Financial Instability

Abstract

In this chapter, the central bank is put aside and we review simple models of financial instability, which will be the basis for the subsequent chapter to explain the role of the central bank as lender of last resort. We first recall that financial instability is mostly triggered by a negative shock on asset prices, and thereby on the solvency of debtors, which in turn worsens access to credit and can set in motion a liquidity crisis with vicious circles. We develop the concepts of solvency “conditional” and “unconditional” on liquidity: a decline in asset prices can lead an unconditionally solvent debtor to become only conditionally solvent, such that sufficient liquidity becomes decisive for preventing its default. We then apply these concepts to the stability of bank funding and introduce the problem of bank runs. We subsequently show why asset liquidity in a dealer market deteriorates during a financial crisis (increased volatility and uncertainty increase the required bid-ask spread); how asymmetric information can lead to a freeze of credit markets in a simple adverse selection model; how declining and more volatile asset prices drive increases of haircut, and how these can force fire sales and defaults of borrowers. We finally discuss the interaction between these various crisis channels.

5.1 Liquidity, Asset Prices, and Default

Financial crises always encompass liquidity crisis, and liquidity crises typically have a funding liquidity and a market liquidity component. In a funding liquidity crisis, the problem is credit availability: debtors cannot obtain credit to roll over their liabilities, as money and capital markets freeze. In a market liquidity crisis, bid-ask spreads and asset fire-sale discounts increase, reducing the total liquidity obtainable through asset liquidation.

Liquidity crises may lead to the inability of debtors to fulfil their contractual obligations and hence to their default, with additional economic damage. Financial crises are typically triggered by strong downward revisions of asset values (Bagehot 1873, Kindleberger and Aliber 2011). Asset price changes can be driven by any major unexpected news the economy, e.g. a natural disaster, a pandemic, an unexpected change of government and of economic policies, the outbreak of a war, the rise of a new major technology, a strong re-assessment of the inflation and thus central bank interest rate outlook, a strong collective re-assessment of the prospects of an asset class (e.g. burst of a housing bubble) etc. can all heavily impact on few, or various asset classes. For example, the outbreak of Covid-19 and the uncertain perspective of a return to the pre-Covid19 normal has reduced the value of assets owned by the international travel and tourist industry (cruise ships, airplanes, hotels, etc.).

Strong declines of asset values have various negative effects on economic agents. Solvency declines, which undermines the ability to access funding sources and the willingness and the ability to undertake risky projects. In the case of banks, a decline in solvency puts at risk compliance with capital adequacy regulations, adding urgency to deleveraging through the shrinking of lending or through asset fire sales. In the case of households, lower wealth reduces consumption, which will have recessionary effects.

The mechanisms of liquidity crises have been similar across time. Already Thornton (1802) noticed the problem of liquidity hoarding and bank runs, and how they relate to a lack of trust.

Bagehot (1873, Chapter VI “Why Lombard Street Is Often Very Dull, and Sometimes Extremely Excited”) argues that while liquidity crises are caused by heterogenous exogenous events, their mechanics, once having been triggered, are similar:

Any sudden event which creates a great demand for actual cash may cause, and will tend to cause, a panic in a country where cash is much economised, and where debts payable on demand are large. …. Such accidental events are of the most various nature: a bad harvest, an apprehension of foreign invasion, the sudden failure of a great firm which everybody trusted, and many other similar events, have all caused a sudden demand for cash. And some writers have endeavoured to classify panics according to the nature of the particular accidents producing them. But little, however, is, I believe, to be gained by such classifications. There is little difference in the effect of one accident and another upon our credit system. We must be prepared for all of them, and we must prepare for all of them in the same way—by keeping a large cash reserve.

Understanding the logic of liquidity crises is a precondition for understanding the role of the central bank in stopping the escalation of liquidity crises and in addressing their economic consequences.

To illustrate how default probabilities increase in financial crises when assets values decline, consider in Table 5.1 the balance sheet of a leveraged financial or non-financial corporate with ε being a random variable impacting asset values.

