Dynamic Strategic Arbitrage∗
Vincent Fardeau† Frankfurt School of Finance and Managment
This version: June 11, 2014
Abstract
Asset prices do not adjust instantaneously following uninformative demand or supply shocks. This paper offers a theory based on arbitrageurs’ price impact to explain these slow-moving cap- ital dynamics. Arbitrageurs recognizing their price impact break up trades, leading to gradual price adjustments. When shocks are anticipated, prices exhibit a V-shaped pattern, with a peak at the realization of the shocks. The dynamics of market depth and price adjustments depend on the degree of competition between arbitrageurs. When shocks are uncertain, price dynamics depend on the interaction between arbitrageurs’ competition and the total risk-bearing capacity of the market.
JEL codes: G12, G20, L12; Keywords: Strategic arbitrage, liquidity, price impact, limits of arbitrage.
∗I benefited from comments from Miguel Ant´on,Bruno Biais, Maria-Cecilia Bustamante, Amil Dasgupta, Michael Gordy, Martin Oehmke (discussant), Matt Pritsker, Jean-Charles Rochet, Yuki Sato, Kostas Zachariadis, Jean-Pierre Zigrand and seminar and conference participants at LSE, INSEAD, FRB, the University of Zurich and the 2013 AFA meetings in San Diego. I am grateful to Dimitri Vayanos and Denis Gromb for supervision and insights. Jay Im provided excellent research assistance. This paper is the first part of a paper previously circulated as “Strategic Arbitrage with Entry”. All remaining errors are my own. †Department of Finance, Frankfurt School of Finance and Management, Sonnemannstrasse 9-11, 60314 Frankfurt am Main, Germany. Email: [email protected].
1 1 Introduction
Empirical evidence shows that prices recover slowly from demand or supply shocks that are unre- lated to future earnings and that these patterns are due to imperfections in the supply of capital.
Patterns of slow price adjustments due to these sluggish capital responses have been documented, for instance, in the catastrophe reinsurance markets (Froot and O’Connell, 1999), and in numerous popular arbitrage strategies used by hedge funds such as convertible debt and merger arbitrage
(Mitchell, Pedersen and Pulvino, 2007).1 These price patterns present a challenge for models of efficient capital markets based on competitive, rational investors, which predict that the supply of capital is perfectly elastic.
This paper proposes a theory based on arbitrageurs’ price impact to explain why capital appears to be slow-moving. There is widespread evidence that maket depth is limited, even in actively traded markets, and that investors develop strategies to minimize their price impact.2 Arbitrage in particular typically involves setting up positions in relatively illiquid markets that are often dominated by a few highly specialized sophisticated investors.3 I consider a model in which supply shocks and market segmentation generate price discrepancies between two identical assets. I show that when arbitrageurs recognize their price impact, these price discrepancies are not eliminated instantaneously. Strategic arbitrageurs break up their trades to minimize price impact and exploit price discrepancies longer, which results in gradual price convergence. The interesting feature of the model is that arbitrageurs’ price impact and the quantity they trade are both endogenous.
The model generates additional predictions which match well-documented empirical findings, such as V-shaped price patterns around anticipated (and uninformative) supply shocks, time- varying market depth, and changes in the sign of basis between identical or closely-related assets.
Moreover, the model is extended to analyze arbitrageur’s behaviour when the supply shocks are
1See Duffie (2010) for additional examples. 2See Chan and Lakonishok (1993, 1995) and additional references in Section 2. 3Anecdotal evidence reveals that some strategies can be dominated by a one or a handful of arbitrageurs. For instance, LTCM was nicknamed the central bank of volatility (Lowenstein, 2000) as it was selling large amounts of out-of-the-money put options and hedging in the underlying securities. Recently, JP Morgan was put under the spotlight after one of it’s traders in the Chief Investment Office accumulated large positions in the CDS markets. The trader, who was called the ’London Whale’ by the press and other market participants, was reportedly involved in yield curve arbitrage. See ‘London Whale’ Rattles Debt Market”, Wall Street Journal, April 6, 2012.
