SOME REFLECTIONS ON THE RECENT FINANCIAL CRISIS
Gary B. Gorton
https://www.nber.org/system/files/working_papers/w18397/w18397.pdf
September 2012
11:
4. Crisis Theory 11 Theoretically, a financial crisis is defined by two essential points. First, a crisis is a singular event. It is a rending, a sundering, or a rupturing, of the normal state of affairs in money markets. A financial crisis is not the worst outcome on a continuum of bad events. There is no continuum in an important sense. There are booms and recessions, and then here are crises. A crisis is a distinct event. Something happens to make a crisis fundamentally different from the usual economic downturn. Second, while each crisis has important unique features, crises have a common root cause. There is a structural feature of bank debt that makes the debt vulnerable to runs. And the bank debt in question is not just demand deposits. Financial crises are always about bank runs. The bank runs either occur or would have occurred had the government or central bank not intervened or been expected to intervene. The first point says that a “crisis" is not simply a particularly "bad state" of the world. A crisis is fundamentally different, a different regime. There are normal non-crisis states and there is an extraordinary crisis state. This is why Anna Schwartz (2007) said that “a decline in asset prices of equity stocks, real estate, commodities; depreciation of the exchange value of a national currency; financial distress of a large non-financial firm, a large municipality, a financial industry, or sovereign debtors-are pseudo-financial crises" (p. 245). They may be bad events, wealth may be destroyed or cleanup costs high, but they are not crises. A financial crisis is a systemic event. The entire financial system is engulfed. The failure of a large firm or problems in one sector, e.g., savings and loans or the auto industry, are not crises in this sense. Financial crises repeatedly occur in market economies. The second point is that there is a reason for this. There is a root cause. Agents in the economy need private money to transact. But, this money is vulnerable to runs. Bank runs are crises. Financial crises are caused by bank runs. The root of the financial crisis problem was elegantly identified by Diamond and Dybvig (1983).⁹ Diamond and Dybvig studied a setting where banks must use long-term collateral to back demand deposits. Agents need the demand deposits because of potential shorter-term liquidity needs. The investments are "long" in the sense that if they are liquidated early there is a very low return. "Long" also means relative to the required frequency of agents' transactions for consumption or other shortterm needs. The agents need demand deposits to smooth consumption, which is uncertain as some 9 There is a large literature on the Diamond and Dybvig model, many extensions and discussions, but I will, for the most part, not go into this literature.
4. 危機理論
11 理論的には、金融危機は2つの重要なポイントによって定義される。第一に、危機とは特異な出来事である。金融危機とは、金融市場の正常な状態を破壊する、引き裂く、破裂させるような出来事である。金融危機は、連続した悪い出来事の中で最悪の結果をもたらすものではない。重要な意味での連続性はない。好景気と不景気、そして危機がある。危機は別個の出来事である。危機を通常の景気後退とは根本的に異なるものにする何かが起こるのである。第二に、それぞれの危機には重要な特徴があるが、危機には共通の根本原因がある。銀行債務には、債務不履行に陥りやすい構造的な特徴がある。そして、問題となる銀行債務は、要求払い預金だけではありません。金融危機は、常に銀行の取り付けに関わるものである。銀行経営は、政府や中央銀行が介入しなかった場合、あるいは介入を期待された場合に発生するか、あるいは発生したであろうものである。第一のポイントは、「危機」とは、単に世の中の特に「悪い状態」ではない、ということです。危機とは、根本的に違う、別の体制である。通常の非危機的な状態があり、異常な危機的な状態がある。アナ・シュワルツ（2007）が、「株式、不動産、商品などの資産価格の下落、自国通貨の交換価値の下落、非金融大企業、大規模自治体、金融業界、政府債務者の財政難-これらは疑似金融危機」（P245）と述べているのは、このためである。これらは悪い出来事であり、富が破壊されたり、後始末のコストが高くついたりするかもしれないが、危機ではない。金融危機とは、システミックな出来事である。金融システム全体が巻き込まれるのである。大企業の破綻や、貯蓄貸付や自動車産業のような一部門の問題は、この意味での危機ではない。金融危機は市場経済において繰り返し発生する。第二のポイントは、これには理由があるということです。根本的な原因があるのです。経済のエージェントが取引をするためには、私的な貨幣が必要である。しかし、この貨幣は、暴落に弱い。銀行融資は危機である。金融危機は、銀行の取り付けが原因である。金融危機の問題の根源は、Diamond and Dybvig (1983) によって見事に特定された⁹ Diamond and Dybvig は、銀行が要求払い預金の裏付けとして長期担保を使用しなければならない設定を研究した。エージェントは、短期的な流動性ニーズのために要求払預金を必要とする。この投資は、早期に清算された場合のリターンが非常に低いという意味で「長い」ものである。また、「長い」とは、消費または他の短期的なニーズのためにエージェントが必要とする取引の頻度に比べて相対的に長いことを意味する。9 Diamond and Dybvigモデルについては、多くの文献があり、多くの拡張や議論がなされているが、ここではほとんど触れないことにする。
12:
agents may want to consume early. An essential feature of the model is that the interest rate offered
on the demand deposits to achieve this smoothing is such that if all agents want to consume early (by
withdrawing from the bank), then the bank cannot satisfy these demands. This is the critical fragility in
the economy.
