Introduction Theory Equilibrium Data and Methodology Results conclusion. Toxic Arbitrage. Wing Wah Tham. Erasmus University Rotterdam

Transcription

1 Toxic Arbitrage Thierry Foucault Roman Kozhan HEC University of Warwick Wing Wah Tham Erasmus University Rotterdam National Bank of Belgium May 27-28, 2015

2 Arbitrage ˆ Arbitrage is a cornerstone of finance...

3 Arbitrage ˆ Arbitrage is a cornerstone of finance... ˆ To make a parrot into a learned financial economist, he only needs to learn the single word: arbitrage (Ross (1987, American Economic Review)

4 Arbitrage ˆ Arbitrage is a cornerstone of finance... ˆ To make a parrot into a learned financial economist, he only needs to learn the single word: arbitrage (Ross (1987, American Economic Review) ˆ Is arbitrage a good thing? What is the social value of arbitrage?

5 High frequency arbitrageurs: Heroes or Villains? ˆ SEC (2010): U.S. concept release on equity market structure. The Commission requests comment on arbitrage strategies and whether they benefit or harm the interests of long-term investors and market quality in general.[...] (Securities Exchange Commission, 2010)

6 High frequency arbitrageurs: Heroes or Villains? ˆ SEC (2010): U.S. concept release on equity market structure. The Commission requests comment on arbitrage strategies and whether they benefit or harm the interests of long-term investors and market quality in general.[...] (Securities Exchange Commission, 2010) ˆ High frequency arbitrageurs: those taking advantage of very short lived arbitrage opportunities, lasting fractions of seconds

7 High frequency arbitrageurs: Heroes or Villains? ˆ SEC (2010): U.S. concept release on equity market structure. The Commission requests comment on arbitrage strategies and whether they benefit or harm the interests of long-term investors and market quality in general.[...] (Securities Exchange Commission, 2010) ˆ High frequency arbitrageurs: those taking advantage of very short lived arbitrage opportunities, lasting fractions of seconds ˆ Is high frequency arbitrage a good thing?

8 For high frequency arbitrageurs...yes.. ˆ Budish et al.(2015, fthcoming QJE): estimate annual profits from cross-arbitrage in the ES-SPY ETFs to be $75 million. ˆ This is just one of hundredth of high frequency cross market arbitrage opportunities (see Table B.2 in Budish et al.(2015)).

9 For high frequency arbitrageurs...yes.. ˆ Budish et al.(2015, fthcoming QJE): estimate annual profits from cross-arbitrage in the ES-SPY ETFs to be $75 million. ˆ This is just one of hundredth of high frequency cross market arbitrage opportunities (see Table B.2 in Budish et al.(2015)). For others?

10 Social Value of Arbitrage? ˆ Arbitrage has benefits: 1. Arbitrageurs increase pricing efficiency: they quickly correct mispricings due to noise/liquidity traders. 2. Arbitrageurs provides liquidity (literature on limits to arbitrage). In correcting mispricing, they provide liquidity to noise/liquidity traders = Relaxing constraints should be desirable because arbitrageurs provide liquidity (Gromb and Vayanos (2012))

11 Social Value of Arbitrage? ˆ Arbitrage has benefits: 1. Arbitrageurs increase pricing efficiency: they quickly correct mispricings due to noise/liquidity traders. 2. Arbitrageurs provides liquidity (literature on limits to arbitrage). In correcting mispricing, they provide liquidity to noise/liquidity traders = Relaxing constraints should be desirable because arbitrageurs provide liquidity (Gromb and Vayanos (2012)) ˆ Our paper: Arbitrage has also costs: Some arbitrage opportunities (not all) raise adverse selection costs they make markets less liquid. ˆ Why?

12 Arbitrage 1: Stale Quotes.

13 Arbitrage 2: Transient Price Pressures.

14 Toxic Arbitrage Opportunities ˆ Arbitrage opportunities due to stale quotes are a source of adverse selection (picking off risk; cf Copeland and Galai (1983)) for liquidity suppliers They are toxic.

15 Toxic Arbitrage Opportunities ˆ Arbitrage opportunities due to stale quotes are a source of adverse selection (picking off risk; cf Copeland and Galai (1983)) for liquidity suppliers They are toxic. ˆ They make the market less liquid. ˆ They do not generate gains from trade: the arbitrageurs gains are his/her counterparties losses.