Table 5.1 Balance sheet of an indebted firm

Asset values can be thought of as being subject to periodic random shocks ε. Assuming simplistically that ε is N(0, σ_ε) then the probability of default (PD) of a company, in the sense of the probability that its asset values will fall below the value of debt in the next time period, could be estimated as (Φ(·) is the cumulative standard normal distribution):

$$PD = P\left( {E + \varepsilon < 0} \right) = P\left( {A + \varepsilon < D} \right) = \varPhi \left( { - \frac{A - D}{{\sigma_{\varepsilon } }}} \right)$$

This formula for the probability of default is however a strong simplification, since asset values are not normally distributed, variables such as σ_ε are not directly observable, time is continuous, and there is no unique horizon to consider. Moreover, default does not need to occur exactly when A + ε touches D, since default is eventually triggered by illiquidity. Merton’s structural credit model (Merton 1974) is a more sophisticated version of this basic default probability model.

Rating agencies provide comprehensive statistics on how rated debtors performed. For example, Standard & Poor’s regularly publishes a statistical default study (e.g. S&P 2020). According to Table 5.2, the following annual default rates applied in the period 1981 to 2019.

Table 5.2 Annual default probability of rated debtors according to Standard and Poor’s S&P 2020

The differences between good years (minimum) and bad years (maximum) are significant. The time series of annual default shows a sort of financial cycle with peaks in default frequency around 1991, 2001, and 2008. Creditors do not only care about the probability of default, but also about what losses occur in case of default. If all debtors rank pari passu, and if default occurs exactly when A + ε = D, then the “Loss-given default” (LGD, = 1 – recovery ratio) should be zero (the recovery ratio should be 1). However, evidence collected by rating agencies suggests that LGDs are on average around 50% (depending also on the debt instrument). This can have two explanations: (i) default often occurs only when A is already clearly below D (i.e. creditors did not realize that the company had negative equity, or there was no debt redemption date); (ii) the default event itself is costly, as default implies that organisational and human capital is destroyed and specific assets are liquidated at fire-sale prices (e.g. a sophisticated machine is sold at its raw material value, minus the costs of removing, transporting and dismantling the machine).

The corporate finance literature provides estimates of the costs of default between 10 and 44% (e.g. Glover 2016; Davydenko et al. 2012). The cost of default is one key reason for central banks trying to prevent defaults of sound companies due to illiquidity. We will see below that in settings of asymmetric information which are typical for the high uncertainty prevailing in financial crisis, credit and asset markets can break down such that illiquidity and default can occur even for firms which are solvent and viable.

The “credit channel” literature has analysed since the 1980s how high credit riskiness and low equity have been identified for a while as an issue for monetary policy transmission. According to this literature, low equity implies higher agency costs in the lending between banks and corporates. Lower bank equity implies higher agency costs between holders of bank liabilities and banks. Higher agency costs result from a deterioration of the alignment of incentives between debt and equity owners when equity levels fall (e.g. Holmström and Tirole 1997). In addition, debtors with insufficient equity will attempt to restore their creditworthiness by aiming at deleveraging, causing economic contraction and deflationary tendencies.

To sum up, it is important to distinguish the following four key concepts for troubled debtor:

• defaulted: a missed payment obligation, possibly defined also by the number of days the payment day passed;

• illiquid: inability to identify money for fulfilling (forthcoming) payment obligations;

• insolvent: debt exceeds assets, and therefore equity is negative;

• over-indebtedness: Relating to insolvency but less linked to a strict threshold. Overindebted companies may still have positive capital, but insufficient capital to grant them healthy and sufficiently cheap market access, so that in the medium-term insolvency/illiquidity looms.

An indebted corporate can default despite being solvent. It may be illiquid because a systemic liquidity crisis situation made all possible lenders stop lending, and it has to refinance a maturing loan or debt instrument. A corporate may be insolvent without yet defaulting, because no debt payment is due. Eventually, a corporate which is clearly insolvent will also end up being illiquid and default because it does not make sense for creditors to give fresh loans to an insolvent debtor.

