japan 9.3 US and Europe Korea * In the previous version, I left out Coinbase: Europe EUR by mistake. 9.3 US and Europe * In the previous version, I left out Coinbase: EUR by mistake. Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 16

17 Extension in Progress Expanding sample to 30 exchanges in 16 countries allows us to look at the correlation structure between price deviatios across countries Data from Bitcoinchart.com, Kaiko and several exchanges directly Calculate arbitrage spread relative to the US for each country Arbitrage spreads co-move and are asymmetric relative to the US: Rest of the world trades at a premium to the US (and Europe) Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 17

18 Co-Movement of Arbitrage Spreads Correlation Matrix: Bitcoin Arbitrage Index Across Regions Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 18

19 Extension in Progress I Use standard Hodrick-Prescott Filter to calculate the smoothed Bitcoin price at the weekly level in the US Calculate deviations of the actual log price from the smoothed log price to provide metric of "buying pressure" in the US Regress arbitrage spreads of individual countries relative to US price on our measure of buying pressure We find a strong positive beta: Countries outside the US repond strongly to price pressure in the US Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 19

20 US Price Deviations from Trend Difference between US bitcoin price and smoothed btcoin price series using Prescott filter Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 20

21 Extension in Progress II Correlate a countriy s average bitcoin premium realtive to the US Bitcoin price with the bitcoin beta of the country Countries that have a higher average bitcoin premium over the US, also show larger arbitrage deviations in times when the buying pressure in the US goes up. Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 21

22 Sensitivity to Buying Pressure in the US Regression of Regional Bitcoin Premia (to the US) on our measure of Buying Pressure Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 22

23 The Role of Capital Controls Regression of pairwise correlation between arbitrage spreads on pairwise measure of capital control: PCont ij = 1 γ i γ j (1) Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 23

24 How to Interpret the Findings? The marginal investor outside the US is willing to pay more for bitcoin in response to positive news. Possibly because the value of cryptocurrencies is higher in countries with poor financial markets Good news about crypto currencies or changes in sentiment affects buying pressure in the US but even more so in other countries To observe sustained price deviations markets must be segemented and arbitrage capital flows slowly Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 24

25 Implementation of Arbitrage In a frictionless world if prices are different across exchanges there is a riskless arbitrage: Exch 1: P 1 = 100 Exch 2: P 2 = 200 B1 B1 $100 $200 Transactions take time need to buy and sell bitcoin simulteneously Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 25

26 Implementation of Arbitrage II Ideally, an arbitrageur would like to short sell bitcoin on the market where the price is high often not feasible, because many exchanges do not allow short-sales Two solutions: Trading on margin similar to short-sales, but does not allow for physical settlement convergence risk Hold a positive balance of bitcoin on both exchanges and simultaneously buy and sell bitcoins across the two exchanges whenever the price on one exchange deviates from that on the other price risk To mitigate the price risk the arbitrageur can Short-sale bitcoins Borrow bitcoin from people who hold big amounts of bitcoin without an interest to sell (hodlers) Use futures contracts (from December 2017) Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 26

27 Frictions I: Transaction costs Buying and selling bitcoins on an exchange: bid-ask spread (1-10bp), exchange fees (0-10bp) Sending bitcoins across exchanges via bitcoin protocol (very small for large transactions) Exchange deposit/withdrawal fees (vary, small for large transactions) For large players the round-up trading costs should be within 50 to 75 bp very low compared to the arbitrage spreads Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 27

28 Frictions II: Exchange Governance Risk To trade on an exchange the arbitrageur has to give up control of her coins to the exchange if the exchange is hacked (and many were) the arbitrageur can loose her funds Not a compelling explanation: Arbitrage spreads are much larger across than within regions for exchange risk to explain this pattern the exchange risk must be region specific Concerns about the governance risk of an exchange should affect its volume and possibly bid-ask spreads There is significant heterogeneity in the liquidity of exchanges within a region but nevertheless arbitrage spreads are small between them Arbitrage spreads have common component Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 28

29 Frictions III: Capital controls The arbitrageur has to able to trade across multiple exchanges and transfer capital between them Many retail investors face restrictions on which exchanges they can trade. Not binding for large institutions Capital controls for fiat currencies (e.g. Korea, binding for retail investors, for large financial institutions - unclear) Arbitrage is much smaller for crypto-currency pairs sign that capital controls contribute to the limits of arbitrage In the presence of capital controls the arbitrageur can still bet on the price convergence across the two regions. But capital controls reduce the efficiency of arbitrage capital Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 29

30 Conclusion Document persistence of large arbitrage spreads in the price of cryptocurrencies to fiat currencies across exchanges Linked to capital controls across regions Not driven by transaction costs or differential governance risk across exchanges Effects are much smaller for exchange rates between cryptocurrencies Results point to the importance of limits to arbitrage At times the arbitrage capital seems to get overwhelmed by noise traders who drive up the price in certain markets or lose heart when negative information about bitcoin comes out One interpretation of our results is that this is a particular form of slow moving capital (Duffie (2010) AFA presidential address) Arbitrage spreads are correlated across regions and time Countries with tighter capital controls and worse financial markets show higher arbitrage spreads Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 30