16 Predictions ˆ Composition effect: This is not the number of arbitrage opportunities that matters but the nature of these opportunities. Illiquidity is higher 1. On days in which toxic arbitrage opportunities are more frequent; 2. In pairs of related assets (ETFs/Underlying basket) in which toxic opportunities are more frequent.

17 Predictions ˆ Composition effect: This is not the number of arbitrage opportunities that matters but the nature of these opportunities. Illiquidity is higher 1. On days in which toxic arbitrage opportunities are more frequent; 2. In pairs of related assets (ETFs/Underlying basket) in which toxic opportunities are more frequent.

18 Predictions ˆ Composition effect: This is not the number of arbitrage opportunities that matters but the nature of these opportunities. Illiquidity is higher 1. On days in which toxic arbitrage opportunities are more frequent; 2. In pairs of related assets (ETFs/Underlying basket) in which toxic opportunities are more frequent. ˆ Speed effect : Illiquidity is higher when arbitrageurs become relatively faster in reacting to toxic arbitrage opportunities.

19 Contribution to the Policy Debate ˆ Budish, Cramton, and Shim (2014) s batch auctions proposal (see CFTC, Comments Letter): market correlations completely break down at high-frequency [...], which creates technical arbitrage opportunities available to whomever is fastest [...] the arms race [...] is socially wasteful, and it harms liquidity. The arms race and these negative effects are thus a consequence of flawed market design. As an alternative, we recommend frequent batch auctions [...] at frequent but discrete time intervals, such as once per second.

20 Contribution to the Policy Debate ˆ Budish, Cramton, and Shim (2014) s batch auctions proposal (see CFTC, Comments Letter): market correlations completely break down at high-frequency [...], which creates technical arbitrage opportunities available to whomever is fastest [...] the arms race [...] is socially wasteful, and it harms liquidity. The arms race and these negative effects are thus a consequence of flawed market design. As an alternative, we recommend frequent batch auctions [...] at frequent but discrete time intervals, such as once per second. ˆ Randomized execution of market orders: 10 milliseconds delay would encourage the liquidity providers to do so because [...]they will have time to adjust their quotes following sudden events (in Interactive Brokers Group Proposal to address High Frequency Trading ),

21 Contribution to the Policy Debate ˆ However, no empirical analysis of the effect of high speed arbitrage on liquidity : 1. We are not sure the problem exists: Our analysis shows that it does but suggests that it could be time and pair specific because the composition of arbitrage opportunities varies over time and across pairs. 2. We have no idea of the counterfactual: By how much would liquidity increase if (a) one could suppress all toxic arbitrage opportunities or (b) one could reduce the likelihood that an arbitrageur is able to exploit a toxic arbitrage opportunity by x%?

22 Theory ˆ Similar to Foucault, Röell and Sandas (2003, RFS) with two assets X and Y. ˆ Payoffs θ X = σ θ Y at t=2. (1 share of X = σ shares of Y ) ˆ 3 types of participants 1. Two risk neutral market makers: One specialized in asset X and one specialized in asset Y. They set bid-ask quotes in each asset. 2. One risk neutral arbitrageur 3. Liquidity traders who buy or sell asset X or Y with equal probabilities.

23 Market Makers Quotes ˆ Market makers: at t = 1, they simultaneously choose their bid-ask spreads: S X and S Y and post quotes for the asset: Ask Price in asset j : a j = m j + S j 2 Bid Price in asset j : b j = m j S j 2 ˆ m j = Midquote for asset j = market maker j s valuation for the asset. ˆ m X = E (θ X ) = Unconditional expected payoff of the asset.

24 Market-Maker Y s Valuation

25 Case 1: Stale Quotes.

26 Case 2: Transient Price Pressures.

27 Arbitrageur s Payoff ˆ The arbitrage portfolio is: (a) Short σ shares of Y -Long one share of X or (b) Long σ shares of Y -Short one share of X. 1. Cost of (a): a X σb Y = m X σm Y + (S X + σs Y )/2 = σjump + (S X + σs Y )/2 ˆ 2. Cost of (b): σa Y b X = σm Y m X + S X /2 + σs Y /2 = +σjump + (S X + σs Y )/2 If σjump > (S X + σs Y )/2, the arbitrageur s optimal arbitrage strategy is: Portfolio (a) if Jump > 0 and Portfolio (b) If Jump < 0. ˆ Here Jump = 1/2