5.2 Conditional and Unconditional Insolvency, and Bank Runs

Debtors that are solvent conditional on the access to funding may become insolvent if funding constraints force them to undertake fire sales of some of their assets at some specified horizon. This relates to problems explained in Sect. 1.2 that fair book values of assets are normally higher than their (short-term) liquidation value. The following assumes that the fair value of the assets of an indebted company is 1, and that these assets are ordered from the most liquid to the least liquid (x-axis). Assume further that L(x) is the liquidity generated by liquidating at some time horizon, say one week, the share x of assets that is most liquid. It follows that L(0) = 0 (if no assets are sold, no liquidity is generated), L(1) ≤ 1 (a sale of all assets provides at the very maximum their book value), dL(x)/dx ≥ 0 (the liquidity generated cannot decrease with more assets being sold) and d²L(x)/d²x ≤ 0 (each new unit of asset sold does not provide more liquidity than the previous one, as assets were ranked from the most to the least liquid). In other words, the liquidity generation can be described by a concave monotonously non-decreasing function of x. We can also define similarly the fire-sale loss function F(x) = x – L(x) which indicates the fire-sale losses generated by selling a share of assets x, starting with the most liquid assets. The function f(x) = dF(x)/dx is the marginal fire- sales loss function indicating the size of fire-sale losses resulting from selling the asset ranked at x, and q(x) = dL(x)/dx = 1 – f(x) is the marginal liquidity generated by selling the asset ranked x.

Figure 5.1 illustrates the simple case when f(x) = x, and therefore q(x) = 1 – x, F(x) = x²/2 and L(x) = x – x²/2. For example, if the company needs to generate a liquidity of L₁ in order to meet a due payment to its bondholders, then it needs to solve the quadratic equation L₁ = x –x²/2 or x²/2 – x + L₁ = 0. The relevant solution to this problem is x = 1 – ($\sqrt {1 - 2L_{1} } )$. For example, if L₁ = 0.4, then the required fire sales are x = 0.55. Figure 5.1a draws the marginal fire sale loss function f(x) for the assets x ordered from the most liquid to the least liquid. Figure 5.1b shows the total fire-sale loss F(x) and total liquidity generated L(x) for f(x) = x, again for liquidity-ordered assets. Figure 5.1c shows the necessary fire sales of assets as a function of the liquidity to be generated, i.e. x = L⁻¹(L₁). Last but not least, Fig. 5.1d illustrates the concrete case when the firm needs to generate the cash flow of 0.4, whereby a share of 0.55 of the assets need to be sold.

Fig. 5.1

Liquidity generation and losses due to fire sales—example with F(x) = x

Assume the company has equity E, and has to generate through fire sales a cash flow at a certain time horizon T. For instance, because it needs to repay a debt instrument at maturity and is unable to roll it over or find another form of financing. For every time horizon T the function L_T(x) is non-decreasing and therefore invertible, so that we can write the inverse function x_T(L), describing the share of assets that must be liquidated for generating liquidity of a specified size L. Then, the company is solvent conditional on the need to generate a cash flow L₁ at time horizon T = 1 if, and only if, the fire-sales losses caused by the liquidation are lower than the equity of the company:

$${\text{F}}_{\text{1}} \left( {{\text{x}}_{\text{1}} \left( {{\text{L}}_{ 1} } \right)} \right) \, < {\text{ E}}$$

We expect that the higher the time horizon, the lower the losses: if T₁ > T₂, then for every x, we expect that f_T2(x) ≥ f_T1(x). To distinguish fire-sale losses due to time constraints from fire-sale losses due to asset specificity, one could say that the marginal fire-sale losses due to asset specificity (f_AS) are the marginal fire-sale losses without any time pressure, i.e.

$$f_{AS} \left( x \right) = \mathop {\lim }\limits_{T \to \infty } f_{T} \left( x \right)$$

Consequently, the marginal fire-sale losses due exclusively to time pressure (f_T,P) can be defined as the difference between the marginal loss function and the asset-specificity related marginal loss function.

$$f_{T,P} \left( x \right) = f_{T} \left( x \right) - f_{AS} \left( x \right)$$

Figure 5.2 provides an example of marginal fire sale loss functions for the same company at different time horizons.