A very important point is that there is no way around this basic horizon problem in any market
economy. People eat lunch every day, but it takes a long time to build a factory and produce output.
People need to pay for their lunch before the output is realized. This timing is fundamental. Bank debt
used for transactions can only be backed by these long investments (which have a low return if
liquidated early). The private sector cannot produce riskless assets. These basic facts mean that financial
intermediaries will always be involved in "maturity transformation," a term which just restates this fact.
"Maturity transformation" is not a choice. It can't be regulated away. It is inherent in any economy
which produces private bank money, that is, any market economy. It is a fundamental fact. Bank
money can only be backed by longer-term investments. As we will see later, agents in the economy will
strive mightily to design bank debt to overcome this problem. But, without the government, bank debt
will always be vulnerable.
Uncertainty about consumption timing is a risk the agents want to shed using bank debt. The problem
of long-term collateral backing bank debt is a necessary but not a sufficient condition for crises to occur.
To get a crisis-a bank run, Diamond and Dybvig introduce a source of uncertainty that is quite special.
It is the uncertainty that each individual bank depositor faces about the actions of other deposit holders.
Depositors care about the actions of other depositors if there is a common pool of assets on which they
all have pari passu claims-the bank's assets-- but the claims are honored sequentially (so they are not
in fact pari passu). Note that the assumption of sequential service means that the payout of the bank to
an individual depositor depends on the actions of the other depositors. How much a depositor gets
back depends on his place in the line. In this setting, depositors may run if they think other depositors
are going to run. Each depositor has an incentive to be first in line to withdraw at the bank if he believes
that other depositors are going to line up. Beliefs about other depositors' beliefs must depend on
something and in Diamond and Dybvig beliefs are coordinated by an extraneous random signal, a
sunspot.
The bank run, due to the beliefs coordination problem, displays the second essential condition of the
definition of a crisis, discussed above. A run in the Diamond and Dybvig model is fundamentally
のエージェントは、早期に消費したいと考えるかもしれない。このモデルの本質的な特徴は、この平滑化を実現するために要求払い預金に提供される金利が、すべてのエージェントが（銀行から引き出すことによって）早期に消費したいと考えた場合、銀行がこれらの要求を満たすことができないようなものであることである。これが経済における決定的な脆弱性である。非常に重要な点は、どのような市場経済においても、この基本的な地平線の問題を回避する方法はないということです。人々は毎日昼食を食べますが、工場を建てて生産物を作るには長い時間がかかります。生産が実現する前に、人々は昼食の代金を支払う必要があるのです。このタイミングが重要なのだ。取引に使われる銀行債務は、このような長期投資（早期に清算するとリターンが低い）に裏打ちされたものでしかない。民間部門は無リスクの資産を生産することはできない。このような基本的な事実から、金融仲介者は常に "maturity transformation "に関わることになる。「満期変換」は選択できるものではありません。規制で排除することはできない。民間銀行貨幣を生産する経済、すなわち市場経済には必ず存在する。これは基本的な事実である。銀行資金は、長期的な投資によってのみ裏打ちされる。後ほど説明するように、経済の担い手たちは、この問題を克服するために銀行債務を設計しようと必死に努力する。しかし、政府がいなければ、銀行債は常に脆弱である。消費タイミングの不確実性は、銀行借入によって回避したいリスクである。銀行債務の裏付けとなる長期担保の問題は、危機が発生するための必要条件ではあるが、十分条件ではない。DiamondとDybvigは、危機を引き起こすために、非常に特殊な不確実性を導入している。それは、個々の銀行預金者が、他の預金者の行動に関して直面する不確実性である。預金者が他の預金者の行動を気にするのは、預金者全員が同順位に請求権を持つ共通の資産プール（銀行の資産）があり、その請求権が順次履行される（つまり実際には同順位ではない）場合である。順次返済の仮定は、銀行から個々の預金者への支払いが、他の預金者の行動に依存することを意味することに注意。ある預金者がいくら払い戻されるかは、その預金者の順位に依存する。この設定において、預金者は、他の預金者が逃げ出すと思えば、逃げるかもしれない。各預金者は、他の預金者が列に並ぶと思えば、銀行で一番に引き出そうとするインセンティブがある。他の預金者の信念に関する信念は何かに依存しなければならず、Diamond and Dybvigでは信念は外来ランダムシグナルである太陽黒点によって調整される。信念の調整問題による銀行経営は、前述の危機の定義の第二の必須条件を示している。Diamond and Dybvigモデルにおける銀行経営危機は、基本的に次のようなものである。
13:
13
different from the normal state of affairs. There are no "small" crises in the Diamond and Dybvig model.