2 stochastic, i.e. when the arbitrage is risky, and study whether arbitrageurs “lean against the wind”.
I consider a multiperiod model in which arbitrageurs exploit price differences between two identical assets traded in segmented markets (e.g. Siamese twin stocks, a derivative and the equiv- alent security based on the underlying, etc.). In each local market, competitive investors receive endowment shocks that are correlated with the asset payoff. Correlations to the asset payoff are
(for simplicity) opposite across local investors, thus market segmentation causes prices to diverge.
Stronger hedging needs from local investors create a larger arbitrage “supply” and thus larger price differences. Arbitrageurs can trade in both markets and act as middlemen until assets mature and pay off the same dividend.4 Arbitrageurs are imperfectly competitive. While intermediating trades, they understand that they face downward-sloping demand curves in each market and take into ac- count the impact of their trades on market-clearing prices. This setting with complete information is suitable to analzye cleanly the effects of uninformative supply shocks on price dynamics.5
If arbitrageurs take prices as given, the spread between the prices of the two assets drops instantaneously to zero. Since arbitrageurs are perfectly hedged across markets A and B, they are effectively risk-neutral, and eliminate any mispricing immediately, resulting in perfect risk-sharing and liquidity. By contrast, imperfectly competitive arbitrageurs intermediate only a limited amount between the two markets to keep the spread open and maximize capital gains. Further, as trades have a permanent effect on prices, arbitrageurs break up their trades to exploit the arbitrage as long as possible. When there is a constant arbitrage supply, arbitrageurs reallocate the asset slowly from one market to the other, giving rise to slow-moving capital. A similar mechanism implies that when unanticipated supply shocks occur at some stage, their effects on prices die out only gradually. In both cases, the equilibrium speed of arbitrage increases as more arbitrageurs compete in the trade.
This predictable price pattern arises in equilibrium even though all investors have perfect
4The assumptions of market segmentation and identical assets emphasize arbitrageurs’ role as intermediaries but are not essential for the results. 5Apart from hedging needs, supply shocks can arise from portfolio rebalancing of passive investors (e.g. index funds or ETFs) around index recompositions, forced liquidations of market participants following shocks to their portfolios (e.g. LTCM unwound its convertible arbitrage activity following losses on other convergence trades), etc. Gromb and Vayanos (2010) provide additional examples.
3 foresight. What induces local investors in, say, market A, to sell today given that in that market the price of the asset will (on average) increase tomorrow? Local investors are willing to sell today because waiting is risky. New information about the fundamental accrues between today and tomorrow and will affect tomorrow’s actual price.6 Nevertheless, the anticipation of a more favorable price reduces local investors’ willingness to sell today. Hence local investors’ perfect foresight erodes arbitrageurs’ market power, as in the classical durable goods monopoly problem.
The slow-moving dynamics rely crucially on the discreteness of trading and the finiteness of the horizon. The discreteness implies that receiving liquidity today is not exactly equivalent to receiving liquidity tomorrow, while the finite horizon provides arbitrageurs a means to commit to provide only limited liquidity in the future.
The model generates additional predictions, including V-shaped price effects around the real- ization of anticipated shocks, time-varying market depth, and sign changes of price differences in case of negative decreasing supply. V-shaped price patterns have been documented around corpo- rate or Treasury bond issuances, and around pre-announced index recompositions, which trigger shifts in uninformed demand for stocks (e.g. from passive investors such as index funds and ETFs), resulting in changes in the net supply.7 These patterns have proven hard to reconcile with rational behavior of market participants because price decreases and reversals are gradual and predictable.
This type of price patterns emerges in the model around the realization of pre-announced shocks, even though all market participants are rational. After the realization of the shocks, prices adjust gradually as in the case of surprise shocks. Indeed, arbitrageurs avoid providing any addi- tional liquidity ahead of the shock, so that much of the shock absorption occurs after realization.
