MIT LIBRARIES 3 9080 02617 9298 .M415 tiMimumuiUi'ifiiiitu^itiiitniiiti'' no.06-07 2006 i Digitized by the Internet Archive in 2011 with funding from Boston Library Consortium IVIember Libraries http://www.archive.org/details/flighttoqualitycOOcaba OEWEV HB31 .M415 Massachusetts Institute of Technology Department of Economics Working Paper Series Flight to Quality and Collective Risk Management Ricardo J. Caballero Arvind Krishnamurthy Working Paper 06-07 March 15,2006 RoomE52-251 50 Memorial Drive Cambridge, 02142 MA This paper can be downloaded without charge from the Network Paper Collection at Social Science Research httsp:/7ssrn.com/abstract=891856 I MASSACHUSETTS INSTITUTE OF TECHNOLOGY LIBRARIES Flight to Quality Ricardo and Collective Risk Arvind Krishnamurthy* Caballero J. Management This draft: March 15, 2006 Abstract We present a model of flight to quality episodes that emphasizes systemic risk and the Knightian uncertainty surrounding these episodes. Agents edge. They understand their own systemwide shocks. Aversion to make risk management decisions with incomplete knowl- how correlated their shocks are with shocks, but are uncertain of them this uncertainty leads to question whether their private risk man- agement decisions are robust to aggregate events, generating conservatism and excessive demand safety. We wasteful in show that agents' actions lock-up the capital of the financial system in a the aggregate and can trigger and amplify a financial accelerator. lective of conservative agents are incomplete knowledge. A guarding against impossible, and is lender of last resort, even less if known The for manner that is scenario that the col- to be so even given agents' knowledgeable than private agents about in- dividual shocks, does not suffer from this collective bias and finds that pledging intervention in extreme events is The valuable. benefit of such intervention exceeds its direct value because it unlocks private capital markets. JEL Codes: E30, E44, E5, F34, Gl, G21, 022, 028. Keywords: Locked collateral, flight to quality, insurance, safe and risky claims, financial intermediaries, collective bias, lender of last resort, private sector multiplier, collateral shocks, robust control. 'Respectively: We MIT and NBER; Northwestern Universitj'. are grateful to Marios Angeletos, Olivier Blanchard, Phil William Hawkins, Burton Hollifield, Conference, Bank of England, issues as (and hence replaces) "Financial NY Fed Liquidity Conference, and the Central Bank of Chile provided excellent research assistance. Caballero thanks the more abstract model. caball@mit.edu, a-krishnamurthy@northwestern.edu. Gali, Michael Golosov, Bengt Holmstrom, Dimitri Vayanos, Ivan Werning and seminar participants DePaul, Imperial College, LBS, Northwestern, MIT, Wharton, Summer E-mails: Bond, Jon Faust, Xavier Gabaix, Jordi NSF System Risk and Flight for their for financial to Quality," comments. NBER MIDM at Columbia, Conference, UBC David Lucca and Alp Simsek support. This paper covers the same substantive NBER WP # 11996 (2005) with a simpler but 1 Introduction Polic}' practitioners operating "... under a risk-management paradigm may, at times, be led to When undertake actions intended to provide insurance against especially adverse outcomes confronted with uncertainty, especially Knightian uncertainty, medium human beings invariably attempt liquidity.... The immediate response on the part of the central bank to such financial implosions must be to inject to disengage from to long-term commitments in favor of safety' and Alan Greenspan (2004). large quantities of liquidity..." Modern Flight to quality episodes are an important source of financial and macroeconomic instability. examples of these episodes US in the include the Penn Central default of 1970: the stock market crash of 1987; the events of the Fall of 1998 beginning with the Russian default and ending with the bailout of LTCM; as well as the events that followed the attacks of 9/11. Behind each of these episodes meltdown that may lead Japan during the 1990s, or even a catastrophe slowdown as to a prolonged in lies the specter of a like the Great Depression.^ In each of them, as hinted by Greenspan, the Fed intervened early and stood read}' to intervene "as Two much as needed" to prevent a questions immediately arise to quality episodes. First, how can when meltdown. comments on flight private wealth create so much considering these examples and Greenspan's events that appear small relative to US havoc? Second, how can interventions by the Fed, that also appear small relative to in preventing a meltdown? answers are that during its ability we develop a model wealth, be effective to address these questions. episodes private capital becomes locked up, and flight to quality of intervention derives from Our model In this paper US much In short, our of the power to unlock this capital. builds on two observations about flight to quality events. First, at a basic level these episodes are about an increase in perceived "risk." However, the risk does not circumscribe a purely fundamental shock, and instead centers around the nerve center of the economy, the financial system. For example, the Russian default in the summer of 1998 eliminated a small fraction of the trillions of dollars of Although small, the default created circumstances that severely strained the illiquid assets fell, losses grew in and moved into short-term and 'See Table century. 1 (part A) The primary in Barro (2005) risk withdrew liquid assets. as sophisticated parts of the financial system were relatively unaffected. As wealth. prices of commercial banks, investment banks, and hedge funds, leading investors to question the safety of the financial sector. Investors institutions financial sector. US risk-capital Bottlenecks in the from the affected markets and movement emerged compromised while other sectors of the economy were during this episode was financial system for a of capital comprehensive list of extreme events in risk. developed economies during the 20th Second, the "risk" in these episodes of an extreme nature. is event that causes agents to re-evaluate tlreir Most episodes models of the world. In the are triggered by an unexpected Fall of 1998, the comovement of Russian government bond spreads, Brazilian spreads, and U.S. Treasury bond spreads was a surprise to even sophisticated market participants. These high correlations rendered standard market participants searching obsolete, leaving financial decisions using "worst-case" scenarios uncertainty. Our model more than in their markets. and theirs just risk. Agents contract with financial intermediaries to cover liquidity shocks that may These management risk decisions are treat this uncertainty as Knightian. grow concerned that shocks may in models, in a manner suggestive of Knightian "stress-testing" arise in other their ability to deliver Knightian uncertainty or a risk as well as the on fall in intermediaries' assets. own else receives shocks. will deplete the resources of intermediaries and Such riskiness rises either through an increase Thus our model combines the financial S3'stem flight to quality episodes. for safety. Second, we show that Knightian agents respond to markets by requiring financial intermediaries to lock-up some capital to markets' shocks, regardless of what happens in other markets. to dedicate capital to cover their markets different from aspects of flight to quality episodes. First, the increase in perceived and demand their uncertainty regarding other devote to their markets that in intermediaries have limited resources, agents If financial Knightian uncertainty that underlies riskiness generates conservatism model describing shocks their financial contracts. The model captures two important arise made with incomplete knowledge. Agents understand shocks, but are uncertain about the probability compromise Agents responded by making focuses on agents' perceptions regarding the ability of financial intermediaries to deliver on their financial contracts. own new models. management models Indeed, the extreme disengagement from risky activities that Greenspan highlights suggests that agents respond to their and for risk own By forcing intermediaries shocks, agents ensure that their shocks are covered regardless of However, once locked-up, intermediaries' capital is not free to move who across markets in response to shocks, resulting in bottlenecks and market segmentation. The central capital bottlenecks have bank action during macroeconomic consequences, and lend support to Greenspan's flight to quality. call for While each Knightian agent covers himself against an extreme shock, collectively these actions prevent intermediaries from moving capital across markets to expediently offset shocks as they arrive. We show that this inflexibility leaves the economy overexposed aggregate shocks that are manageable by the private sector in the absence of the scenario that the collective of conservative agents are guarding against so even given agents' incomplete knowledge. Collectively, agents make poor is to (moderate) flight to quality. impossible, and risk management Moreover, known to be decisions that lead to avoidable losses. In this context, a lender of last resort, even own market, can unlock if less informed than private agents about each agent's private capital and stabilize the economy during moderate shocks. It does so by committing to intervene during extreme events wliere the financial intermediaries' capital is depleted. Importanth', because these extreme events are highly unlikely, the expected cost of this intervention minimal. If credible, the pohcy derives its public intervention in the extreme event is power from a private sector matched bj' but because it each pledged dollar of a comparable private sector reaction