28 Traders payoffs Termination Liq. trader arrives (with prob 1 α) Traders Payoffs A toxic arbitrage happens (with prob αϕ) A non toxic arbitrage happens (with prob α(1 ϕ)) Liq. trader trades Arb. trades (with prob π) X cancels (with prob 1 π) Arb. trades (with prob 1) X cancels (with prob 0) Arb s expected payoff 0 X s expected payoff Y s expected payoff Aggregate expected payoff (Arbs + Market Makers) σ S X σs Y 2 0 S X4 (σ S X 2 ) 0 σs Y 4 σs Y 2 0 σ S X σs Y 2 0 S X2 0 σs Y 2 0 S X +σs Y σ 2 0

29 The Arbitrage Race ˆ Traders choose their average speed of reaction ( latency ) to arbitrage opportunities (i.e., λ 1 or γ 1 ) but being faster is costly: an increase in speed by one unit cost c j to trader j {X, Arb}. ˆ π = γ γ+λ = A measure of arbitrageurs relative speed. ˆ Expected duration of an arbitrage opportunity: E (D) = ϕ(e (Min{D a, D X })) + (1 ϕ)e (D a ) = ϕ λ + γ (1 ϕ) +. γ

30 Equilibrium ˆ We solve for equilibrium spreads, speeds, duration of arbitrage opporunities and π in two steps: 1. For given bid-ask spreads, Nash equilibrium in speeds: traders choose the speed that maximizes their expected profit: 1.1 In equilibrium traders speed decrease with bid-ask spreads Arbitrage opportunities last longer when spreads are larger. 2. Then, we solve for zero profit spreads.

31 Equilibrium ˆ Let ρ = cx = Relative Cost of Speed: higher ρ means c Arb the cost of speed becomes relatively smaller for arbitrageurs. ˆ In equilibrium: Relative Speed : π = γ /(λ + γ ) = ρ 1 + ρ

32 Equilibrium ˆ Let ρ = cx = Relative Cost of Speed: higher ρ means c Arb the cost of speed becomes relatively smaller for arbitrageurs. ˆ In equilibrium: Relative Speed : π = γ /(λ + γ ) = ρ 1 + ρ Illiquidity : S X = ϕ π α(2 π )σ απ (2 π )σ + (1 α).

33 Equilibrium ˆ Let ρ = cx = Relative Cost of Speed: higher ρ means c Arb the cost of speed becomes relatively smaller for arbitrageurs. ˆ In equilibrium: Relative Speed : π = γ /(λ + γ ) = ρ 1 + ρ Illiquidity : S X = ϕ π α(2 π )σ απ (2 π )σ + (1 α). Duration: E (D) = (1 + ρ ϕ) γ (ϕ, ρ, SX )(1 + ρ) }{{} Arb s Speed

34 Testable implications ˆ Imp.1a (Composition effect): An increase in the fraction of arbitrage opportunities that are toxic (ϕ) causes an increase in illiquidity.

35 Testable implications ˆ Imp.1a (Composition effect): An increase in the fraction of arbitrage opportunities that are toxic (ϕ) causes an increase in illiquidity. ˆ Imp.1b (Speed effect): An increase in arbitrageurs speed relative to dealers speed (π) causes an increase in illiquidity.

36 Testable implications ˆ Imp.1a (Composition effect): An increase in the fraction of arbitrage opportunities that are toxic (ϕ) causes an increase in illiquidity. ˆ Imp.1b (Speed effect): An increase in arbitrageurs speed relative to dealers speed (π) causes an increase in illiquidity. ˆ Imp.2: A decrease in arbitrageurs relative cost of speed (an increase in ρ) triggers an increase in arbitrageurs relative speed (π) and, through this channel, causes an increase in illiquidity.

37 Testable implications ˆ Imp.1a (Composition effect): An increase in the fraction of arbitrage opportunities that are toxic (ϕ) causes an increase in illiquidity. ˆ Imp.1b (Speed effect): An increase in arbitrageurs speed relative to dealers speed (π) causes an increase in illiquidity. ˆ Imp.2: A decrease in arbitrageurs relative cost of speed (an increase in ρ) triggers an increase in arbitrageurs relative speed (π) and, through this channel, causes an increase in illiquidity. ˆ Imp.3: An increase in the fraction of arbitrage opportunities that are toxic (ϕ) causes a reduction in the duration of arbitrage opportunities.

38 Testable implications ˆ Imp.1a (Composition effect): An increase in the fraction of arbitrage opportunities that are toxic (ϕ) causes an increase in illiquidity. ˆ Imp.1b (Speed effect): An increase in arbitrageurs speed relative to dealers speed (π) causes an increase in illiquidity. ˆ Imp.2: A decrease in arbitrageurs relative cost of speed (an increase in ρ) triggers an increase in arbitrageurs relative speed (π) and, through this channel, causes an increase in illiquidity. ˆ Imp.3: An increase in the fraction of arbitrage opportunities that are toxic (ϕ) causes a reduction in the duration of arbitrage opportunities. ˆ Imp.4: A decrease in arbitrageurs relative cost of speed (an increase in ρ) reduces the duration of arbitrage opportunities. Faster arbitrageurs reactions to toxic arbitrage opportunities make the market less liquid but always more price efficient.

39 Numerical Example Bid Ask Spread bps Likelihoodtoxic arbitrage φ Arbitrageur's relative speed Bid Ask Spread bps Ρ c m c a ArbitrageDuration sd Likelihoodtoxic arbitrage φ ArbitrageDuration sd Arbitrageur's Cost of Speed

40 Data ˆ Tick-by-tick data ( ) from Reuters D-3000: an interdealer limit order book in the FX market ˆ Three currency pairs: $/e, $/ and e/ ˆ All orders: limit, market, cancellations etc ˆ Time-stamped accuracy at the one-hundredth of a second

41 Triangular Arbitrage

42 Triangular arbitrage opportunities ˆ Two ways to buy euros with dollar: 1. Direct: Buy e1 at A $/e, the ask price in dollar for euros. Cost: A $/e 2. Indirect: Buy A /e units of pounds at A $/ and then e1 at A /e in the euro/sterling market. Cost: Â $/e =A /e A $/. ˆ Two ways to sell euros against dollar: 1. Direct: Sell e1 at B $/e, the bid price in dollar for euros. Revenue: B $/e 2. Indirect: Sell e1 at B /e in the euro/sterling market and then sell B /e units of pounds at B $/. Revenue: B$/e =B /e B $/.

43 Triangular Arbitrage Opportunities ˆ A triangular arbitrage opportunity exists if: A $/e < OR B $/e  $/e < B $/e ˆ Practically, we just focus on arbitrage opportunities that secure a profit before costs of at least 1bps. ˆ # triangular arbitrage opportunities in sample: 37,689 over two years.

44 Triangular Arbitrage Opportunities ˆ Similar in nature to those exploited by high speed arbitrageurs, i.e.: 1. Short-lived (last for about 1 second and sometimes much less) 2. Almost riskless 3. Deliver a very small profit per opportunity 4. Winner takes it all flavor ˆ All opportunities take place within the same trading platform: Synchronisation of time stamps across markets is not an issue (a major hurdle for empirical studies of high frequency arbitrage).

45 Toxic vs. Non-Toxic Arbitrage opportunities: Classification Panel A: Toxic arbitrage opportunities (permanent shifts in prices) Panel B: Non-toxic arbitrage opportunities (price reversals) ˆ # toxic triangular arbitrages in sample: 15,908.

46 Toxic and Non Toxic Arbitrage Opportunities: Time-Series

47 Arbitrage opportunities breakdown

48 Proxies for Dealers Exposure to Toxic Arbitrage Trades ˆϕ t = # Toxic arbitrage opportunities on day t. # Arbitrage opportunities on day t ˆ π t = # Toxic opportunities closed by a trade on day t # Toxic Arbitrage opportunities on day t

49 Toxic vs. Non-Toxic Arbitrage opportunities Toxic Non Toxic Daily measures Median SD Median SD Duration (msd) Nbr Arb ˆϕ(%) Arb Size (bps) Profit (bps) π (%) ˆ Profit per opportunity are small but the total daily profit on triangular arbitrages (about $5,000) is of the order of magnitude of that found for HFTs on Nasdaq (see Brogaard, Hendershott and Riordan (2012)). ˆ π for toxic and non toxic arbitrage opportunities have a zero correlation (0.08) = do not capture the same phenomenon.