The anticipation of this behavior by local investors generates an increasing spread up to the real- ization. The striking feature of this result is that, between the announcement and the realization
6Vayanos and Woolley (2008) have a similar effect, which they term a bird-in-the-hand effect. 7Keloharju, Malkmaki, Nyborg and Rydqvist (2002) show that heavy funding needs of the Treasury cause V- shaped price patterns around new issuances in the Finnish market. Prices decline ahead of the issuance and recover afterwards. Lou, Yan and Zhang (2013) find similar secondary market price pressure effects around US Treasury issuances. Newman and Rierson (2003) show that bond issuance of one firm in the European telecom market temporarily raises yields of other firms in the sector. Index recompositions are either announced in advance or can be guessed with a high precision due to mechanical rules. See Lynch and Mendenhall (1997), Madhavan (2000), Greenwood (2005) and Boyer (2011) for evidence about the effects of index recompositions on prices. The V-shaped patterns can be observed for added and deleted stocks.
4 of the supply shock, the spread increases only gradually instead of instantaneously. As with unan- ticipated shocks, the discreteness of trading and time to maturity of the asset are important for the result. As time passes, two forces determine the spread dynamics. First, arbitrageurs’ trades accumulate, leading to a decreasing spread. Second, the opportunities to retrade vanish, so that local investors anticipate less and less liquidity provision, increasing the impact of future shocks on the spread. In the case of constant supply or surprise shocks, the first effect dominates and the spread decreases gradually in equilibrium. When shocks are pre-announced, the second effect dominates during the interim period between announcement and realization, leading to a peak of the effect at realization of the shock.
Time to maturity is also a key driver of market depth dynamics. Since the opportunities to further share risks become scarcer as time passes, local investors’ willingness to hold the asset decreases over time. This implies that - controlling for conditional dividend volatility, which de- creases as more information accrues over time - price impact increases as time passes. This effect is stronger when fewer arbitrageurs compete in the trade, since imperfect competition further lim- its risk-sharing opportunities. The non-stationarity of endogenous price impact stands in sharp contrast to the assumption of constant exogenous price impact that is usual in the literature on optimal order execution, showing that for large traders/ orders, it is important to consider jointly optimal trading strategies and market depth.8 Time-varying market depth is a feature of equity markets (Chordia, Roll and Subramanyam, 2000, 2001), Treasuries (Fleming, 1999) and corporate bonds (Bao, Pan, Wang, 2008).
In addition to gradual and V-shaped price adjustment, the model also predicts that the spread can change sign when the supply decreases in a predictable way, even if the supply does not change sign. This implies that assets which are undervalued for liquidity reasons, can become overvalued, i.e. prices overshoot the fundamental value. In other words, deterministic decreases in supply, instead of simply narrowing the spread also change its sign. This result may shed light on puzzling empirical evidence of such change in the sign of bases between closely related assets.
For instance, Bergstresser et al. (2010) find that insured municipal bonds became cheaper than
8The examples of LTCM and JP Morgan given in footnote 3 are good illustrations of this result.
5 uninsured bonds of the same municipal with similar characteristics in the aftermath of the financial crisis, i.e. precisely when liquidity needs were decreasing, which is one interpretation of the decrease in supply in the model.
Finally, I study arbitrageurs’ reactions to arbitrage risk. This analysis speaks to the old question of whether arbitrageurs “lean against the wind” (e.g. Friedman, 1953). When the future supply is stochastic, arbitrageurs’ positions increase in the size of the average future shock. This increase, however, does not reduce the spread because prices also reflect the anticipation of a larger shock, and this effect dominates. Models with financially constrained arbitrageurs (e.g. Shleifer and Vishny,
1997, Gromb and Vayanos, 2002) generate the same price effect, but for a different reason. In these models, the spread increases because arbitrageurs decrease their positions to save on capital for later. I also analyse arbitrageurs and price responses to increases in the volatility of future supply shocks, which is not analysed by the previous models. Since a higher shock volatility makes it more likely to face large price gaps, one could expect that arbitrageurs respond by trading more aggressively. I find that when the volatility of shock increases, arbitrageurs do not necessarily increase their positions, even if they are risk-neutral. This is because a larger shock volatility leads to a thinner market, inducing arbitrageurs to reduce their trades. These two results show that arbitrageurs’ willingness to lean against the wind may not only depend on their risk-aversion or trading capacity (determined, for instance, by inventory considerations, or financial constraints), as emphasized by the previous literature, but more broadly on the interaction between the market structure and the overall risk-bearing capacity of the market.