to free flexible LTCM resources to deal with moderate shocks. In this sense, the effect, multiplier; is bailout was important not for its direct served as a signal of the Fed's readiness to intervene should conditions worsen, which unlocked private collateral. The model we develop and our Literature review. related to the Diamond and Dybvig Diamond and Dybvig model relation to is Our paper bank (1983) model of runs. An in greater depth after We will discuss the we introduce the model. Papers such as a gap in the literature on financial frictions in business cycle models. fills closely important difference relative to the that ours does not not involve a coordination failure. Diamond and Dybvig most prescriptions over the lender of last resort are Bernanke, Gertler, and Gilchrist (1998) and Kiyotaki and Moore (1997) highlight how financial frictions firms amplify aggregate shocks. Instead, uncertainty in response to shocks, and Our paper also studies the sense, our paper is closer to we emphasize how how in financial frictions lead to greater (Knightian) back into the financial accelerator. this rise in uncertainty feeds macroeconomic consequences of frictions in financial intermediaries. Holmstrom and and Krishnamurthy (2003) that emphasize that Tirole (1998) with complete financial contracts, the aggregate collateral of intermediaries is In this ultimately behind financial amplification mechanisms. In terms of the policy implications, Holmstrom and collateral limits private liquidity provision (see also government can The key issue difference limited. In our model, the is but and more There is in the aggregate centrally, it may be As in much for liquidity pro\'ision. and Woodford, Tirole, is that we show how inefficiently used, so that private sector liquidity government intervention not only adds to the private improves the use of private sector's collateral collateral. a growing economics literature that aims to formalize Knightian uncertainty (a partial contributions includes, Gilboa and Schmeidler (1989), Hansen and of aggregate Woodford, 1990). Their analysis suggests that a credible between our paper and those of Holmstrom and provision how a shortage government bonds which can then be used by the private sector even a large amount of collateral also, Tirole (1998) study Dow and Werlang Sai-gent (1995, 2003), Skiadas (2003), Epstein of this literature, Knightian uncertainty is we use a max-min device (1992), Epstein and and Schneider (2004), and Hansen, to describe agents expected utilit}^ most similar to Gilboa and Schmeidler, in that agents Wang et al. list of (1994), (2004)). Om- treatment choose a worst case of among a class of priors. Our paper applies These firms typically max-min expected stress-test their utility theory to agents who we interpret as running financial firms. models to various extreme scenarios before formulating investment The widespread policies. making Corporate liquidity management in practice. We mind. in use of Value-at-Risk as a decision is making also criterion is an example of robust decision done with a worst case scenario for cash-flows view the max-min preferences of the agents we study as descriptive of decision rules that overweight a worst-case rather than as stemming from a deeper psychological foundation. In paper we refer to agents as robust decision makers. making process of the much of the This terminology most closely corresponds to the decision financial institutions that concern us in this paper. Routledge and Zin (2004) also begin from the observation that financial institutions follow decision rules to protect against a worst case scenario. maker averse market sets bids and asks They develop a model of market liquidity to facilitate trade of an asset. Their in which an uncertainty model captures an important aspect of flight to quahty: uncertainty aversion can lead to a sudden widening of the bid-ask spread, causing agents to halt trading and reducing market liquidity. Both our paper and Routledge and Zin share the emphasis on financial intermediation and uncertainty aversion as central ingredients episodes. But each paper captures in flight to quality different aspects of flight to quality. Easley and O'Hara (2005) study a model where ambiguity averse traders focus on a worst case scenario when making an investment in a worst-case scenario Finally, in the home than Like us, Easley and O'Hara point out that government intervention decision. can have large effects. our model agents are only Knightian with respect to systemic events. Epstein (2001) explores bias in international portfolios in a related setup, where agents are local markets.^ As Epstein points out, this modelling also highhghts the difference aversion and aversion to Knightian uncertainty. interact with macroeconomic conditions We of risk aversion. to flight to quahty In Section 2 to quality. we in in the foreign between high risk Moreover, our modelling shows that max-min preferences ways that are not present show that when aggregate intermediary agents behave identically. However more uncertain about when aggregate collateral in is models with an invariant amount plentiful, Knightian and standard collateral falls, the actions of these agents differ, leading Knightian model. describe the environment, while in Section 3 we describe decisions, equilibrium Section 4 derives the value of a lender of last resort in our economy. interaction between aggregate collateral and robustness considerations. problems and other critiques associated with the lender of last resort. and flight Section 5 illustrates the Section 6 discusses moral hazard Section 7 concludes. The Model 2 We study a model conditional on entering a period of turmoil where Knightian uncertainty model is silent on what ^Epstein's model is triggers the episode. closely related to our model in In practice, we think that the occurrence Caballero and Krishnamurthy (2005). is high. Our of an unexpected LTCM event, such as the We focus concerns. or Enron crises, on the mechanisms that play out during the concerns affect prices and quantities? What is causes agents to re-evaluate their models and triggers robustness How do these robustness concerns interact with aggregate constraints? a variant of a bank-run model is The agents agents and intermediaries. (e.g. "eeurly" in Diamond and Dybvig, which case he needs some liquidity immediately. by a financial market We sudden need The agents sign financial contracts with the intermediaries to provide are "early." An Diamond and view the liquidity shocks as a parable specialist (e.g. a trading desk, hedge fund, market maker). them with liquidity insurance they if intermediary can be thought of as a bank which extends a credit line to the specialists, or as the top-level capital allocator in may in the canonical which case he has no special liquidity needs, or he "late" in for a for capital 1983). It has a set of competitive As are subject to liquidity shocks. Dybvig (1983) banking model, an agent may be We do agents' robustness the role of an outside liquidity provider such as the central bank? The model may be How liquidity episode: an investment bank. where the agents perceive that there are interested in situations not deliver the contracted liquidity insurance. is some chance that intermediaries bank run model, a central concern In a of agents is that they will arrive too late to claim liquidity from an intermediary, and possibly not receive the liquidity. We introduce this type of concern by assuming that early agents are further divided into those shocks first and those who receive shocks second. Each intermediary of liquidity/collateral, Z, which (or collateral) also Z, is it delivers to the agents. We is who receive assumed to have a limited amount normalize things so that in aggregate liquidity and no longer distinguish between one intermediary and the aggregate. SHOCKS AND PROBABILITIES Formally, there is U'^ u TZ+ : u'(0) = — * 7^ = { b)' tj u if late u{ci) if early and u(c2) if early and second £ fi = [0, 1]. Agent w receives utility: Probabilitj^ Probability: 0.(1) first Probability: twice continuously differentiable, increasing, strictly concave and satisfies the condition cx). Unlike the occur, there ()!)(1), is a continuum of agents indexed is Diamond and Dybvig model, we assume a first that the shocks are correlated across agents. wave, and then possibly a second wave. The first and the second wave takes place with conditional probability wave 0(2|1). If shocks of shocks occurs with probability We assume that 0(2|1) < 1, so that 1 with 0(2) = > 0(1) > 0(2) > (1) 0(2|1)(/)(1). This condition states that, in aggregate, a single-wave event is more likely than the two-wave event. We will refer to the two- wave event as many an extreme event, capturing an unlikely but severe liquidity crisis in which agents are affected. The shocks are aggregate in that they simultaneously affect a wave occurs, then one-half shock of the agents conditional on the first, first is hit mass by the shock. For agent wave of shocks occurring, u>, equal to is We of agents. assume that if the first the probability that he receives a While Xn) for conditional probability need not be one-half, the average conditional probability across any given w, the agents must be all equal to one-half: ^ n If the second wave occurs, the remaining one-half of the agents hit is by the shock: ^ n SECURITIES There a contracting time, date is We insurance to the agents. an agent s{uj) is shock • r(w) not We if offer financial contracts to provide liquidity shocks are observable and contractible. There during the first w or second wave. The claim a risky claim. This claim pays one to agent he receives its We define is no concern that two types of financial offer: a safe claim. This claim pays one to agent is is all where intermediaries pretend to have a shock and collect on an insurance claim. will contracts that intermediaries • assume 0, if he receives a shock regardless of whether the costs p{s,{jj). u) if he receives a shock during the first wave, but shock during the second wave. The claim costs p(r, w). refer to the r claim as risky In the example of a bank-run, it because it depends on the order of the agent reflects the feature where an agent may need in the sequence of shocks. liquidity but be second in line to claim the liquidity. These two securities that financial securities span the uncertainty depend on the shock in our economy. realizations over every subset of fi. In principle one could introduce But given the securities we have defined, such securities will not be traded.'