There are two outcomes: no crisis and crisis. The crisis is a distinct, very different, event. This is
consistent with the empirical evidence that there are distinct events that can be called "crises" and
which are clearly much worse outcomes than recessions or Anna Schwartz's pseudo-financial crises. The
model lays out a convincing setting and shows that the outcome can be very different than the normal
state of affairs, a run can occur-a crisis. This was the first model that displayed the two essential points
articulated above. In this sense, it provides a coherent picture of a financial crisis.
But, as a theory of crises, the Diamond and Dybvig model is not completely satisfactory. The very
phenomenon we want to explain, why there is a loss of confidence, is not explained - it is "sunspots."
That is, each agent believes that the other agents will run when they observe "sunspots." While the
coordination device is called "sunspots,” this is just a name for the multiple equilibria that can occur in
the model. There is no explanation for why the economy switches from one equilibrium to another.
The issue of belief coordination is especially troublesome. There is no explanation for why a run would
suddenly occur. And so, no empirical predictions or policy implications follow. The empirical evidence
shows that financial crises are preceded by credit booms and are related to the business cycle, and that
agents are prone to run when public information arrives forecasting a recession. The link between the
preceding credit boom and the business cycle provides the structure for belief formation.
Economists have attempted to address the issue of belief formation in the Diamond and Dybvig model
(and other similar models). Using the global games approach of Carlsson and van Damme (1993), if some
noise and asymmetric information are added to the model the multiplicity can be eliminated or reduced.
If each depositor privately observes a signal about the future value of the banks' assets, then the
equilibrium can be unique if their private signals about the banks' assets are sufficiently accurate. In this
way, the belief coordination problem can be linked to economic fundamentals. There is still a threshold
effect, so a crisis is a distinct event. ¹0 Coordination games can generate large changes in agents'
behavior without large changes in economic fundamentals. Agents change their beliefs about the
actions of other agents and this can have a large effect. This is a general statement which applies to
many phenomena, as long as they can be modeled as a coordination game, where the payoff to any one
agent depends on the actions of other agents.
10
The important papers are Morris and Shin (2001) and Goldstein and Pauzner (2005). The multiplicity of equilibria
can be eliminated in other ways; see, e.g., Postlewaite and Vives (1987).
13 正常な状態とは異なる。Diamond and Dybvig モデルでは、「小さな」危機は存在しない。危機なしと危機ありの2つの結果がある。危機は、別個の、非常に異なる、イベントである。これは、「危機」と呼ぶことができる明確な出来事があり、それは明らかに不況やAnna Schwartzの擬似金融危機よりもはるかに悪い結果であるという経験的証拠と一致している。このモデルは説得力のある設定を行い、結果が通常の状態とは大きく異なり、危機が発生する可能性があることを示す。これは、上に述べた2つの本質的な点を示した最初のモデルであった。この意味で、金融危機の首尾一貫した姿を示している。しかし、危機の理論としては、Diamond and Dybvig モデルが完全に満足できるものではありません。我々が説明したい現象、すなわち、なぜ信頼が失われるのか、それそのものが説明されていないのです。それは "黒点 "です。つまり、各エージェントは、"黒点 "を観測すると他のエージェントが走ると信じているのです。調整装置が "sunspots "と呼ばれているが、これは、このモデルで起こりうる複数の均衡の名前に過ぎない。なぜ経済がある均衡から別の均衡に切り替わるのか、その説明はない。特に厄介なのは、信念の調整の問題である。なぜ突然、暴走が起こるのか、その説明がない。そのため、経験則に基づく予測や政策的な示唆は得られない。実証的には、金融危機の前には信用ブームがあり、景気循環と関係があること、不況を予想する公的情報が届くとエージェントがランしやすくなることが示されています。先行する信用ブームと景気循環の関連は、信念形成の構造を提供する。Diamond and Dybvigモデル（および他の類似モデル）において、経済学者は信念形成の問題を扱おうとしている。Carlsson and van Damme (1993)のグローバル・ゲーム・アプローチにより、ノイズと非対称情報をモデルに加えると、多重性を排除または低減することができる。各預金者が銀行の資産の将来価値に関するシグナルを私的に観測している場合、銀行の資産に関する私的シグナルが十分に正確であれば、均衡は一意的となり得る。このように、信念調整問題は経済のファンダメンタルズと連動させることができる。それでもなお閾値効果があるため、危機は別個の出来事である。調整ゲームは、ファンダメンタルズを大きく変化させることなく、エージェントの行動を大きく変化させることができる。エージェントは他のエージェントの行動に関する信念を変化させ、これが大きな影響を与えることがある。これは、あるエージェントのペイオフが他のエージェントの行動に依存する協調ゲームとしてモデル化できる限り、多くの現象に適用される一般論である。
10 重要な論文は、Morris and Shin (2001) と Goldstein and Pauzner (2005)である。均衡の多重性は他の方法で排除することができる；例えば、Postlewaite and Vives (1987)を参照。
14:
21
It is important to note that this is a purely formal fix-up to a vexing problem arising in the Diamond and
Dybvig model. It can't be tested; no one has every articulated the nature of the private information that
bank debtholders might realistically have learned. Kelley and Ó Gráda (2000) and Ó Gráda and White
(2003) study the details of who ran on the Emigrant Industrial Savings Bank in 1854 1nd 1857. It is hard
to see what the nature or role of the alleged private information.