A number of empirical findings support or are consistent with the predictions of the model.
Some papers find evidence that arbitrageurs’ market power is a driver of slow-moving capital and mispricings. Froot (2001) shows that supply-side stories and market power exerted by intermedi- aries are the most likely explanations of slow-moving capital in the catastrophe reinsurance markets.
Ruf (2012) studies the difference between implied and realized skewness in commodities options. He shows that the difference between the two is larger when traders hold more concentrated positions.
Others document more generally the explanatory role of market thinness (the inverse of market depth) for mispricings. For instance, Pelizzon et al. (2014) show that market thinness predicts the
6 bond-futures basis in the European sovereign bond market.
The paper relates to several strands of the literature. First, the paper belongs to the literature studying models where an oligolopoly of large (Cournot) traders trade with a competitive sector
(DeMarzo and Urosevic, 2006, Pritsker, 2009, Kihlstrom, 2000). My main contribution is to high- light new price effects caused by time variation in the asset supply, and to extend the analysis to risky asset supply.9 Pritsker’s paper is closest to this paper. He studies the effects of distressed sales on price dynamics, but does not consider random changes in asset supply. Further, arbitrageurs have heterogeneous levels of risk-aversion in Pritsker’s model, and price dynamics in the presence of distressed sales are obtained via numerical simulations. In this paper, the arbitrageurs are effec- tively risk-neutral when the supply is deterministic and risk-averse when the supply is stochastic.
This allows me to characterize the full dynamics of price adjustments in closed-form and to derive new comparative statics. Another related paper is Oehmke (2010), who studies the price effects of large arbitrageurs in a dynamic setting, but models competitive investors as exogenouly given demand curves. Instead, market depth is endogenously time-varying in my model.10
Vayanos (1999), and more closely Rostek and Weretka (2011), study dynamic extensions of Kyle
(1989), where an oligopoly of large traders compete in demand schedules (and there is no competi- tive sector). Rostek and Weretka study the price effects of unanticipated and pre-announced supply shocks, but do not study random shocks. The main difference between our predictions concerns price dynamics. In the present paper, prices are affected by market illiquidity.11 This implies that unanticipated shocks generate gradually adjusting prices and that pre-announced shocks generate
V-shaped returns around the realization date, which are consistent with empirical evidence. By contrast, equilibrium prices coincide with competitive prices in the absence of shocks in Rostek and Weretka’s model (i.e. the market is illiquid, but there is no liquidity premia). As a result, pre-announced shocks affect prices at announcement and realization but not in-between, as returns
9Kilhstrom studies a three-period model with fixed asset supply and focuses on the analogy between asset pricing with market power and the durable goods monopoly problem. DeMarzo and Urosevic extend Kihlstrom’s model to multiple periods and focus on a moral hazard problem, whereby the arbitrageur is a firm insider whose incentives to exert effort depends on his asset holdings. 10Zigrand (2004) considers a static setting and focuses on strategic arbitrageurs’ entry decisions. 11This feature corresponds to empirical findings. See for instance, Amihud and Mendelson (1986, 2002) and Acharya and Pedersen (2005).
7 coincide with competitive returns in the interim period. It is the interaction between the market illiquidity and the uncertainty about the fundamental value (which decreases as time passes in my model, whereas it is realized in one go in the final period in Rostek and Weretka’s setting) that generates liquidity premia in the present model. Thus the comparison of our results shows that liquidity affects prices when trading and the realization of dividend news take place simultaneously.