^ INFORMATION Agents purchase financial claims from the intermediaries to insure against their liquidity shocks. making the insurance decisions, agents have a probability assume that agent-w knows for the indexation. We =<^J1)+^J2). example, we could imagine a security that pays one bj' of their liquidity shocks in mind. his probability of receiving a shock: 0, •'For model In if both ui and uj' receive a first shock. However, agent w has no use However, and centrally to our model, agent uncertain about the correlation between his ui is know aggregate conditions. In particular he does not own shock and among the relative likelihood of being the first wave or second wave. Agents treat the latter uncertainty as Knightian. We define, = -JW' ^-> Agent UI does not know the true probabilities 0tj(l) and d>^{2). unknown We use the notation 0J^(1), know Since agents agent-w's perception of the relevant true probability. agent-w's ^ etc., to denote a sufficient statistic for 4>^, 6'^ is Normalizing these probabilities by the aggregate shock probabilities probabilities. ' is convenient in the analysis. Agents consider a range of probability-models insurance portfolios that are robust to their model uncertainty. Maxmin Expected Utility representation of the max min K captures the extent of agents' J5o[t/'^|^"] problem and We wq is [1 impinge on their how activities. The}' may have the behavior of agents in other markets feared gridlock in the unsure whether payments money owed to sj'stem. , 1 (3) the initial endowment of agents. how aggregate conditions own markets, but are unsure of them. For example during 9/11, market participants affect arrive. In K], and design < Wq Each participant knew how much money he owed him would + write: a good understanding of their may /\ and Schmeidler's (1989) In a flight to quality event, such as the Fall of 1998 or 9/11, agents are unsure of will — follow Gilboa p(s,a;)s(w) -|-p(r, w)r-(w) s.t. uncertainty, while with support in the set 0, 9'^ to others but was our modeling, agents are certain about the probability of receiving a shock, but are uncertain about the probability that their shocks will occur early relative to others, or late relative to others. We view agents' max-min preferences in (3) as descriptive of their decision rules. worst-case scenario analysis in decision making by financial firms of such agents. is The widespread use of an example of the robustness preferences "* SYMMETRY To may simplify our analysis be different, the The symmetry this information We is 9ui wc assume that the agents for every agent is 0. While each agent's true 6,^ drawn from the same 9. applies in other dimensions as well: common are sj'mmetric at date 4>^, K, wq, and u{c) are the same for all ui. Moreover, knowledge. Eissume that the aggregate shock probabilities, cf){l) and (p{2), are known by all agents. We may think that agents observe the past behavior of the economy and form precise estimates of these aggregate "See also Routledge and Zin (2004) for a discussion of max-min behavior among financial specialists. probabilities. However the same past data does not reveal whether a given more ui is likely to be in the first wave or the second wave. INTERMEDIARIES Each intermediary Financial intermediaries have a limited capacity to deliver liquidity insurance. endowed with some goods to deliver to its liquid funds or as the collateral/capital of may have more goods and can much of these goods. For varying aggregate collateral Z claimholders. These goods can be thought of as the intermediarj^'s A an intermediary. thereby credibly of our analysis, sell more better capitalized/coUateralized intermediary liquidity insurance. In aggregate, there are we assume that Z is and safe claims sold The the intermediaries pay out ^^.^ units effect of R and S the aggregate amount Agents can observe total respectively. liabilities and collateral agents require that each intermediary has sufficient assets to cover the the world. liabilities in all states of Z and Moore, 1997). (as in Kiyotaki by intermediaries, assets for each intermediary. we study the constant. In Section 5 Agents recognize that intermediaries have limited resources. Denote by of risky If If the wave occurs, one-half of the agents first will wave occurs, the remaining one-half the second intermediaries pay out ^. Then, the collateral constraint for the intermediaries is be affected and affected and the is: § + S<Z A representative less intermediary's objective is U' = of <p{l), 0(2), ci It maximize the date first and ex-ante symmetry of agents are at the rate P>0 (5) a standard expected utihty maximizer, is sufficient to in selling insurance. compute the expectation Knowledge in equation (5). (1983) important distinction between our model and the Diamond and Dybvig (1983) model of bank runs constraint creates a possible coordination failure. runs and policy intervention. to write ^ and second shock agents, discounted The intermediary our assumption of complete state-contingent markets. agents revenue from the sale of the insurance does not need to know the true 9^ of each agent RELATION TO DIAMOND AND DYBVIG An (4) + Eo[P{ci+4)], collateral constraint in (4). with linear preferences. is to a cost of the insurance resources disbursed to both and subject to the is is Since In Diamond and Dybvig The coordination Our model does not impose a the sequential service failure informs the discussion of bank sequential service constraint. Agents are able complete state-contingent contracts on the order of the shocks. The amount of early liquidation by predetermined by date it is common knowledge the shock in the continuum of first clients, wave, it R/2 faces contracts rather than by coordination failure that agents are symmetric, agents know tiiat r(a)) = among agents. R. Since one-half of the agents receive of risky claims are settled by the intermediaries. Also note that since each intermediary has a no uncertainty on the quantities to be delivered in each aggregate state. Diamond and Dybvig, In the bank run triggers unanticipated withdrawals and default by the bank. may In our model, agents recognize that intermediaries default because they have limited the collateral constraint of equation (4). However, in our complete markets This equilibrium. intermediary's is Z and impose model default never occurs in because, with complete markets, agents anticipate the state of the world where the liabilities lead it and rewrite contracts so as to reduce the to default, point where no default occurs. Thus default is liabilities central in the calculations of agents but, and exactly to the as a result, does not occur in equilibrium.^ Knightian Uncertainty and Flight to Quality 3 In this section we describe agents' decisions and equilibrium, and connect robustness concerns with flight- to-quality episodes. The 3.1 cost of locking collateral The problem and of a financial intermediary q{r,Lj), in order to is maximize revenue to sell r(aj) risky claims less / [{p{r,uj) safe claims, at prices p(r, w) the actuarial cost of providing liquidity insurance, subject to the collateral constraint. Following the objective in max and ^(w) (5), - Pcfjl))^^) + intermediaries solve: {pis,io) - Picf^Jl) + ^J2)))s{uj)]d^. Because agents are ex-ante symmetric, their decisions are identical and intermediaries same R prices to and S all it affected. Likewise, intermediary objective is is if common knowledge that if in the decisions and price functions and the let aggregate shock hits, one-half of the agents are it is affected. Since the has sold, we can rewrite as: ( practice, liquidity shocks trigger p(r) - /? p(.) - ^^^i^-M /3^ '^)-(' dynamic problem objective is -f ) dynamic trading features from our framework through a complete ( in asset and not ) 5. martcets - for example, by shedding risky assets and all ex-ante. While interesting, Arrow-Debreu markets assumption which, we moving deliberately removed these as usual, allows us to state a as a static one. to provide an endogenous explanation for why an exogenous segmentation/incomplete markets assumption. isolates the first only concerned about the aggregate paj'out on the financial claims to Treasury Bills - so that agents' responses are ex-post Our u> the second aggregate shock hits, the other one-half of the agents rnax ''In we omit the argument insurance at the denote the aggregated claims sold to agents. Recall that its agents. For brevity, offer mechanism we highlight. As capital does not The assumption move across markets, as opposed to making of complete liquidity-shock-contingent claims a matter of interpretation, the ex-ante insurance contracts can be thought of as collateralized contingent credit-lines. 10 The intermediary must satisfy the collateral constraint in (4): Because of the linearity of the objective function, to the intermediary's problem yields that R and S are P{'r)=P^For example, when If /? = if the collateral constraint does not bind the solution an at and p(s) = interior only /3 if, . prices are actuarially fair at an interior solution. 