There are still problems. First, the issue of belief coordination only arises for some forms of bank
money. Demand deposits are claims on a common pool of assets the bank's portfolio of loans, the
case where belief coordination arises as a problem. Other forms of bank money can differ from the
Diamond and Dybvig model in important ways. There may be a maturity date on the claim, even if it is a
short maturity, and there may be no common pool problem. If a "depositor" does not have a claim on a
common pool of bank assets, then the actions of other depositors are irrelevant; beliefs about other
agents' beliefs then do not matter. Or, if there is no sequential service, no lining up, then claims really
are pari passu. But, financial crises are not just about demand deposits. All forms of bank money are
vulnerable.
Bank money is short-term debt. The critical feature of bank money is that it retains value so that it can
act as a short-term store of value or such that other agents unquestioningly accept it in a transaction,
without suspicion of private information held by the counterparty. Bills of exchange and negotiable
instruments generally are bank money. This includes private bank notes, commercial paper, bankers'
acceptances, money market funds, sale and repurchase agreements, and sight drafts. In fact, the history
and evolution of various forms of bank money is rich and complicated. There are many kinds of bank
money. See, e.g., Usher (1914), DeRosa (2001) and Ferderer (2003). Longer term bank debt that by
design resembles government debt may also be included, that is, securitizations.
Checking accounts have not always been the primary form of bank money, and even today checks are
being replaced by ATM machines and on-line banking. See Quinn and Roberds (2008). The issue of
whether all forms of bank money are vulnerable to runs was brought to the fore by the recent crisis. The
recent financial crisis was not a case of household depositors running on banks. It involved firms,
financial and nonfinancial, foreign and domestic, running on shadow banks in the repo and asset-backed
commercial paper markets. And, even this type of wholesale run is not new. See Quinn and Roberds
(2012) and Schnabel and Shin (2004) who study a run in the wholesale market in Amsterdam in 1763.
And, see Flandreau and Ugolini (2011) on the Overend-Gurney Panic of 1866 in England. It seems clear
that runs have occurred under a variety of bank money forms.
これは、Diamond and Dybvigモデルで生じた厄介な問題を、純粋に形式的に修正したものであることに注意することが重要である。銀行債の保有者が現実的に知り得た私的情報の性質について、誰も明確にしていないのである。Kelley and Ó Gráda (2000) and Ó Gráda and White (2003) は、1854年と1857年のEmigrant Industrial Savings Bankの経営者の詳細について研究している。私的情報とされるものがどのような性質や役割を持つのかが見えにくい。問題が残っている。まず、信念の調整の問題は、ある種の銀行貨幣の場合にのみ生じる。要求払い預金は、銀行の融資ポートフォリオという共通の資産プールに対する請求権であり、信念の調整が問題となるケースである。その他の銀行貨幣の形態は、Diamond and Dybvigモデルとは重要な点で異なる可能性がある。たとえ満期が短くても、債権に満期がある場合があり、共通プール問題がない場合がある。預金者」が銀行資産の共通プールに対する債権を持っていない場合、他の預金者の行動は無関係であり、他のエージェントの信念に関する信念は重要でない。あるいは、順次サービスや整列がない場合、債権は本当に同順位である。しかし、金融危機は要求払預金だけの問題ではない。すべての銀行貨幣は、脆弱である。銀行貨幣は、短期債務である。銀行貨幣の重要な特徴は、短期的な価値貯蔵として機能するように、あるいは取引相手が持つ個人情報を疑わずに、他のエージェントが取引において疑うことなく受け入れるように、価値を保持することである。為替手形や譲渡可能金融商品は、一般に銀行貨幣である。これには、民間銀行券、コマーシャルペーパー、銀行引受証、マネーマーケット・ファンド、売戻契約、サイトドラフトなどが含まれる。実は、様々な形態の銀行貨幣の歴史と進化は、豊かで複雑である。銀行貨幣には多くの種類がある。例えば，Usher（1914），DeRosa（2001），Ferderer（2003）などを参照。また、設計上、政府債務に類似した長期銀行債務、すなわち証券化も含まれる場合がある。当座預金は常に銀行貨幣の主要な形態であったわけではなく、今日でも小切手はATM機やオンライン・バンキングに取って代わられつつある。Quinn and Roberds (2008)を参照。銀行貨幣のすべての形態がランに対して脆弱であるかという問題は、今回の危機で前面に押し出された。今回の金融危機は、家計の預金者が銀行に逃げ込んだというケースではない。金融機関、非金融機関、国内外を問わず、企業がレポや資産担保コマーシャルペーパー市場でシャドウバンクに走ったのである。そして、このようなホールセール・ランも新しいものではない。Quinn and Roberds (2012)やSchnabel and Shin (2004)は、1763年にアムステルダムの卸売市場で発生した取り付け騒ぎを研究している。また、Flandreau and Ugolini (2011)は、1866年のイギリスのOverend-Gurney Panicを研究している。このように、様々な銀行貨幣の形態でパニックが発生していることは明らかであろう。
15:
Another special feature of the Diamond and Dybvig model is the fact that agents do not actually meet and trade, so there are no prices in the model. In the model, terms are set on the bank contracts initially and there are no subsequent prices because there is no subsequent trading among agents. In reality there are two complications. First, with many forms of bank money, including demand deposits and private banknotes, agents directly transact. One agent meets and, for example, writes a check to another agent in exchange for goods. Second, other forms of bank money have maturities; agents do not have the contractual right to withdraw any time.