50 Liquidity measures ˆ Note: Bid-ask spreads are expressed in pips: 1 pip = 1 bps. Minimum trade size = 1 million of the base currency in each pair.

51 Main Test ˆ We estimate the following regression for the three currencies in our sample: Ill it = α i + β t + b 1 πˆ t + b 2 ϕˆ t + b 3 Arbsize t + b 4 ˆα it + b 5 Vol it + + b 6 Trsize it + b 7 #Orders it + b 8 Illiq EBS it + ɛ it Predictions: b 1 > 0 and b 2 > 0. ˆ Control variables: daily average arbitrage size (σ in the model), daily ratio of the number of arbitrage opportunities to number of trades (α), daily realized volatility, daily average trade size in millions, daily number of orders, illiquidity on EBS platform

52 IV Approach ˆ Reverse Causality Problem: Illiquidity also affects π: Arbitrageurs have less incentive to be fast when trading costs are large. ˆ Instrument: the introduction of AutoQuote (API) by Reuters D-3000 in July ˆ AutoQuote API (Application Programming Interface): Enable traders using Reuters D-3000 to automate order entry onset of algo trading on Reuters. ˆ Increase in traders speed. Should affect π without directly affecting illiquidity.

53 Findings spread espread slope 1 st stage 2 nd stage 1 st stage 2 nd stage 1 st stage 2 nd stage AD (4.09) (4.12) (4.10) ˆπ (3.91) (3.70) (3.96) ˆϕ (-0.31) (2.29) (-0.31) (3.68) (-0.28) (2.61) ˆσ (-2.14) (4.93) (-2.17) (9.94) (-2.11) (4.39) vol (-0.75) (3.72) (-0.77) (8.65) (-0.76) (3.87) trsize (0.66) (-0.30) (0.84) (-0.98) (0.76) (-1.09) nrorders (0.27) (-0.77) (0.22) (-2.62) (0.30) (-1.01) illiq EBS (-3.88) (0.79) (-3.85) (-0.43) (-3.89) (0.08) Adj.R % 34.40% 2.34% 62.18% 2.35% 25.56% Fstat Currency pair FE Month dummies YES YES YES YES YES YES

54 Economic size of the effects ˆ A 1% increase in the likelihood that a toxic arbitrage terminates with an arbitrageur s trade ( ˆπ) raises bid-ask spread by about 4% (0.08bps) ˆ This effect translates in a quite large increase in trading costs given the trading volume for the currencies in our sample (average trade size of about 1.8 mio with about 2,500 trades per day). We estimate that the increase in trading costs due to a 1% increase in: ˆ ˆπ is $161,296 (about $40 mio per year) ˆ ˆϕ is $14,047 (the daily standard deviation of ˆϕ is 10%)

55 Arbitrage and Pricing Efficiency (Implications 3 and 4) Dep.Var: log(duration Arb.) Toxic AD (-3.04) (-2.93) vol (-3.15) (-4.53) ˆϕ (-2.95) (0.68) ˆσ (6.59) (9.22) trsize (0.18) (0.14) nrorders (-7.29) (-7.40) Adj.R % 33.33% All ˆ The introduction of Autoquote API ( AD ) reduces by about 6.8% the duration of toxic arbitrage opportunities (i.e., by about 62 milliseconds; the average duration is 890 milliseconds).

56 Conclusions 1/2 ˆ Arbitrage and liquidity: 1. The mix of arbitrage opportunities matters: more arbitrage opportunities due asynchronous price adjustments are associated with less liquidity. 2. Arbitrageurs speed of reaction matters: Faster arbitrageurs reaction to toxic opportunities lower liquidity. ˆ Future Work: Do benefits of high speed arbitrageurs offset adverse selection costs? 1. Faster price discovery? Do we care about prices being right 60 ms faster? Why? 2. Faster response to transient liquidity shocks? Maybe...needs to be modeled and quantified, however.

57 Conclusions 2/2 ˆ How to solve the problem? Lower arbitrageurs relative speed π: 1. Batch auctions 2. Random delay to the execution of market orders. ˆ Calibration of regulatory intervention: should depend on the composition of arbitrage opportunities. Unlikely to be the same across all asset pairs! ˆ Need to look more systematically at the composition of arbitrage opportunities across asset pairs (ETFs, options, futures etc.