1, the collateral constraint is binding, then: .M^/sf^ + f a„d.W.^*l+ie)+,. where /i > is the Lagrange multiplier on the collateral constraint. A ,6) positive multiplier in the cost of insurance reflects that the intermediary has limited resources and hence the opportunity cost of selling one type of insurance claims is selling less of the other. Safe claims use up twice the risky claim, given our assumption that one-half of the agents are affected in a sells one less safe claim, then it can credibly sell factor twice the cost of the collateral constraint We are most interested writing liquidity insurance. drop the in the when To simphfy some is The logic leads the intermediary to we assume for now that /? = and thereby (6) to yield: = ^. (7) the main result from modelhng the supply side of the economy. equilibrium cost of "locking" collateral. In Section 5 3.2 shock. If the intermediary selling safe claims over risky claims.^ of our expressions, pM This pricing expression of collateral as a opportunity cost of using collateral, as opposed to the actuarial cost of term on the right-hand-side of first two more risky claims. This first amount we revisit the case of /3 > It captures the 0. agent's decision problem Facing prices p{r) and p(s) each agent solves the decision problem in max min cPZC^Mci) + Ci^Mco) + r,s (l (3): - 4,Ju (8) fl";ee ^Note that to if the first and second waves were mutually exclusive events, then the intermediary could use the same back liquidity insurance in each event (i.e., it would separate the first and second wave collateral constraints). collateral In this case the tradeoff of collateral would only be one for one between risky and safe claims, and the only binding constraint would be that for claims sold for agents hit on the first wave. In contrast, mutually exclusive. 11 in our models waves take place sequentially and hence are not where, The off = r j.'-n^ "^-(^^ = 0(1) 0(1) + 0(2) "^0(1)+O(2)' expressions for cj and C2 only in the first first + C2-S, s, come from our let = 0„-0Jl). '^w(2) earlier definitions of risky first and safe claims: the risky claim pays and second shocks. us consider the problem in (8) for a given value of 6*^ (i.e we drop the min operator). order conditions are, cf>^{l)u'{c^) where +p(r)r <wo, p{s)s shock, while the safe claim paj's off in both For the moment, The ci tp is + (PZilWici) = Mr) cj>^{2)u'{c2) = Ms) the Lagrange multiplier on the agent's budget constraint. Dividing the second equation by the first equation yields: C(2)u^^p(£2 + 0-(l)V(ci) From the definition of 9'^ in (2), we rewrite this expression as, 0(2) _ ^.'(C2) Xl)u'(ci) The right hand right hand side is An affects the agent's decisions agent who considers a higher value of fully collateralized to set aside more by the intermediary, We now when he does is alternative values of 6'^. first analysis of the intermediary's problem, C ^ in 9'^ chooses higher cn and lower safe claims and demanding more likely to first Ci. less risky claims. ' or second. He can Since all achieve this claims have to safe claims, the agent also forces the bank happen second. In terms of the bank-run example we be last in line at the bank, requires the bank to set aside arrive at the bank. turn to the robustness step. theory, the agent 0(2) collateral to cover shocks that will have used, an agent who thinks that he resources for From our by altering the perceived odds of being consumption pattern by holding relatively more be p(r) equal to one. Thus u'(ci) is, 6"^ p(.) side of this expression are equilibrium prices. we know that the That ^''> p(r) The agent makes his insurance choices while being robust to In the game-theoretic language often used to describe chooses r and s, then "nature" chooses given his choices. 12 6*^ € max-min expected utility to minimize the utility of agent A, If is, > u{ci) when the agent his true 9'^ 9'^ u{c2), = is u{ci) The agent makes that ^" is (8), we see that nature will set more insured against being equal to is — K. At 1 from = 1 + On K. u(c2), the expected utihty Ci > C2 so that u{c2), nature will for the agent behavior by nature. First, suppose that nature will choose 9'^ = I is that do the opposite and from the agent's decisions are independent of his decisions anticipating this near one. In this case < u(ci) if That as high as possible. than second, the worst-case hit first the other hand, and thus 6", (l)'^{2), + K. The K set 9'^. small so is agent chooses cj and Co to satisfy: U'(C2) ^ 0(1) li'(ci) As K rises, 1 the right-hand side of this expression + 1 0(2) A' and ci/co falls 0(2) the right-hand side is equal to one, and = ci C2. We ci < C2, if refer to this situation as the fully robust case since = then nature will set 0^ 2p(r) = 2p, = 1 — K. economy rises. < K<K= ^Lt — 1, The ,^ - if the agent chooses refer to this as the "partially robust" case. For K > K, if _ w'(ci) We agents' decisions are as ci = = C2. price, and decrease an equilibrium in the robust economy: constitute U'{C2) if = 9" 0(1) 0(2) 1 + K: 1 1 + ^ K ^ K = K: U\C2) ^ u'ici) 3.3 because Agents' demands are infinite at p agents' decisions are as 0(2) We is following proposition summarizes these results: 0(1) • This Given the supply determined relative unique. is effects. Proposition 1 The following insurance decisions For 0(^(2). Anticipating this action, the agent will prefer to choose a change in p only has income monotonically toward zero as p • and K rises past K, the agent will not change his decision. Finally, note that equilibrium in this p{s) toward one. When, ' agents' decisions are robust to their uncertainty over ^^^(l) Notice that falls 0(1) 1 <t>{2)l + K = 1 refer to this as the "fully robust" case. Flight to quality In the partially robust solution, every agent chooses choose r = ci — C2 and s = C2. As K rises, ci > C2. To achieve this consumption pattern, agents so that the uncertainty for agents rises, ci 13 and co converge to each other. Agents lower r and increase The pattern At the s. fully robust solution, agents set ci of shedding risky claims in favor of safe claims is The facilities.^ The concern of each agent bank's limited credit from when that his shock is facility will ^^'ill = 0. K generates a "run" on lines. occur after the average agents' shocks, at which point the By have been depleted. ° prearranged at date is demands towards can also interpret the By decision. own agents' choosing r "committed capital" As ) = effect of > 0, s , in the a rise in setting r = 0,s > 0, agents insulate themselves by an agent preemptively drawing down a when r falls of capital where think of these actions as agents shifting unlikely that the intermediary's resources it is in flight to quality episodes. K on equilibrium in terms of intermediaries' capital allocation and there is is free is less ( "trading capital" ) to allocate to the first is locked-up agents who trading capital to allocate flexibly. In the fully robust locked-up and there latter interpretation captures a feature shared movement we can Alternatively, are complete in our model, the future these shocks occur, s of the capital of an intermediary and r of the capital 6'^ rises, As markets credit line agents force intermediaries to reserve some resources to cover the 0, case, all of the intermediary's capital The 0. common shift is shocks, regardless of receive shocks. reflected well capitalized intermediaries, be depicted. This portfolio We may be the fear of other agents' shocks arises. action by each agent ( and r s this concern. at a date will = agents recognize that banks have limited resources to back up their credit In practice, the robustness action their C2 central to flight to quality episodes. Interpreting the financial claims of agents as credit lines from banks, the rise in banks' credit = is no capital mobility. by most Bottlenecks arise flight to quality episodes. and markets appear segmented. For example, markets where hedge funds specialized were particularly affected by the in the Fall of crisis, 1998 episode the as (abundant) capital did not flow into these markets. For a fixed case in more K , an alternative way to generate flight to quality in detail in Section 5, but the intuition risk as agents is clear enough. our model A fall in Z terms of financial intermediary 'Although the actions of agents level, agents' to reduce Z. in We explain this increases flnancial intermediary grow concerned that intermediaries maj' not have enough resources Agents worry about being second and raise 6^. Some episodes, such as the in is to cover their shocks. Fall of 1998, can be thought of risk. our model are most naturally interpreted as a run on banks' credit actions are also related to the more commonly analyzed run on banks' deposits. facility, at a deeper As Diamond and Dybvig (1983) emphasize, a deposit contract implements an optimal shock-contingent allocation of liquidity. In the fully robust equilibrium of our model, agents choose an allocation that their own allocations 'We is non-contingent on each other's shocks. Agents each hoard liquidity to cover shocks, independent of other markets' shocks. In this sense, robust agents' preference for shock-independent liquidity is related to the behavior of panicked depositors in a can also interpret robustness of an investment bank. Each u> in terms of the internal lobbies for more risk capital, risk bank run. management of an investment bank. because of fear that other agents the bank's total risk capital. 