Diamond and Dybvigモデルのもう一つの特徴は、エージェントが実際に会って取引をしないので、モデルには価格が存在しないことです。このモデルでは、最初に銀行契約の条件が設定され、その後のエージェント間の取引はないため、その後の価格は存在しません。現実には、2つの複雑な問題があります。第一に、要求払預金や民間銀行券など、多くの形態の銀行貨幣では、エージェントが直接取引を行う。あるエージェントが他のエージェントに会って、例えば、商品と交換するために小切手を書く。第二に、他の銀行貨幣は満期があり、エージェントはいつでも引き出せるという契約上の権利を持っていない。
16:
14
before there is a central bank, say during the National Banking Era in the U.S. The run starts-for some
reason, time passes, and then agents no longer want to run. Somehow agents' anxiety is assuaged, their
beliefs are revised. But, we don't know how this happens.¹4 If the government or the central bank takes
actions, then agents may revise their beliefs about whatever it was that caused them to run to start
with. The details of what this means and how it happens are unclear. Before the Federal Reserve
System was in existence, this puzzle is clearer. A run would start, usually in New York City, and banks
would suspend convertibility. What happened during the period of suspension that allowed bank to
resume convertibility? A model which can explain how a "loss of confidence" occurs needs also to
explain how confidence is recovered. Clearly, a model with multiple equilibria as the "explanation" for a
crisis has difficulties here. ¹5
15
17
The Diamond and Dybvig setting is compelling. Private agents cannot produce debt that is invulnerable
to runs. Only long-term private assets are available to back bank debt, which is needed to facilitate
shorter-term transactions that some agents need to make to smooth consumption. But, the bank debt
is vulnerable. And a crisis in Diamond and Dybvig is a distinct event. Building on Diamond and Dybvig
requires a model in which a state of the world occurs causing everyone to run. Clearly, there is much
work to be done. Incorporating credit booms into a crisis theory, explaining why there is an association
between crises and prolonged recoveries, and explaining how a crisis ends, are all open questions.
16
5. Bank Debt
Let's take a step back and ask a general question: why is bank debt used for transactions? Agents could
issue their own money. Or firms could issue money. In principle, the "money" could be equity or debt,
or indeed, any security. Many such securities are traded in markets that are often described as "liquid."
So, a basic question is why bank debt is used as money. Why banks? And why debt?
These questions are related to the notion of "liquidity," a term that is used in different ways in the
economics literature. A central contribution of Diamond and Dybvig is their notion of "liquidity" as
consumption smoothing. But, there is another notion of liquidity, a quite natural one first articulated by
14
We know that the clearinghouses acted during crises, but we do not know how agents' beliefs were revised in
response. We just know that eventually suspension of convertibility was lifted.
That is, a "reverse" sunspot just compounds the problem of a lack of an explanation.
15
16
There are other models of runs, as well. Diamond and Rajan (2001) show a model of bank fragility that is
different than Diamond and Dybvig. It connects the asset side of banks to the liability side more specifically,
showing that a kind of fragility is required, and displays a collective action problem. Another interesting example is
Rochet and Vives (2004).