Previous research has shown that gradual price recovery after shocks can result from financial constraints on arbitrageurs or search frictions to contact intermediaries (Duffie and Strulovici,
2012).12 Our model offers a complimentary mechanism and delivers different predictions. It predicts that these gradual adjustments can be observed not only in bad times (that is, when constraints bind), and that they are slower in more concentrated markets.13
Few models with rational investors generate V-shaped price patterns. That is, most models generate either gradual price declines, or gradual reversals, but not both “at the same time”.
Vayanos and Woolley (2013) and Albuquerque and Miao (2014) are two notable exceptions. Vayanos and Woolley consider a model of delegated investment where gradual flows in and out of active funds generate momentum and reversals. Albuquerque and Miao obtain momentum and reversal in a model where some investors have advance information about future earnings that is unrelated to current earings. However, as noted above, gradual price declines and reversals occur also around uninformative supply shocks. This paper shows that gradual declines and reverals can occur also in the absence of asymmetric information and even if arbitrageurs are not subject to short-term outflows (e.g. thanks to lock-up periods).
The paper proceeds as follows. In section 2, I outline the three-period baseline model. Many of the results can be derived in this framework with deterministic supply shocks, as shown in Section
3. I extend the analysis to risky arbitrage in Section 4. Section 5 considers the T -period model and analyse surprise and pre-announced shocks. Section 6 summarizes the empirical predictions and
12See for instance Gromb and Vayanos (2002), Brunnermeier and Pedersen (2009) for an analysis of financially- constrained arbitrage. 13Of course, binding constraints or arbitrageurs’ defaults can lead to more market concentration in bad times, and search frictions may be lower in concentrated markets. However, Duffie and Strulovici (2012) show that more competition among intermediaries can reduce the speed of price adjustment when search frictions are present.
8 concludes. All proofs of Sections 3 and 4 are in the appendix. The proofs of Section 5 are available in the supplementary appendix.
2 Baseline model
In this section, I introduce the three-period baseline model. I consider an economy in which imperfectly competitive arbitrageurs exploit price differences between two identical assets traded in segmented markets.
2.1 Set-up
Assets and timeline. The economy has three periods (t = 0, 1, 2), and consists of two identical risky assets (A and B) and a risk-free asset. The risky assets are in zero net supply and pay off a liquidating dividend at time 2,
D2 = D + 1 + 2,
2 where t are i.i.d. normal variables with mean 0 and variance σ . The innovations t are realized and revealed to all market participants at time t = 1, 2. I denote Dt = Et (D2), the conditional expected value of the dividend. The risk-free asset is in perfectly elastic supply and its return r is normalized to 0. Trading takes places at time 0 and time 1, and consumption at time 2.
Local investors. The economy is made of two types of investors: competitive local investors and imperfectly competitive arbitrageurs. There is continuum of local investors in markets A and B, represented by a competitive agent with CARA utility with absolute risk-aversion coefficient a in
9 each market:14
k k u C2 = − exp −aC2 , k = A, B
Local investors in market k can trade only the k-risky asset and the risk-free asset. Thus, markets A and B are segmented from the point of view of local investors. Market segmentation can stem from institutional constraints or behavioral biases. For instance, mutual funds are usually restricted by their mandates to trade in the cash market (e.g. bonds), while other market participants such as banks or insurance companies might prefer to use derivatives (e.g. CDS). Siamese stocks trade on different exchanges and local investors may incur costs in the form of taxes, agency frictions or behavioral costs inducing them to trade in their domestic exchange (Froot and Dabora, 1999).15
Market segmentation limits valuable risk-sharing between local investors, as follows. Local investors in each market receive endowment shocks that are correlated with the payoff of their risky
k asset. Specifically, investors in market k receive a shock st−1t at time t = 1, 2. st−1 represents the magnitude of local investors’ exposure to dividend risk and will determine local investors’
A hedging demand in period t − 1. For instance, if investors in market A have exposure s0 > 0, they will be willing to sell the asset at time 0 to insure against their endowment risk. Notice that
k although the endowment shock is realized at time t, the exposure st−1 is known (at the latest – A B see next paragraph) at time t − 1. Shocks are opposite across markets: st−1 = −st−1 = st−1, so that local investors could perfectly share risk by trading with each other. However, market segmentation prevents direct trading between local investors and creates a role for arbitrageurs.