14 Each tj is a trading desk will suffer losses soon, reducing Collective Bias 4 and the Lender In the fully robust equilibrium of Proposition 1 of Last Resort agents insure equally against and second shocks. To first arrive at the equal insurance solution, robust agents evaluate their first order conditions (equation 9) at conservative probabihties: C(l) = C(2). Suppose we compute the probabihty of one and two aggregate shocks using agents' conservative probabilities. Then, ^(1) = 2 /C(l)'i^in where the two in this expression reflects the fact that only half of agents are affected by the first aggregate shock. Likewise, 0(2) = 2 / C(2)f^. Together, these computations imply that agents' conservative probabilities are such that, 0(1) Implicit in the conservative probabilities, the aggregate shocks. But we know (and all = 0(2). economy is perceived as equally likely to receive one or two agents know) that actually 0(1) > 0(2), which implies that agents' conservative probabilities are collectively biased. Since each agent is concerned about the scenario in which he receives a shock last and the intermediary's resources have been depleted, robustness considerations lead each agent to bias upwards the probability of receiving a shock later than the average agent. But collectively, every agent cannot be later than the "average." These points carry over to the partially robust equilibrium. 1 + min[A', A'], we K > 0, robust agents set B'^ = implying that, 0-(2) If For any = (l + min[A^A'])^0::;(l). take the average of the conservative probabilities on the left and (11) right hand side of equation (11), we find. When A" > the last inequality is strict, showing that agents conservative probabilities are collectively biased. Note that each agent's conservative probabilities are individually over 6uj, it is possible that agent from the aggregate does it ui plausible. Given the range of uncertainty has a higher than average probability of being second. Only becomes apparent that agents' 15 when viewed collective probabilities are implausible. These observations motivate us to study how a central bank, which is interested in maximizing the collective, can improve on outcomes. Central bank objective and information 4.1 The central average, also bank maximizes the average we assume that the central utility that bank uses its own assume that the central bank equally weights the because agents are observationally identical at date When computing this probability assessments rather than the agents'. We agents derive from their choices. utility of every agent. The The former assumption 0. is latter more In assumption follows delicate. agents have non-Savage preferences, at least two approaches seem defensible: The central bank objective uses both the agents' preferences and the agents' conservative probability assessments; Or, the central bank takes a more paternalistic approach in from consumption, but not evaluated at the agents' We which it is concerned with agents' ex-post average utility collectively biased conservative probability assessments. follow the latter path but conclude the section by showing that both approaches have similar implications for the welfare gains associated to the presence of a lender of last resort. As noted earlier in the paper, decision rules of financial specialists we think (e.g., they reproduce the behavior of agents in uncertainty. Prom function that Thus, let this perspective, may it is of the robustness preferences as a realistic depiction of the worst-case scenario analysis). These preferences are useful because flight to quality situations not obvious that a central bank should build biases into lead to obvious average losses, just because agents exhibit these biases. us consider the foUowing central V= ci,wiC2,w are the where they are faced with Knightian J bank objective objective ^° function:^' + 0^^(2)«(C2,„)] [t'^'^ilMc:,^) its doj. consumption resulting from agents' insurance decisions, while bank's assessments of the probabilities of the relevant events. We 4>^ (1), etc., are the central note again that the central bank uses agents' utility function of u{c) to evaluate their consumption decisions, but discounts u{c) using its own probability assessments, rather than the agents' conservative probability assessments. Importantly, we do not assume that the central bank knows the true biases stem from a lack of knowledge of central bank how O^^s of agents. aggregates impinge on their activities, lacks knowledge over 0^. In fact, for all of our results in this section we Since the agents' also assume that the we can assume that the '"Sims (2001) has made a related point in questioning the application of robust control to central bank policymaking. He argues that max-min preferences are simply shortcuts to generate observed behavior of economic agents, but should not be seen as deeper preferences. robust control for its ''We have omitted Note, however, that the collective bias we identify would be solved by a central bank, even decisions. the constant term involving u in this objective. 16 if it uses bank knows central strictly less than the agents; that the i.e. K of the central bank is larger than the K of agents. The central bank, like the agents, affected knows both aggregate and that one-half of the agents is by each aggregate shock: U^^\l)±o = t^ U^J{2)d^ and, ^ Jn We may probabilities Jn imagine that the central bank and individual agents are able to form precise estimates over the aggregate behavior of shocks from observing past data. 4.2 Collective risk management and wasted collatercd Consider a central bank that alters agents' decisions by increasing decreasing collateral, C2,u> by the same amount. This can be implemented, by requiring agents to hold two more units of r and one Ci,„ by an infinitesimal amount, and amount of intermediaries' The value of the reallocation for a given less unit of s. is: V^'' = ^ [0S^(l)«'(ci,J In the fully robust equilibrium, the first - 0^^(2)«'(c2,j] order condition for agents dio. is, W'(Cl,a;) The budget constraint for every agent that, is p{r)r Since the central bank knows that conclude that Ci^^ = ci for all w. all wq. agents face the same prices and have the same wealth Then, we can v^'' + p{s)s = = «'(co/j€^(i)-€^(2)]rfw = ^«'(ci)(</.(l)-0(2)) > 0. u'jco,,^) ^ 4>{1) Ci for all uj. Then is, 1 ^ Again, given that agents face the same prices and have the same wealth = can write, Similarly, in the partially robust case, the first order condition for agents that Ci^u level, it substituting the agents' first 17 level, the central bank can conclude order condition into the central bank's objective, we find, y«^ '(1) = = We summarize these Proposition 2 For any 1 1 2 2 1 + K > 0, is Thus by Z). many The tie up as much about individual B^s. In important point The somewhat at some ^" If may is much is where the more likely away from collateral as safe claims do toward more risky claims and central this sense, the central shock (the important when aggregate collateral (i.e. the collateral constraint less safe claims, may is is intermediaries never occur. bank reaches these conclusions requiring only knowledge that the central Lender of last resort maximizes the expected reallocating agent insurance of their resources to cover a second-shock that As we have remarked, the probabilities. differ central bank that safe claims in a scenairio portfolio of insurance shifting agents' decisions Finally, note that the central may A insurance valuable to the central bank because from the central bank's perspective agents are is insurable with risky claims. do not have to 4.3 first. economy can improve outcomes by limited because risk}' claims lock only half as R/2+S < much and toward risky claims. reallocation one) A' agent decisions are collectively biased. Agents choose too wasting aggregate collateral by using too first A' results in the following proposition: (ex-post) utility of agents in the The dw + 0. against receiving shocks second relative to receiving shocks safe claims 1 1 <^(iy(ci) > 0(2) <j){l) n 1 <Pt''m 0(i)«'(ci) of aggregate bank may be much more confused than individual agents bank may be the bank does not suffer from least informed agent of the economy. The collective bias.^^ last resort abstract reallocation experiment discussed above highlights how central bank policy choices from robust agents' policy choices. More concretely, we now focus on the value of a lender of (LLR) which cost, the central it (i.e. in assume that the central bank obtains resources ex-post, can credibly pledge to agents bank were imagine a situation liquidity crises We in the robust equilibrium. to be uncertain about the values oi <t>{l) and which the central bank is However, two-shock event. ^'^ In practice, this pledge <^(2), then we could overturn our uncertain about these probabilities, and the incidence of both shocks occurring). •'overinsurance" bias of agents. in the this "bias" is its In this case, the biased central of a different nature than the one may result. In particular, we objective function overweights bank will also we emphasize be subject to the as it would not be collectively inconsistent with conditional probabilities. ^^As in at least, it collateral. Woodford (1990) and Holmstrom and Tirole (1998), the LLR has access to collateral that intermediaries do not (or has access at a lower cost). Woodford and Holmstrom and Tirole focus on the direct value of intervening using this Our analysis focuses on the gains beyond the direct value of the intervention. 18 . be supported by costly ex-post and carried out by guaranteeing, against inflation or taxation liabilities of financial intermediairies who have We sold financial claims to both markets. default, the analyze the impact and marginal benefit of such a guarantee. Formally, the central bank credibly expands the collateral of the financial sector in the two-shock event by an amount Z'^ Since we ai^e . Thus, the collateral constraints on intermediaries are altered to computing the marginal benefit interested in of intervention, we study an infinitesimal inter- vention of Z'^ If in the central bank offers more insurance against the two-shock event, this insurance has a direct benefit terms of reducing the disutility of an adverse outcome. The direct benefit of the where the 2 in front reflects the fact that consumption for the group (of an extra unit of collateral measure one half) that is for the LLR is, second shock yields two units of by the second wave. hit However, the anticipation of the central bank's second-shock insurance leads agents to insurance decisions. reoptiiTiize their Agents reduce insurance against the second-shock (reduce safe claims) and increase their first-shock claims (increase risky claims). The total benefit of the intervention includes both the direct benefit as well as any benefit from portfolio reoptimization: duj. Since Ci,„ + ct.w = {R + S) +S= 2{Z + Z'^), the reoptimization of each agents' portfolios must sum up to twice the increase in total resources: As we have shown, a reoptimization dci^u dc2,u dZ^ dZ(^ of agents' portfolio _ ~ r, "' away from safe claims and towards risky claims has a first-order benefit in the robust equilibrium. The following proposition confirms this additional value in the context of the lender of last resort; Proposition 3 For any K > 0, the total value of the lender of last resort policy exceeds its direct value: total .^ jrdirect V za > v^Q Proof. We rewrite the total value of intervention as. V'S"' = /^ [C^(l)«'(ci,.)^ = 0(2)u'(c2)+/ i0^^(l)u'(ci,„)-0^^(2)u'(c2.^)|^cL;. 19 + </>5^(2)k'(c2,J (2 dZ° duj We can now substitute using the first order condition for agents' decision to arrive Tlie last step follows since agents substitute some at, of the second-shock insurance of the LLR toward purchasing first-shock insurance. Welfare equivalence and paternalism 4.4 We conclude this section by revisiting the non-paternalistic case, where we compute welfare using agents' subjective and conservative probabilities. beyond its objective direct value. There but the direct benefit rises since the is It turns out that there is still an important value to the LLR no indirect benefit from reallocation (by the envelope theorem), agents collectively exaggerate the likelihood of the extreme two-shocks event Jn That is, the LLR = ^(2)u'(C2) > 0(2)li'(c2) has an extra benefit which comes from reducing the "anxiety" of robust agents. Moreover, in the fully robust case we note that to imply: ,,airect,int >di7-ect,inf ^zo _ - 0(1) +(j>{2) 'VK^J 'i^\^l „,li„ ^,^ ' W 7) \ ICoj __ -Trtotal - l'2G 2 This means that in the fully robust case the value of anxiety reduction (from the agents' perspective) to the value of insurance reallocation (from the central bank's perspective). is equal While the exact equivalence result carries over to the partially robust case only for the case of log-preferences, case there is even in the partially robust an extra benefit due to anxiety reduction. Comparing the paternalistic and non-paternalistic perspectives, we note that agents value the intervention more because they exaggerate the probability of the event that is being subsidized, while the central bank values the intervention because the subsidy reduces private collateral waste. the econom}' were to be drawn infinite times, all agents the direct ex-post benefit of intervention. The agent The latter effect would receive higher average ex-post means that utility if beyond perceives this additional benefit ex-ante, although for the "wrong" reasons. 20 Price of Liquidity and Collateral Risk 5 Up now we have analyzed to of liquidity insurance (i.e., /? a model where intermediaries factor zero actuarial cost = For this case, the intermediaries' 0). setting the prices order condition pins a particularly straightforward algebraic analysis, fixing relative prices allows for down the be constant and independent of the agents' portfolio decisions. relative price of safe to riskless claims to Although first when it hides some of the mechanisms underlying the model. When P > agents' robustness concerns have important price effects. In the 0, we elaborate on these price effects to further clarify our model. In the second part The considerations exacerbate the costs of collateral reductions. latter point is first part of this section, we show that robustness the main substantive result in this section. As in Kiyotaki and Moore (1997), reductions financial accelerator. Our model lateral lead to a fiight-to-quality We the set first 5.1 /? = 1, in collateral (or collateral values in their illustrates a further channel for amplification. which locks already scarce now so that intermediaries Reductions in aggregate and amplifies the collateral model) trigger a col- financial accelerator. factor in the actuarial cost of disbursing insurance resources in and second shocks. Abundant collatereJ Z Starting with a benchmark, suppose that financial sector is enough so large is that, at equilibrium levels of demand, the unconstrained in selling insurance. In this case the prices of the two financial claims just reflect probabihties: Pi-r) = -^ and p{s) = . Facing these prices, robust agents make their insurance decisions to Given p{r) and p{s) we can rewrite the At K= 0, agents choose ci = C2. cP{2)u'{c2) _p{s) -0(l)u'(cx) p{r) first It is order condition for agents easy to verify that ior This choice remains optimal because by choosing probabilities, When concerns. and the robust decision there is rule still abundant intermediary Agents know that intermediaries satisfy: ci = C2, K > as. agents continue to choose agents' decisions are robust to the ci = C2. unknown does best. collateral, the equilibrium will is unaffected by agents' robustness not default on their insurance claims. 21 Thus, they are not concerned about their shock being or second relative to other agents. Their uncertainty about the sliock first ^^ ordering becomes irrelevant. Using the budget constraint for agents, we can calculate that facing the actuarially fair prices p{r) and p{s), agents will choose, = r s 0; = = Ci = C2 In order for intermediaries to not be collateral constrained in equilibrium,'^ Price of liquidity and aggregate shortage 5.2 Z Suppose next that falls below Then, at actuarially Z_. fair prices, agents' demand saturates the collateral constraint of the intermediaries and prices rise to reflect the collateral Umitation: where > fi We Q is M 4>{l) ^ , J , 0(1) X + 0(2) the Lagrange multiplier on the intermediaries' collateral constraint. can compute these prices explicitly using the budget constraint and the intermediary collateral constraint {^ Proposition 4 When Z< = Pir) Z_, +s= the prices of risky 5+ (0(1) for the agent {p(r)r and safe claims as a function of Ci - 0(2))|| and p{s) = ^- (0(1) - fall in Ci and equal rise in co causes the prices of both risky and safe claims 2. A fall in Ci and equal rise in c^ causes the relative price ratio, ^W — 3. The economy has an aggregate liquidity shortage. thereby decreasing claims (point cj 1), as '"The irrelevance previous section, and increasing Ct. when and \, to rise. to rise. agents' uncertaintj' rises they purchase These actions lead co are: 0(2))^^- A in the wq) Z): 1. As we have shown + p[s)s = more to a rise in the equilibrium price of safe claims, both financial well as to a rise in the relative price of safe claims over risky claims (point 2). result in the case of abundant collateral also relates to discussions in the asset pricing literature as to when the actions of "rational" agents will eliminate any effects stemming from the biases of "irrational" agents. In our model, the intermediaries are Bayesian and risk-neutral, but are subject to a constraint whereby they have to hold collateral to back-up any short-sales of financial and prices securities. When there is sufficient reflect the rational intermediaries' valuations. It is intermediary collateral, the short-sale constraint does not bind a feature of our model that agents' insurance decisions are not collectively biased at the actuarially fair prices. '^In the insurance /3 = case we have studied earlier Z is infinite because intermediaries are always at a corner solution sales. 22 in their Point 3, regarding the aggregate liquidity shortage, A shocks. rise in K who also are subject to first and second shocks and wish to insure against these increases the price of liquidity insurance for Savage agents. In this sense, our model flight to quality in clarify this Suppose we introduce a small number of Savage (Bayesian) point, consider the following thought experiment. agents into the economy an important aspect of our model. To is is all agents in the economy, both robust and not just a problem affecting the decision rules of robust agents. It has a larger cost because the actions of robust agents raise insurance prices and thereby distort the insurance of all agents in the economy. The mechanism behind ularly transparent have shown if earlier, the increase in prices we examine the is the collateral lock-up The risky claim. is indirect. This demand free to R/2 (i.e. The back risky claims. That -H 5 < is, Agents increase their demand since safe claims lock up twice the amount of amount Z is used inefficiently, leading to a smaller effective Z we Thus causing for safe claims, of collateral that collateral as risky claims Z), as robustness concerns rise, agents force the intermediaries to lock intermediaries' limited partic- for risky claims as robustness rises. intermediaries to lock-up collateral to cover their shocks and thereby decreasing the is becomes effect risky claim price rises as robustness rises; but, as robust agents actually decrease their the increase in price of the risky claim eiTect. of the up more collateral. economy and higher insurance prices. We aries can think of claim prices as the marginal cost of liquidity provision have a limited amount of collateral to back economy the capacity of the liquiditj' provision, at the aggregate level, Our model shows to provide liquidity to agents. economy. Since intermedi- in the Z parameterizes that the actions of robust agents decreases the effective amount of liquidity provided to the entire economy. Reductions liquidity in aggregate liquidity provision premia are two central ingredients market structure - would also be 5.3 We e.g. if each w was next study how a fall in in most descriptions of flight to quality events. rise in financial With a richer a separate asset market - the rise in the cost of liquidity provision reflected as higher liquidity Collateral reduction by financial intermediaries and a corresponding and premia flight in the asset markets where the agents are active. to quality Z, for fixed K, affects the equilibrium of the economy. described a flight to quality event in terms of increasing K. We now Thus far we have show that our model generates similar implications from a decrease in Z. The first order condition for an agent's decision 0(2) u'{c2) '^</.(l)u'(ci) ^ is: 2wa21.0 + p{s) p(r) where the second equality follows from Proposition 4. 23 (0(1) (0(1) - 0(2))ci ^' 0(2))C2 ^ Additionally, from the collateral constraint of inter- know mediaries we that 1 + 1=^. Equations (12) and (13) in examining how c\ jointlj' define and the equiUbrium values of ci and c^ as a function of Z. Z change as cn falls below and study the function a{Z). As we have remarked Proposition 5 Suppose comparing Z is, Z as = earlier, for Z> Moreover, suppose that 0. = 2a{Z)Z Let us wTite ct Z_. economies with different equilibria across That falls. K that (13) levels of Z Z_, ci is — c ",.'^? and We are interested c\ = 2(1 — equal to co so that a(Z) is weakly decreasing in the (unique) equilibrium, value of , q:(Z))Z, = Then c. a |. falls as the ratio of insurance devoted to the first shock over that to the second shock falls, rises. When Proof. K equals zero, - F(^"' 7\ = 6'^ From equation 1. ^~^(l)w'(2(l-a)Z) The first order condition in (12) sign{^) equals the negative is satisfied of sign ( - 1 0(2) u'{co) ( Z <p{l)u'{ci)\ when F[a[Z), Z) = da This term is + u"{c2) T C2—7 - (0(l)-0(2))(l-a) Uo-(0(l)-0(2))(l-a)Z on o^ 0(2)^'(C2) (U"{C2) = This term is is negative. Therefore sign{j^) negative for The falls, (0(1) K= u'(ci) to the first shock first We + (0(1)-0(2))qZ ; also note that the equilibrium is unique since ^ is more likely we have derived in the previous sections. As reflects the limited availability of Z intermediary then the second shock, the central bank allocates more resources than to the second shock. affected by Z. For Z> agents act as this will leave positive. and second shock insurance Consider next the robust economy where If u-o >v case corresponds to the central bank's solution the allocation between below Z. ^;o-(0(l)-0(2))(l-a)Z \ / all a. resources. Since the first shock is - 0(2))a \ - 0(2))aZy' (0(1) u"{c^) - 0(2))qZ is + ^ (0(1)-0(2))Z + (0(1) t^o u{c). 0(1) u'(ci) \u'{c2) Wo find that u'(ci) - 0(2))qZ da we u"(ci) positive given the regularity condition 9^ Implicitly differentiating F, Ci u'{c2) (0(1) 0. ' J 7i;o-(0(l)-0(2))(l-Q)Z / u;o define, -g- \ OF dZ we ^° ~ ('^(^^ ~ '^^^^^(^ ~ "^^ wo + (0(1) - 0(2))aZ ^'(^^'^) '^(^^ (12) agents set Z^, if S^ them exposed = 1, they ci K> 0. equal to will We begin with a heuristic description of how equilibrium co. Starting from this point, suppose that decrease C2 and increase to the possibility that 9^^ is 24 high. ci slightly (see Taking Z Proposition falls slightly 5). However, this into account, agents will consider when making 6'" a larger decisions. For Z close to As Z falls sufficiently the fully robust insurance decisions. a small increase in 0" Z_, is choose sufficient for agents to below Z, 6" reaches maximum \+K, its with c\ larger than C2. In this case, agent decisions are partially robust. Proposition 6 Suppose — c l^,,7 that weakly decreasing in is Z across economies with different levels of agents make equilibrium decisions as we have , Then comparing c. (the unique) equilibria For a given value of Z, robust the following results. if, = eZ 1 + min{K, K(Z - Z)), whCTG — Z) Since K{Z_ • For Z such that 0" • For Z small enough that We Proof. increases as = a = An for \. 1 choose falls. fully robust. agent decisions are partially robust. , 61;::; = 1 + A' if a < = i; 61;;;; economy - 1 is /"ST if - u^p q > a a value i; - (0(1) + w;o ' equilibrium in the robust and + K) =0. d to F{a, Z;9'^ = — is 1 For a value of a value of Z > \ g'^' •^' First, and ^(2))(1 - a)Z (0(l)-</.(2))aZ ' indifferent over is e 61^ = that satisfies F[a, Z; 0") is we note that the proof of Proposition negative for any 0" uniquely determined since — K) = Z satisfies for 0. Since F{a, Z\ which d < , a, the equilibrium is 0, - [1 where + K] A', 1 if optimal is 0'^, first sets 6*" order conditions |. F{a = ^, Z= Nature is Nature sets at d. order conditions Nature (i.e. as the solution to we define we have that d > increasing in 0", is d a. Similarly, all = — 5 directly generalizes so that Next, we define 0. negative for first at a. = at is 9'^) is a is > ^, the equilibrium the equilibrium market clearing and agents' > ^ market clearing and agents' which Q > ^ for satisfies Oi = 1+K Z nature at that a. positive of 9'^ 4>{2))Z {4>il)-4>i2))Z- K{Z_ — Z), agent decisions are + - increases (weakly) as 0(2) v^^^-flc. "^" "4'{l)u'{2[l-a)Z) ' We proceed as follows. is falls, 6'^ u'{2aZ) ^ will Z (.^(1) define, F( Nature 1 v^ ^- I • _ 0(1) S^'o ^- ^{2)2wo + ry(^ . (i.e. — 1 F() = A' 0). 0'^ F(') if = = a > K = as the solution and , Finall}', for indifferent over the choice of 9'^ \i a < \, and this Likewise, for a value 0). ^ F[a,Z]9'^ '^ a. + 1 a g'^' at this value of a value of and Z for a which sets 9'^ such that F{a^\,Z-9"J=Q. We know in a, that at Z = Z, we must have that d > 5 at Z= Z_;9'^ = 1) = 0. Since F is Z. Moreover since decreases in 25 increasing in Z 6*" and F is decreasing decrease the function F(-), d must as fall Z below \ The Two falls. . conclusions follow: The statements d'^ of an agent is It is not possible that a function of both The agents directly increaes &^. this sense, our 5.4 Risky K and d'^ Z. in agents' response to a model generates collateral risk, the risk in for Z sufficiently low a must fall falls, fall in their Z is robust agents grow increasingly concerned own similar to from either a flight to quality rise in how K K Likewise as decisions. rises, agents respond to a or a fall in the rise in Z. and policy collateral (as in Kiyotaki Z As making In practice, the collateral of financial intermediaries economy and of the proposition follow immediately. about receiving a shock second and increase K. In a>\\ risky is and Moore, 1997). Although comparative static with respect to it Z is and varies with the underlying state of the beyond the scope of this paper to endogenize provides some insight into the effects of collateral our economy. Starting from a state of the world where suppose there financial claims, become safe claims is Z is plentiful and agents are confident that they hold a shock that reduces Z. As collateral falls, safe agents recognize that previously risky as they lose their collateral backing. In response, agents act aggressively to shield themselves from risk by shedding risky claims and purchasing safe claims, in a manner consistent with a flight to quality. These actions lock-up the the effective collateral of the economy. collateral of financial intermediaries, Our model shows and thereby further reduce economy exacerbates the that the robust collateral shortage caused by reduced aggregate collateral, amplifying the standard financial accelerator highlighted by Bernanke and Gertler (1989) and Kiyotaki and Moore (1997), among We to i.e. is can also think of the lender of increased collateral To a The will deliver on The problem last resort policy in these terms. that agents respond large extent, private sector collateral will always be perceived as risky - A is intermediaries can never insulate themselves against default risk. "riskless." 6 risk. others. central bank promise, if credible, lender of last resort policy derives benefits because agents are certain that the central bank promise. its The value of such riskless promises rises when private sector collateral is risky. Moral Hazard and Irrelevance Critiques In this section central bank we discuss our lender of last resort intervention. The moral hazard recommendation critique is in light of that the anticipation of an explicit or implicit insurance policy leads the private sector to underinsure against shocks and to backfire. The irrelevance critique is two prominent critiques of may lead the central bank policy that the events warranting insurance, such as the Great Depression, are so rare in economies with deep and sound financial markets, that while in principle the central insurance is useful, in practice it has minimal ex-ante welfare implications. 26 bank In discussing tliese critiques, Moral hazard 6.1 The moral hazard we also extend our model to tire case oi critique In our model, in keeping with the moral hazard critique, agents activities. the reallocation improves (ex-post) outcomes on average.^® We waves of aggregate shocks. predicated on agents responding to the provision of public insurance by cutting is from the publicly insured shock. However, when reallocate insurance awa}' in 2 critique back on their own insurance complements N> flight to quality is Public and private provision of insurance are our model. note that the central bank can achieve the same distribution of insurance intervene in the the concern, first shock. However the expected intervention, since the central is bank rather than the is much instead it commits to larger than the extreme event private sector bears the cost of insurance against the Agents would reallocate the expected resources from the central bank to the (likely) single-shock event. two-shock event, which cost of this policy if exactly the opposite of what the central bank wants to achieve. In this sense, interventions in intermediate events are subject to the moral hazard critique. More ^= formally, let us return to the simpler shock in the first model and note that the direct benefit of intervention is; V^r^''^'^^' = 2 / 0^^(l)u'(ci,^)do. = d{l)u'{C2). Jn The Kjc^'i remains unchanged because the relevant constraint remains the constraint on total benefit, two-shock insurance; V -total ^S^(l)-'(=i, = J^ = J§f + g^) + [05^(l)«'(c:,J - [0^^(l)^/(c,,J (2 4>il)n'ic,) - - 0^^(2)^.'(C3, J (2 - €^(2)u'(c2,.) <^^«(2)k'(c2,„)] j^ •I = The r direct, first ""-^ -— "(^^^l^r^\^) f/ \ ^ ( t l + \ min(A-,A-) 1 dC2.ui du) ^d^ "^2,cj ^ xr direct, first jdZ^^^^- Agents reduce insurance against the first-shock (reduce risky claims) and increase their second- shock claims (increase safe claims). This private insurance reallocation policy. has to be a ^^Note that That last if is, the lender of last resort facility, to be effective offsets some of the benefits of the and improve private financial markets, not an intermediate resort. the direct effect of intervention inconsistent. This result the policy dw ) anticipation of the central bank's first-shock insurance leads agents to reoptimize their insurance decisions. LLR ^ is is insufficient to justify intervention, then the lender of last resort policy is time not surprising as the benefit of the policy comes precisely from the private sector reaction, not from itself. 27 Multiple shocks 6.2 It is clear that the for late shocks; LLR but if should not intervene during early shocks and instead should only pledge resources we move away from our two-shock model potential waves of aggregate shocks, To answer experience The [T < A'' central this question how injects 1/(A'' —T+ into the model. The LLR 1/A^tli of the agents. 1) units of liquidity for all N). For simplicity we focus on the fully more realistic context with multiple late? is we introduce multiple shocks waves of shocks, each affecting bank The value late to a We assume the economy may policy takes the following form: shocks after (and including) the Tth wave robust case. of the intervention as a function of T is, N i=T = where we used the insurance. As fact that T one unit of additional collateral allows intermediaries to sell A'' units of additional before, the anticipation of the central bank's insurance leads agents to reoptimize their private insurance decisions. Agents reduce insurance against the publicly insured shocks and increase their private insurance for the rest of the shocks. well as The total benefit of the intervention includes any benefit from portfolio reoptimization: N '^ i=l with: V^^-A^ ^ dZG i =l so that, In the fully robust case, Ci and j^ are the same for all i. Then, 7 =1 = «'(ci)^E^(^) 1=1 28 both the direct benefit as Note that this expression is is decreasing with respect to is strictly greater than one Of course, the above it independent of the intervention rule T. In contrast, T since the T> for all 1 (/i(i)'s and is are monotonically decreasing. amount of resources available for intervention, the — 1st shock, it LLR should first completely replaces private insurance, it pledge should and so on. Irrelevance critique 6.3 The the ratio: increasing with respect to T. resources to the A''th shock and continue to do so until N Then V^]^^<^^ result does not suggest that intervention should occur only in the A''th shock. Instead suggests that for any given then move on to the apparent that it is multiple shock model also clarifies another benefit of late intervention. insured by the LLR become resources for the central increasingly less bank is likely. If we take the As T rises, case where the constant, the expected cost of the LLR policy events that are being shadow falls as T cost of the rises, LLR while the expected benefit remains constant. In other words, as T rises, it is more private resources to early and quality the private sector that increasingly improves the allocation of scarce likely aggregate shocks, phenomenon. In contrast, the central bank plays a decreasingly small value of resources actually disbursed, as Thus, while a well designed (hence the irrelevance critique), benefits 7 thereby reducing the extent of the flight-to- come from the impact LLR its T terms of the expected increases. policy main role in may indeed ha\'e a direct effect only in highly unlikely events benefits are during more likely and felt less extreme events. These of the policy in unlocking private collateral. Final remarks a pervasive phenomenon that exacerbates the impact of recessionary shocks and finan- Flight to quality is cial accelerators. In this paper we present a model of this financial specialists. We show that when aggregate phenomenon based on robust intermediation collateral is decision plentiful, robustness consider- ations do not interfere with the functioning of private insurance markets (credit lines). However, think that aggregate intermediation collateral is making by when agents scarce, they take a set of protective actions to guarantee themselves safety, but which leave the aggregate economy overexposed to recessionary shocks. In this context, a Lender of Last Resort policy events. The main benefit of this policy is useful if used to support extreme rather than intermediate comes not so much from the 29 direct effect of the policy during extreme events, more which are very rare, but from its abiHty to unlock private sector collateral during milder, and far frequent, contractions. The implications of the framework extend bej'ond the particular interpretation and policymakers. For example, the polic3'maker as the IMF in we have given to agents the international context one could think of our agents as countries and or other IFFs. Then, our model prescribes that the IMF not support the first country hit by a sudden stop, but to commit to intervene once contagion takes place. The benefit of this policy comes primarily from the additional availability of private resources to limit the impact of the initial pullback of capital flows. 30 References [1] Barro, R.J., 2005, "Rare Events and the Equity Premium," Mimeo, Harvard University. [2] Bernanke, B., M. Gertler, and S. Gilchrist, 1999, "The Financial Accelerator in a Quantitative Business Cycle Framework," Handbook of Macroeconomics, Taylor and Woodford, eds. [3] WP # [4] [5] 11996. Diamond, D.W. and P.H. Dybvig, 1983, "Bank Runs, Deposit Insurance, and Political Economy Dow, and J. Portfolio," [6] NBER Caballero, R.J. and A. Krishnamurthy, 2005, "Financial System Risk and Flight to Quality," 91, pp. 401-419. Werlang, 1992, "Uncertainty Aversion, Risk Avei'sion, and the Optimal Choice of S. Econometrica 60, p. 197-204. Friedman, B., and Kuttner, K., 1993, Stock, J., Liquidity," Journal of Watson, M. (Eds.), New "Why does the paper-bill spread predict real economy activity." In; research in Business Cycles: Indicators and ForecEisting. University of Chicago Press, Chicago. [7] Easley, David and Maureen O'Hara, 2005, "Regulation and Return: The Role of Ambiguity," working paper, Cornell University. [8] Epstein, L., 2001, "Sharing Ambiguit}'," [9] Epstein, L. and American Economic Review, M. Schneider, 2004, "Recursive Multiple 91(2), 45-50. Priors," Journal of Economic Theory, 113(1), 1-31. [10] Epstein, L. and T. Wang, 1994, "Intertemporal Asset Pricing Under Knightian Uncertaintj',' Economet- rica 62, 283-322. [11] Gatcv, Evan, and Phil Strahan, 2005, "Banks' Advantage in Hedging Liquidity Risk: Theory and Evidence from the Commercial Paper Market," forthcoming. Journal of Finance. [12] Gilboa, I. and D. Schmeidler, 1989, "Maxmin Expected Utihty with Non-unique Mathematical Economics, [13] Priors," Journal of 18, 141-153. Greenspan, A, 2004, "Risk and Uncertainty in Monetary Policj'," Remarks for the AEA Meetings, January. [14] Hansen, L. and T. Sargent, 1995, "Discounted Linear Exponential Quadratic Gaussian Control,' Transactions on Automatic Control 40, 968-971. 31 IEEE [15] Hansen, L.P. and T.J. Sargent, 2003, "Robust Control of Forward-looking Models," Journal of Monetary Economics [16] 50(3), 581-604. Hansen, L.P., T.J. Sargent, G.A. Turmuhambetova, N. Williams, 2004, "Robust Control and Model Misspecification," April, [17] Mimeo. Holmstrom, Bengt and Jean Economy, 106(1), 2/98, pp. [18] Tirole, "Private and Public Supplj' of Liquidity," Journal of Political 1-40. Kashyap, Anil, Jeremy Stein and David Wilcox, "Monetary Policy and Credit Conditions: Evidence from the Composition of External Finance," Am.erican Economic Review, 83(1), 3/93, pp 78-98. Moore, 1997: "Credit Cycles," Journal of Political Economy, 105 [19] Kiyotaki, N. and [20] Krishnamurthy, Arvind, 2003, "Collateral Constraints and the Amplification Mechanism," Journal of J. Economic Theory, [21] (2), 211-248. 111(2), 277-292. Krishnamurthy, Arvind, 2002, "The Bond/Old-Bond Spread," Journal of Financial Economics 66(2), 463-506, 2002. W. Xiong, 2001, "Contagion as a Wealth Effect", Journal of Finance 56, 1401-1440. [22] Kyle, A. and [23] Longstaff, R-ancis, 2004, "The Flight-to-Liquidity Premium in U.S. Treasury Bond Prices," Journal of Business 77(3), 511-526. [24] Routledge, Bryan and Stanley Zin, 2004, "Model Uncertainty and Liquidity," working paper, Carnegie Mellon [25] Universit}'. Sims, Cliristopher A., 2001, "Pitfalls of a Minimax Approach to Model Uncertainty," working paper, Princeton University. [26] Skiadas, Costis, 2003, "Robust Control and Recursive Utihty," Finance and Stochastics [27] Stock, J. and Watson, M., 1989. New 7, 475-489. indexes of coincident and leading economic indicators. NBER Macroeconomics Annual, 351-394. [28] Woodford, M., "Public Debt as Private Liquidity," American Economic Review, Papers and Proceedings 80, May 1990, pp. 382-88. 32 52 7