SOME REFLECTIONS ON THE RECENT FINANCIAL CRISIS
Gary B. Gorton
Working Paper 18397 http://www.nber.org/papers/w18397 NATIONAL BUREAU OF ECONOMIC RESEARCH 1050 Massachusetts Avenue Cambridge, MA 02138
https://www.nber.org/system/files/working_papers/w18397/w18397.pdf
September 2012
4. Crisis Theory
11
Theoretically, a financial crisis is defined by two essential points. First, a crisis is a singular event. It is a rending, a sundering, or a rupturing, of the normal state of affairs in money markets. A financial crisis is not the worst outcome on a continuum of bad events. There is no continuum in an important sense. There are booms and recessions, and then here are crises. A crisis is a distinct event. Something happens to make a crisis fundamentally different from the usual economic downturn. Second, while each crisis has important unique features, crises have a common root cause. There is a structural feature of bank debt that makes the debt vulnerable to runs. And the bank debt in question is not just demand deposits. Financial crises are always about bank runs. The bank runs either occur or would have occurred had the government or central bank not intervened or been expected to intervene. The first point says that a “crisis” is not simply a particularly “bad state” of the world. A crisis is fundamentally different, a different regime. There are normal non-crisis states and there is an extraordinary crisis state. This is why Anna Schwartz (2007) said that “a decline in asset prices of equity stocks, real estate, commodities; depreciation of the exchange value of a national currency; financial distress of a large non-financial firm, a large municipality, a financial industry, or sovereign debtors—are pseudo-financial crises” (p. 245). They may be bad events, wealth may be destroyed or cleanup costs high, but they are not crises. A financial crisis is a systemic event. The entire financial system is engulfed. The failure of a large firm or problems in one sector, e.g., savings and loans or the auto industry, are not crises in this sense. Financial crises repeatedly occur in market economies. The second point is that there is a reason for this. There is a root cause. Agents in the economy need private money to transact. But, this money is vulnerable to runs. Bank runs are crises. Financial crises are caused by bank runs. The root of the financial crisis problem was elegantly identified by Diamond and Dybvig (1983). 9 Diamond and Dybvig studied a setting where banks must use long-term collateral to back demand deposits. Agents need the demand deposits because of potential shorter-term liquidity needs. The investments are “long” in the sense that if they are liquidated early there is a very low return. “Long” also means relative to the required frequency of agents’ transactions for consumption or other shortterm needs. The agents need demand deposits to smooth consumption, which is uncertain as some 9There is a large literature on the Diamond and Dybvig model, many extensions and discussions, but I will, for the most part, not go into this literature.
agents may want to consume early. An essential feature of the model is that the interest rate offered on the demand deposits to achieve this smoothing is such that if all agents want to consume early (by withdrawing from the bank), then the bank cannot satisfy these demands. This is the critical fragility in the economy. A very important point is that there is no way around this basic horizon problem in any market economy. People eat lunch every day, but it takes a long time to build a factory and produce output. People need to pay for their lunch before the output is realized. This timing is fundamental. Bank debt used for transactions can only be backed by these long investments (which have a low return if liquidated early). The private sector cannot produce riskless assets. These basic facts mean that financial intermediaries will always be involved in “maturity transformation,” a term which just restates this fact. “Maturity transformation” is not a choice. It can’t be regulated away. It is inherent in any economy which produces private bank money, that is, any market economy. It is a fundamental fact. Bank money can only be backed by longer-term investments. As we will see later, agents in the economy will strive mightily to design bank debt to overcome this problem. But, without the government, bank debt will always be vulnerable. Uncertainty about consumption timing is a risk the agents want to shed using bank debt. The problem of long-term collateral backing bank debt is a necessary but not a sufficient condition for crises to occur. To get a crisis—a bank run, Diamond and Dybvig introduce a source of uncertainty that is quite special. It is the uncertainty that each individual bank depositor faces about the actions of other deposit holders. Depositors care about the actions of other depositors if there is a common pool of assets on which they all have pari passu claims—the bank’s assets-- but the claims are honored sequentially (so they are not in fact pari passu). Note that the assumption of sequential service means that the payout of the bank to an individual depositor depends on the actions of the other depositors. How much a depositor gets back depends on his place in the line. In this setting, depositors may run if they think other depositors are going to run. Each depositor has an incentive to be first in line to withdraw at the bank if he believes that other depositors are going to line up. Beliefs about other depositors’ beliefs must depend on something and in Diamond and Dybvig beliefs are coordinated by an extraneous random signal, a sunspot. The bank run, due to the beliefs coordination problem, displays the second essential condition of the definition of a crisis, discussed above. A run in the Diamond and Dybvig model is fundamentally
different from the normal state of affairs. There are no “small” crises in the Diamond and Dybvig model. There are two outcomes: no crisis and crisis. The crisis is a distinct, very different, event. This is consistent with the empirical evidence that there are distinct events that can be called “crises” and which are clearly much worse outcomes than recessions or Anna Schwartz’s pseudo-financial crises. The model lays out a convincing setting and shows that the outcome can be very different than the normal state of affairs, a run can occur—a crisis. This was the first model that displayed the two essential points articulated above. In this sense, it provides a coherent picture of a financial crisis. But, as a theory of crises, the Diamond and Dybvig model is not completely satisfactory. The very phenomenon we want to explain, why there is a loss of confidence, is not explained – it is “sunspots.” That is, each agent believes that the other agents will run when they observe “sunspots.” While the coordination device is called “sunspots,” this is just a name for the multiple equilibria that can occur in the model. There is no explanation for why the economy switches from one equilibrium to another. The issue of belief coordination is especially troublesome. There is no explanation for why a run would suddenly occur. And so, no empirical predictions or policy implications follow. The empirical evidence shows that financial crises are preceded by credit booms and are related to the business cycle, and that agents are prone to run when public information arrives forecasting a recession. The link between the preceding credit boom and the business cycle provides the structure for belief formation. Economists have attempted to address the issue of belief formation in the Diamond and Dybvig model (and other similar models). Using the global games approach of Carlsson and van Damme (1993), if some noise and asymmetric information are added to the model the multiplicity can be eliminated or reduced. If each depositor privately observes a signal about the future value of the banks’ assets, then the equilibrium can be unique if their private signals about the banks’ assets are sufficiently accurate. In this way, the belief coordination problem can be linked to economic fundamentals. There is still a threshold effect, so a crisis is a distinct event. 10 Coordination games can generate large changes in agents’ behavior without large changes in economic fundamentals. Agents change their beliefs about the actions of other agents and this can have a large effect. This is a general statement which applies to many phenomena, as long as they can be modeled as a coordination game, where the payoff to any one agent depends on the actions of other agents.
10 The important papers are Morris and Shin (2001) and Goldstein and Pauzner (2005). The multiplicity of equilibria can be eliminated in other ways; see, e.g., Postlewaite and Vives (1987).
It is important to note that this is a purely formal fix-up to a vexing problem arising in the Diamond and Dybvig model. It can’t be tested; no one has every articulated the nature of the private information that bank debtholders might realistically have learned. Kelley and Ó Gráda (2000) and Ó Gráda and White (2003) study the details of who ran on the Emigrant Industrial Savings Bank in 1854 1nd 1857. It is hard to see what the nature or role of the alleged private information. There are still problems. First, the issue of belief coordination only arises for some forms of bank money. Demand deposits are claims on a common pool of assets – the bank’s portfolio of loans, the case where belief coordination arises as a problem. Other forms of bank money can differ from the Diamond and Dybvig model in important ways. There may be a maturity date on the claim, even if it is a short maturity, and there may be no common pool problem. If a “depositor” does not have a claim on a common pool of bank assets, then the actions of other depositors are irrelevant; beliefs about other agents’ beliefs then do not matter. Or, if there is no sequential service, no lining up, then claims really are pari passu. But, financial crises are not just about demand deposits. All forms of bank money are vulnerable. Bank money is short-term debt. The critical feature of bank money is that it retains value so that it can act as a short-term store of value or such that other agents unquestioningly accept it in a transaction, without suspicion of private information held by the counterparty. Bills of exchange and negotiable instruments generally are bank money. This includes private bank notes, commercial paper, bankers’ acceptances, money market funds, sale and repurchase agreements, and sight drafts. In fact, the history and evolution of various forms of bank money is rich and complicated. There are many kinds of bank money. See, e.g., Usher (1914), DeRosa (2001) and Ferderer (2003). Longer term bank debt that by design resembles government debt may also be included, that is, securitizations. Checking accounts have not always been the primary form of bank money, and even today checks are being replaced by ATM machines and on-line banking. See Quinn and Roberds (2008). The issue of whether all forms of bank money are vulnerable to runs was brought to the fore by the recent crisis. The recent financial crisis was not a case of household depositors running on banks. It involved firms, financial and nonfinancial, foreign and domestic, running on shadow banks in the repo and asset-backed commercial paper markets. And, even this type of wholesale run is not new. See Quinn and Roberds (2012) and Schnabel and Shin (2004) who study a run in the wholesale market in Amsterdam in 1763. And, see Flandreau and Ugolini (2011) on the Overend-Gurney Panic of 1866 in England. It seems clear that runs have occurred under a variety of bank money forms.
One of the most important forms of bank money historically was private bank notes. Private bank notes were issued by banks in many countries. Schuler (1992) finds sixty cases of such free banking in history. In some cases these notes were claims on a common pool of assets and in some cases they were not. In the U.S. under state free banking laws banks were required to back their notes with state bonds. In the case of a bank failure—an inability to honor requests for cash from noteholders—the state bonds would be sold (by the state government) and the note holders paid off pro rata. Note holders were paid off pro rata, so there was no common pool problem. Yet, there was a run on banks (banknotes and deposits) during the Panic of 1857. The recent financial crisis centered on sale and repurchase agreements (repo). 11 In a sale and repurchase agreement (a repo) one party lends/deposits money typically overnight at interest and this depositor receives a specific bond as collateral from the bank borrower. The lender/depositor must return the collateral at the maturity of the repo contract. There is no common pool of assets upon which the “depositor” has a claim. 12 If the borrower/bank fails, then the lender/depositor can unilaterally terminate the contract and sell the collateral. Of course, a depositor need not renew the loan, and will not if there are concerns about the joint event of (1) the solvency of the bank and (2) the value of the collateral. Repo and free banknotes are two examples of bank money where there is no common pool problem. Demand deposits and asset-backed commercial paper are examples where there is a common pool problem; these forms of bank debt are backed by a common portfolio of assets. We observe runs on both forms of bank money, suggesting that the common pool problem is not the inherent vulnerability. Another special feature of the Diamond and Dybvig model is the fact that agents do not actually meet and trade, so there are no prices in the model. 13 In the model, terms are set on the bank contracts initially and there are no subsequent prices because there is no subsequent trading among agents. In reality there are two complications. First, with many forms of bank money, including demand deposits and private banknotes, agents directly transact. One agent meets and, for example, writes a check to another agent in exchange for goods. Second, other forms of bank money have maturities; agents do not have the contractual right to withdraw any time.
11 See Gorton (2010) and Gorton and Metrick (2012). 12 Although see Martin, Skie, and von Thadden (2010). 13 Jacklin (1987) discusses some of the trading restrictions in the Diamond and Dybvig model.
In the Diamond and Dybvig model, once the agents have deposited money in the bank, there are no later transactions between depositing agents in the model. Some agents, perhaps all agents, go to the bank to withdraw prior to the realization of the investment payoffs. But, they do not transact directly with each other at some price, the price of goods in terms of the bank money. So, there are no prices in the model at the date when agents form beliefs about the actions of other agents. But, in reality, agents do meet and trade goods or services for bank money. Before the U.S. Civil War when agents transacted they used private bank notes, the liabilities of banks denominated as money (i.e., one dollar bills, five dollar bills, etc.). An agent would go to the store and offer to buy goods with these notes. But, these notes did not trade at par. There was an exchange rate between the notes and gold. That is, there was a price. And prices contain information. It could be that one agent writes a check to another agent, for example. In this case, the relative price of the bank money in terms of goods plays a role, as in other markets. With demand deposits the price is usually par, except in a crisis when checks were discounted. There are two cases. First, suppose there is a common pool problem. What is the effect of prices? Atkeson (2001) raises this point. In this case of the coordination problem, it is not clear that the multiplicity of equilibria disappears when prices are introduced. Economists have tried to address this issue and in related settings have found that the multiple equilibria remain in the presence of prices. See, e.g., Angeletos and Werning (2006) and Hellwig, Mukherji and Tsyvinski (2006). We would like to have a detailed theory of how beliefs are formed. This is an ongoing area of research. The second case occurs where there is no common pool problem. There is no common pool problem in repo, for example. In a repo transaction there is a depositor who lends money and a bank borrower. The depositor receives interest on the loan, which is usually overnight. And the borrower delivers collateral to the depositor, which must be returned when the transaction matures. The collateral is sometimes “haircut,” which means that the depositor lends less money than the market value of the collateral provided. For example, $90 million is lent and the collateral is worth $100 million at market prices. In repo, haircuts and interest rates depend on the identity of the counterparty if the collateral is private bonds. Even in an over-the-counter market, at any moment, agents in the market (eventually) know these prices. These prices are formed somehow and are related to agents’ beliefs. Another issue concerns how a crisis ends. If agents run on banks because they believe other agents will run, or because fundamentals have deteriorated, how does the crisis end? It is clearest to think of this before there is a central bank, say during the National Banking Era in the U.S. The run starts—for some reason, time passes, and then agents no longer want to run. Somehow agents’ anxiety is assuaged, their beliefs are revised. But, we don’t know how this happens. 14 If the government or the central bank takes actions, then agents may revise their beliefs about whatever it was that caused them to run to start with. The details of what this means and how it happens are unclear. Before the Federal Reserve System was in existence, this puzzle is clearer. A run would start, usually in New York City, and banks would suspend convertibility. What happened during the period of suspension that allowed bank to resume convertibility? A model which can explain how a “loss of confidence” occurs needs also to explain how confidence is recovered. Clearly, a model with multiple equilibria as the “explanation” for a crisis has difficulties here. 15 The Diamond and Dybvig setting is compelling. Private agents cannot produce debt that is invulnerable to runs. Only long-term private assets are available to back bank debt, which is needed to facilitate shorter-term transactions that some agents need to make to smooth consumption. But, the bank debt is vulnerable. And a crisis in Diamond and Dybvig is a distinct event. Building on Diamond and Dybvig requires a model in which a state of the world occurs causing everyone to run. 16 Clearly, there is much work to be done. Incorporating credit booms into a crisis theory, explaining why there is an association between crises and prolonged recoveries, and explaining how a crisis ends, are all open questions.
0 件のコメント:
コメントを投稿