31 Thank You! Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 31

32 Appendix: Net order flow and prices There is a strong positive relationship between net order flows and prices in traditional financial markets Currency markets: Evans and Lyons (2002) Bond markets: Brandt and Kavajecz (2004) S&P 500 futures market: Deuskar and Johnson (2011) US stock market: Hendershott and Menkveld (2014) Usually attributed to price discovery. It is less clear what the fundamentals are in the case of cryptocurrency markets and whether there are any traders who have more information than others Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 32

33 Net order flow and prices (cont.) A common way to estimate the impact of net order flow is to regress returns on the signed volume The complication in the bitcoin market is that the same asset is traded simultaneously on multiple exchanges and often at different prices Therefore, when forming their demand investors might not only look at prices on their own exchange but also take into account prices on the other exchanges where bitcoin is traded Hence, a regression of returns on signed volume in each market separately may give a biased picture of the true impact of net order flow Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 33

34 Model: signed volume s it = s i + β s i s t + ŝ it, β s = 1 (2) i s it is signed volume on exchange i s t is the common component for all exchanges ŝ it is an exchange specific component E[s t ] = 0, E[ŝ it ] = 0 E[s t ŝit] = 0, E[ŝ it ŝ jt ] = 0, for i = j Linear model: s t = w s i s it, β s i ws = 1 i Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 34

35 Model: returns r it = r i + β r i s t + ˆr it (3) r it is log-return on exchange i r t is the common component for all exchanges ˆr it is an exchange specific log-return E[r t ] = 0, E[ˆr it ] = 0 E[r t ˆr it ] = 0, E[ˆr itˆr jt ] = 0, for i = j Linear model: r t = w r i r it, w r = 1 i Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 35

36 Estimation: signed volume Bitfinex Coinbase USD Bitstamp USD Gemini Kraken USD Kraken EUR Coinbase EUR Bitstamp EUR bitflyer Quoine Zaif Bithumb Poloniex Bittrex 5-min frequency β s i w s i R hourly frequency β s i w s i R daily frequency β s i w s i R Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 36

37 Estimation: returns Bitfinex Coinbase USD Bitstamp USD Gemini Kraken USD Kraken EUR Coinbase EUR Bitstamp EUR bitflyer Quoine Zaif Bithumb Poloniex Bittrex 5-min frequency β r i w i r R hourly frequency β r i w i r R daily frequency β r i w i r R Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 37

38 Sytematic price impact r t = λs t T + λτ s τ 1 + ϵ t τ=1 5-min frequency λ 10 4 (%) hourly frequency λ 10 4 (%) daily frequency λ 10 4 (%) s t s t 1 s t 2 s t 3 s t 4 s t (80.06) (86.19) (88.05) (35.12) (39.7) (40.41) (16.92) (19.93) (18.96) (-36.54) (-32.24) (-16.53) (-15.67) (-4.05) (-3.62) (-11.68) (-3.71) (-0.2) (-7.56) (-1.22) (-0.76) (-6.88) ( -3.00) (-1.71) (-5.24) (-1.33) (1.57) R Makarov and Schoar, Trading and Arbitrage in Cryptocurrency Markets 38

39 Exchange-specific price impact p it = p t + ˆp it, 3 ˆp it = λ i ŝ it + a i,s ˆp it s + ϵ it s=1 Bitfinex Coinbase USD Bitstamp USD Gemini Kraken USD Kraken EUR Coinbase EUR Bitstamp EUR bitflyer Quoine Zaif Bithumb Poloniex Bittrex 45 5-min frequency λi 10 4 (%) (16.49) (22.83) (9.18) (14.35) (21.14) (27.66) (25.64) (7.43) (22.26) (6.58) (13.34) (25.13) (12.28) (14.00) a1i (48.44) (16.28) (56.57) (34.58) (43.48) (40.07) (29.02) (25.25) (40.69) (26.36) (14.73) (50.95) (54.99) (61.34) a2i (17.07) (5.58) (21.47) (13.5) (14.75) (11.51) (4.18) (18.78) (4.8) (5.55) (0.08) (6.45) (19.32) (21.32) a3i (12.84) (5.51) (21.89) (11.18) (19.68) (9.62) (4.1) (13.54) (1.79) (2.59) (3.3) (3.64) (16.56) (18.78) R-square Table 8. This table reports the results from time-series regressions of the idiosyncratic component of the signed volume on each of the exchange list on the top of the column, regressed on the deviation of the price from the common price component and past three lags of the idiosyncratic component of Makarov the signedand volume Schoar, of the Trading same exchange and Arbitrage. The idiosyncratic in Cryptocurrency components, ŝitmarkets and ˆpit and are obtained as the residual values of signed volume and returns 39