Indeed, arbitrageurs can buy from investors with low valuation and sell to investors with high valuation to capture the price difference (spread).
In the baseline case, all investors know in advance the values of s0 and s1. Thus there is 14The assumption of a representative agent for competitive investors is standard in the finance literature on dynamic markets with large traders (see DeMarzo and Urosevic (2006), Pritsker (2009), Kihlstrom (2000)) but is not without consequences for the analysis. Under this assumption, only unilateral deviations by strategic arbitrageurs affect the outcome of the game. Gul, Sonnenschein and Wilson (1985) consider a game between a monopolist seller and non- atomic buyers but impose regularity conditions on buyers’ strategies that imply that unilateral deviations by buyers do not affect other buyers’ actions nor the monopolist’s actions. (See also Ritzberger (2002)). 15One can also think of local investors as hedgers with opposite hedging needs. See Gromb and Vayanos (2010) for additional examples of supply or demand shocks and market segmentation.
10 no uncertainty about the spread and arbitrageurs face a textbook arbitrage opportunity, which
A B disappears at time 2 when assets A and B pay off (p2 = p2 = D2). In Section 4, I extend the baseline model to the case where investors know only the distribution of s1 at time 0. In this case, arbitrageurs face the risk that the spread increases or decreases at time 1. This case is called risky arbitrage.
At time 2, local investors consume their entire wealth:
k k k k C2 = W2 = Y2 D2 + E2 , k = A, B
k k where Y2 denote the position in the risky asset and E2 the position in the risk-free asset. I denote k k k k k yt the time t trade in risky asset k and pt its price. The law of motion of positions is: Yt = Yt−1+yt k k k k k 16 for the risky asset and Et = Et−1 − yt pt + st−1t, for the risk-free asset, for k = A, B. Local investors solve the following problem:
h ki for k = A, B, max E u C2 (1) Y k ( t )t=0,1
k k k s.t. C2 = Y2 D2 + E2
k k k Yt = Yt−1 + yt
k k k k k Et = Et−1 − yt pt + st−1t
Arbitrageurs. Arbitrageurs can trade all assets without restriction. Thus they can intermediate trades by buying from investors with low valuation (say, in market A) and selling to investors with high valuation (in market B). Doing so, arbitrageurs will contribute to integrate markets A and B and provide liquidity to local investors. The main assumption is that arbitrageurs recognize the price impact of their trades.
Price impact costs are a first-order concern for sophisticated investors seeking to profit from
16In the three-period model, there is no trading at time 2, and no endowment sock at time 0, implying that
k k k k k k k k k E2 = E1 + s12,E1 = E0 − y1 p1 + s01,E0 = E−1 − y0 p0
k Since local investors have CARA preferences, we can set their initial endowment E−1 = 0 without loss of generality.
11 misvaluations such as hedge funds. Given that mispricings are often caused by some unmet liquidity demand, limited liquidity is a natural issue for arbitrageurs. A well-known example is that of LTCM: many of the hedge fund’s trades involved betting on assets with identical or closely related payoffs to exploit price discrepancies. However, as Perold (1999) and Lowenstein (2000) report, as the fund size increased, the impact of LTCM’s trades on prices were limiting the profitability of it’s strategies. 17
Price impact costs are likely to be a major issue in trades where only a few arbitrageurs are active or in specialized markets (Attari and Mello, 2002). For instance, convertible arbitrage involves buying illiquid convertible debt to exploit price differences with the synthetic asset made of the underlying securities. Further, there is a large body evidence showing that even very active markets such as the equity, Treasury and currency markets of developed markets may offer limited depth.18
Price impact is modeled via a Cournot game of complete information among arbitrageurs.
There are n imperfectly competitive arbitrageurs indexed by i (i = 1, . . . , n < ∞), each endowed with CARA utility with absolute risk-aversion coefficient b: