Contagion involves run behavior, whereby fears of widespread financial collapse lead to the withdrawal of funding from banks and other financial institutions. The problem of contagion is of hoary vintage.1 The term “contagion” denotes the spread of run-like behavior from one financial institution to an expanding number of other institutions, reducing the aggregate amount of funding available to the financial system and in turn to the economy.2 Contagion can also spread to short-term capital markets that fund the complex and growing assortment of nondepository financial institutions in the financial system. The special feature that distinguishes contagion from other causes of systemic instability is the tendency of contagious runs to propagate indiscriminately, apart from connectedness. Contagion is “indiscriminate” when it afflicts healthy markets and solvent institutions.3
Financial institutions are vulnerable to contagion due to their dependence on short-term borrowing to fund long-term investment activity. When short-term debt investors suddenly refuse to extend funding, institutions relying on such funding engage in fire sales of assets and ultimately fail. The fire sales may generally decrease asset values of other institutions who may also fail, as well. Runs of short-term creditors may be rational or irrational and need not be indiscriminate. A run by short-term creditors can be targeted to a single or limited number of financial institutions, for example, those known to have incurred significant losses.4 During a contagious and indiscriminate run, however, investors may also withdraw funding from multiple institutions or markets that are not themselves facing any objective business distress. In such an environment the decision to exit is made not on the basis of specific information but rather because investors possess insufficient information to differentiate their own risks from those that others are—or appear to be—facing. This dynamic, one former central banker has warned, may “lead to failures of other financial intermediaries, even when [they] have not invested in the same risks and are not subject to the same original shocks.”5 If these intermediaries fund themselves using short-term capital instruments, contagion effects may spread to the markets where these instruments trade. Sudden demand for liquidity by investors in intermediaries that normally hold these instruments or a refusal on the part of interbank lenders to renew their funding can trigger liquidations or freeze-ups in these markets, triggering fire sales, decimating asset prices, and halting lending activity.
Contagion presents risk of a singular nature to financial institutions.6 Contagion is a structural feature of the financial system that is endogenous to the economics of maturity transformation and, in my judgment, is not likely to be resolved through better risk management or improved prudential oversight. Absent affirmative steps to contain it, the problem of contagion will continue to haunt the financial system.
Financial historians disagree as to the extent of contagion’s role in the periodic disruptions of the US financial system over the past two centuries. Clearer is the role that contagion has played in the development of the US Federal Reserve System, which was created in reaction to the Panic of 1907, although the notion of a central bank lender of last resort had been contemplated throughout the late nineteenth century in response to successive panics and waves of bank failures.7 Critical in the creation of the Fed was the belief that private efforts of the clearinghouses to provide liquidity to their members were not sufficient to forestall panics.8
Some economic historians trace the frequency of bank failures in US history to an abnormally high level of concentration risk promoted by the decentralized structure of the US banking system. This decentralization was a by-product of the restructuring of the American banking industry during the National Banking Era, stimulated by distinctive legislative changes in the United States that were not duplicated elsewhere.9 According to this account, branching restrictions embodied in the National Bank Act of 1864 propelled a thirteenfold increase in the total number of US banks over the next fifty years.10 By 1914 the unprecedented expansion of banking in the United States had culminated in a unit banking system comprising 22,030 institutions nationally.11 The massive proliferation of small banks managing localized loan portfolios created a banking system in which each bank’s balance sheet was concentrated in a local economy. This lack of diversification left each bank vulnerable to idiosyncratic shocks, and therefore increased the overall failure rate.12 The small banks might also have suffered from poorer management relative to their larger and more sophisticated peers. This background of “too small to survive” is ironic in view of the current concern with “too big to fail.”
Countries whose banking systems did not develop along decentralized lines have not faced repeated waves of financial panic. In Canada, for example, bank failures have been rare, even though the Canadian macroeconomic environment has generally tracked the US experience.13 Between 1870 and 1913, Canadian banks underwent 23 liquidations, compared with 3,208 recorded in the United States over the same period.14 No banks failed in Canada between 1923 and 1985, whereas between 1930 and 1933 alone, 9,000 US banks suspended operations.15 Such discrepancies are not attributable to the variance in the performance of the Canadian and US economies but may reflect dramatic differences in industry consolidation.16 Then again, the different bank failure rates across the two systems might also be the result of any number of other political, regulatory, and social factors distinguishing the Canadian and US environments from each other during the Great Depression.
In their study of the Chicago Banking Panic of 1932, Charles Calomiris and Joseph Mason conclude that worsening economic conditions can cause depositors to withdraw their money from weak banks in favor of stronger ones, showing that most bank failures resulted from homogeneous balance sheet impairments caused by the collapse in asset prices after the onset of the Great Depression.17 This finding is striking given the tight geographic focus of the Panic, in which some 40 Chicago-area banks failed, including 26 during a single turbulent week in June 1932.18 Although these characteristics appear to resemble a classic bank run, the authors reject this interpretation, insisting instead that most of the banks that did succumb were “distinguishable months before the panic,”19 the evidence of their preexisting mass insolvency “reflected in stock prices, failure probabilities, the opinions of bank examiners, debt composition, and interest rates.”20 The authors find, by contrast, that solvent banks did not fail during the Chicago Panic.21 Part of the explanation for the sharply differentiated performance of solvent banks may be that the healthy banks were able to supplement lost deposits by coordinating private interbank lending facilities, to which insolvent institutions did not have access.22
Some scholars and finance professionals have minimized the role of contagion in the unfolding of the 2008 financial crisis. Financial Times columnist Martin Wolf, for example, assigns primary blame for the crisis to asset shocks and macroeconomic instabilities linked to long-term international imbalances in global trade, savings rates, and investment.23 In another study, Nicolas Dumonteaux and Adrian Pop scrutinize the impact on financial institutions of the Lehman bankruptcy, determining that contagion effects, to the extent any existed at all, were “firm-specific, rational and discriminating rather than industry-wide-specific, ‘pure’ panic-driven or undifferentiated.”24 Like the Chicago Panic, firms that were most affected by the collapse of Lehman possessed comparable core business characteristics, operating fundamentals, and a performance record that was measurably correlated with Lehman.25 Appraising the totality of the evidence, Dumonteaux and Pop conclude that the effects of Lehman’s failure on financial institutions were neither indiscriminate nor contagious.26
Other commentators offer a more persuasive, contrary view. William Sterling has argued that studies finding an absence of indiscriminate contagion are based on an incomplete set of indicators.27 Using the Bloomberg Financial Conditions Index, which incorporates a broader set of indicators, Sterling showed that the Lehman failure was an immediate and massive shock to an already stressed financial system.28 Other studies have found that Lehman was a dangerous transmitter of contagion, with Jian Yang and Yinggang Zhou finding, for example, that an increase in the price of a Lehman CDS caused subsequent increases in the CDS prices of other financial institutions.29 Lehman’s effect on other firm’s was the largest among global financial institutions.30 JPMorgan was apparently a prime beneficiary of this kind of funding transfer during the financial crisis, as retail customer deposits31 and prime brokerage assets32 flowed out of weakened commercial and investment banking institutions and into JPMorgan’s insured deposit and prime brokerage accounts. Writing in his annual letter in 2009, Jamie Dimon, JPMorgan’s chief executive officer, advised shareholders that the bank received a net inflow of deposits as investors fled lower quality institutions. “As we entered the most tumultuous financial markets since the Great Depression,” Dimon wrote, “we experienced the opposite of a ‘run on the bank’ as deposits flowed in (in a two-month period, $150 billion flowed in—we barely knew what to do with it).”33
Several important US financial firms that arguably possessed considerably stronger business models than Lehman, such as Morgan Stanley and Goldman Sachs, do in fact appear to have been affected by some degree of run behavior after the Lehman failure. Although “[s]ignificant bank runs were not a feature of the financial crisis,” some large banks did fall prey to runs. Wachovia faced a bank run on its deposits prior to its acquisition by Wells Fargo.34 Washington Mutual faced a similar fate before its eventual sale to JPMorgan.35 Importantly, nonbank financial institutions, beginning with Bear Stearns36 and later spreading to critical segments of the short-term capital markets, as well as the money market funds, also underwent serious runs.37 Although no significant financial institution sharing Lehman’s basic business attributes collapsed as a result of Lehman’s failure, this was likely a reflection of the bailout signals transmitted by the Federal Reserve’s rescue of AIG as well as by the multifaceted public support programs instituted by the US Treasury and the Federal Reserve.38 As discussed at greater length below, the evidence shows substantial contagion effects elsewhere in the financial system. These effects were transmitted initially through the Reserve Primary Fund (RPF) to other prime money market funds; certain segments of the asset-backed, financial, and corporate commercial paper markets; and unsecured interbank lending and secured repo borrowing markets. Ultimately, they resulted in serious runs on other investment banks as investor confidence in the vitality of the independent investment banking business model deteriorated.
The best-developed theory of systemic risk attributes financial panic to run behavior by short-term creditors that spreads to multiple financial institutions.39 Depositors, operating under the constraints of asymmetric information, withdraw from all banks indiscriminately. Douglas W. Diamond and Philip H. Dybvig have suggested that the “shift in [creditor] expectations” can “depend on almost anything,” whereas others have attributed contagion to a change in the business cycle,40 as originating from a lack of timely market information,41 or as one instance of a more general form of crowd behavior documented in nonfinancial contexts.42 According to the first view, contagion can be triggered by purely random phenomenon, or economic “sunspots.” The latter view attributes contagion to “informational cascades” in which individual market participants use the actions of peers as cost-effective surrogates for actual data collection about an underlying reference entity. When peers run, they run too.43 Each of these explanations shares the recognition that contagion is not conditioned on insolvency. Instead, contagion is a liquidity-driven phenomenon that reflects the maneuvering of short-term creditors in response to informational constraints, rational incentives, and structural vulnerabilities uniquely characteristic of financial intermediaries dependent on short-term borrowing and longer term assets. Such constraints can provoke short-term creditors to withdraw from institutions preemptively, even if they are fundamentally well-capitalized and have no exposure to losses connected to an asset shock, as occurred in money market funds with no exposure to Lehman during the financial crisis.
Contagion theory historically focused on runs by uninsured depositors to explain the wave of bank failures of the 1930s and elsewhere in modern financial history, although the underlying economic explanation for contagious runs applies equally to the behavior of nondeposit short-term creditors.44 Contagion can spread indiscriminately to solvent institutions, causing “real economic problems because even ‘healthy’ banks can fail.”45 Financial institutions that succumb to contagion may be solvent immediately beforehand and may not display characteristic warning signs of distress detectable in advance by regulators.46 Importantly, contagion and runs on individual financial institutions are distinct, albeit related, phenomena. An isolated run by short-term investors on a single financial institution is not an example of contagion. Contagion only occurs when a run at one institution or some other event induces short-term creditors of multiple other institutions to run, including from institutions that are adequately capitalized and have no financial linkage to the same set of problematic risk exposures.47 Under certain circumstances individual runs can generate systemic contagion effects, provoking further runs. Contagion can therefore develop from a generalized fear of failure on the part of short-term creditors as much as it may represent a reaction to specific cases of real distress.
The Diamond and Dybvig model establishes that banks and other financial intermediaries exist at “multiple equilibria.”48 Because maturity transformation requires an intermediary to finance long-term illiquid assets (e.g., mortgages with maturities spanning multiple decades)49 with short-term or demand liabilities redeemable at par, one of these equilibria is a run:
Banks are able to transform illiquid assets by offering liabilities with a different, smoother pattern of returns over time than the illiquid assets offer. These contracts have multiple equilibria. If confidence is maintained, there can be efficient risk sharing, because in that equilibrium a withdrawal will indicate that a depositor should withdraw under optimal risk sharing. If agents panic, there is a bank run and incentives are distorted. In that equilibrium, everyone rushes in to withdraw their deposits before the bank gives out all of its assets. The bank must liquidate all its assets, even if not all depositors withdraw, because liquidated assets are sold at a loss.50
The core of this account is what some commentators have labeled a collective action problem.51 When short-term creditors of a maturity-transforming firm develop suspicions that the firm is verging on insolvency, the creditors have a rational motivation to withdraw funding before the firm’s supply of liquid reserves is drained by others responding to the same incentives. Generating enough liquidity to redeem exiting creditors at par forces the firm into monetizing long-term assets at noneconomic valuations. In the ensuing fire sale, the firm incurs actual losses, thus realizing the concern that had caused creditors to panic in the first place. Even though all short-term creditors would collectively be better served by remaining invested and seeking to maximize their recoveries through an orderly disposition of long-term assets, each individually has a strong incentive to be the first to exit. A downward spiral at one firm becomes contagious when it induces short-term creditors of other firms to develop similar concerns and incentives, initiating multiple distressed liquidations that ultimately engulf healthy financial institutions, drive down asset prices, and cause systemic balance sheet impairment of otherwise solvent firms through forced sales of assets at fire sale prices and mark-to-market accounting losses, where accounting rules force banks to mark down their assets to the fire sale prices.52
Runs can be a self-fulfilling prophecy. Fire sales initiated by affected institutions to (1) fund withdrawals of liquidity, (2) post margin, or (3) cover defaults by counterparties through the liquidation of collateral53 cause asset prices to fall, impairing institutional balance sheets, depleting capital, and driving institutions into insolvency. Milton Friedman and Anna Schwartz observed that at this point the run may become “self-justifying,” since the fire sale “force[s] a decline in the market value of ... the remaining assets” held on institutional balance sheets, which in the worst cases brings about actual insolvency.54
Mason (1998) argues that “pure contagion involves changes in expectations that are self-fulfilling, with financial markets subject to equilibria or sunspots,” and that “changes in expectation … are not related to changes in ... fundamentals.”55 One potential weakness of this theory is that it may be challenging to construct empirical tests for the presence of multiple equilibria. However, Hortacsu et al. (2011)56 constructed a testable model in which companies face a negative feedback loop between their perceived financial health and demand for their products. In this model, financial distress can reduce the demand for a firm’s products by, for example, reducing the expected value of the company’s warranties. In the banking sector, a warranty corresponds to a bank’s promise to convert deposits to cash on demand. The relationship between demand and financial health can generate a vicious cycle: financial distress reduces demand, reduced demand increases financial distress, which further reduces demand. As in the Diamond Dybvig model, perceived distress can be just as pernicious as actual distress, because “if consumers suddenly believe a firm is distressed, even incorrectly, the resulting demand effect could push the firm into distress and even bankruptcy.”57 The potential for multiple outcomes—all dependent on the company’s perceived financial health—yields “multiple equilibria.”
According to the widely cited work of Gorton and Metrick (2012), “[t]he 2007–2008 financial crisis was a system wide bank run.”58 They draw a historical analogy between the recent crisis and nineteenth-century panics, when banks “suspended convertibility [of deposits] and relied on clearinghouses to issue certificates as makeshift currency.”59 In the nineteenth century, evidence of contagion could be found in the discounts at which these certificates traded. In the twenty-first century, contagion is evidenced by “unprecedented[ly] high repo haircuts and even the cessation of repo lending on many forms of collateral.” There were additional runs on asset-backed commercial paper and other products. The authors observe that the potential for traditional banking runs, of the type observed in the nineteenth century, was eliminated by the expansion of Fed discount window lending and deposit guarantees. They argue that collateralized lending arose as a private-sector partial substitute for government guarantees. For example, in a repo transaction an investor lends money to a bank (equivalent to a traditional deposit) by purchasing a security and agreeing to sell it back after a fixed period of time. The collateral fills the role of a government guarantee, as the investor can sell it should the bank become unable to repurchase it. In the traditional bank run, investors simply refused to lend to banks through deposits. In the recent bank run, investors refused to lend through repos. The root cause of the modern run was a refusal to accept collateral, because investors decided its value as a guarantee-substitute was diminished. Hence it is important to ask what drove the contagious refusal to accept collateral.
Gorton and Ordonez (2014)60 and Dang, Gorton, and Holmstrom (2013) offer information economics as an explanation for the run on repo and collateralized lending. They argue that during ordinary periods, short-term collateralized debt is “money-like” in the sense that traders of it are information insensitive. In other words, the prices of these assets are not sensitive to the release of new information, and market participants therefore have limited incentive to generate this information. However, a small idiosyncratic shock can trigger investors to become information sensitive, which creates price drops as negative information is generated. In addition the fact that the assets become information sensitive means that some market participants will have superior information to other market participants (“asymmetric information”). Fearing that their counterparty has superior information, purchasers of these assets will offer prices lower than their expected value to avoid adverse selection (i.e., buying at a price higher than the asset is worth given existing information). According to this model, the sudden shift in information sensitivity, coupled with asymmetric information, is the root cause of contagion. The response to asymmetric information pushes asset prices below their fundamental value. This amplifies the original idiosyncratic shock that caused the market to become information sensitive, and the market plunge can lead to systemic crisis. Because the “adverse selection discount” is rooted in asymmetric information, rather than price-relevant fundamental information, the result is an indiscriminate bank run, spreading from asset-to-asset and institution-to-institution. Once adverse selection discounts are recorded in market prices, banks realize losses by marking their balance sheets to market.61 This may be enough to trigger further contagion.
In the academic literature, the three leading measures of systemic risk are Adrian and Brunnermeier’s (2010) conditional value at risk (CoVaR), Acharya et al.’s (2011) systemic expected shortfall (SES), and Billio et al.’s (2012) measure of interconnectedness.62 However, each such measure is really a measure of correlation, and not of connectedness or contagion. As a result these measures fail to have predictive power for systemic risk that arises from either connectedness (which I show is not the primary source of systemic risk) or contagion (which I show is the primary source of systemic risk). A gap remains in the academic literature for measuring the risk of contagion. For a detailed discussion of each systemic risk measure, see the appendix.