Fig. 5.2

Marginal fire sales loss curve for alternative liquidation horizons

To ensure funding stability, the bank (or any debtor) should ideally be able to ensure liquidity and solvency at all time horizons. For example, if one debt position of an amount L₁ matures in 1 week, and another of L₂ in four weeks, then both conditions

$${\text{F}}_{{ 1 {\text{Week}}}} \left( {{\text{x}}_{{ 1 {\text{Week}}}} \left( {{\text{L}}_{ 1} } \right)} \right) \, < {\text{ E and F}}_{{ 4 {\text{Weeks}}}} \left( {{\text{x}}_{{ 4 {\text{Weeks}}}} \left( {{\text{L}}_{ 1} + {\text{L}}_{ 2} } \right)} \right) \, < {\text{ E}}$$

should be fulfilled in order to ensure that the debtor is stable and can communicate to its current or potential future creditors that it will be fine anyway (even conditional on no roll-over of funding). Chapter 6 will provide a more precise model in which, for a specific functional form of L(x) and F(x), precise conditions for funding stability will be derived.

For the moment, it is useful to retain that (i) solvent debtors can be sub-classified into those which are solvent regardless of assumptions taken with regards their ability to roll over some debt instruments maturing at some horizon, and those which are solvent only conditional on accessing fresh funding; (ii) the latter may create multiple equilibrium situations, as further explained in chapter 6; (iii) both a negative asset value and a deterioration of asset liquidity can push a lender from being unconditionally solvent into being solvent only subject to funding renewal.

Bank runs have been a major issue at least since the nineteenth century, with particularly devastating episodes in the early 1930s, leading to the general introduction of deposit insurance schemes. More recent runs occurred in the UK (Northern Rock in 2008), and in Greece and Cyprus during the euro sovereign debt crisis. The latter runs mainly materialised through electronic transfers of deposits to accounts with non-domestic euro area banks, i.e. without queues in front of the banks to withdraw cash. Bank runs have been extensively modelled in the economic literature, such as in Diamond and Dybvig (1983) and Rochet and Vives (2004). The particularity of bank runs is their self-fulfilling property: once a run on a bank starts, it can lead to the default of the bank, confirming the individual wisdom of those who were first in the queue to withdraw their money. We will illustrate in chapter 7, with a very simple but powerful strategic bank run model, that a bank can essentially be in three states in terms of stability of its short-term liabilities:

• (a) Funding stability: if there is a single no-run equilibrium. This should apply for unconditionally solvent banks.

• (b) Multiple equilibria: there are two equilibria, one in which depositors run, and one in which they do not run. The depositors’ behaviour can in principle switch from one to the other. This situation arises for solvent banks which are however conditionally insolvent, i.e. conditionally on a run.

• (c) Single run equilibrium: depositors and other short-term investors will run in any case when a bank is unconditionally insolvent.

A switch from state a) to state b) can occur if: (i) asset liquidity deteriorates; and/or (ii) asset values decline, implying a decline of equity. A switch from state a) or b) to state c) will occur if asset values fall such that equity becomes negative. In crisis situations, both factors tend to materialise with higher probability than usual, in particular for banks with more limited solvency and liquidity buffers, or with an unlucky asset concentration, i.e. an asset concentration towards those assets which by bad luck suffer from value losses and a decline in liquidity. This will be taken up in chapter 6 and in particular in the model of Sect. 6.5.

5.3 Illiquidity in Credit and Dealer Markets

In this section we attempt to shed further light on the mechanisms leading to illiquidity in credit and asset markets.

5.3.1 Credit Markets

Information asymmetries can lead to a freeze of credit markets and related economic damage (e.g. Stiglitz and Weiss 1981; Bolton and Freixas 2006). The following basic model of a credit market freeze is based on Flannery (1996). It is assumed that entrepreneurs who would like to borrow are either “Good” or “Bad”, i.e., will repay or not, respectively. The proportion of Good entrepreneurs is g, while the proportion of Bad entrepreneurs is (1 – g). The entrepreneur needs a unit bank loan to finance her project. At the end of the period, Good entrepreneurs’ projects will be worth V_G > 1, which suffices to repay loans, assuming that the lending interest rate was not higher than V_G – 1.

Banks are imperfect in assessing loan applicants. The creditworthiness will be assessed correctly with a probability p. The bank obtains either a signal, $S_{G}$ (good borrower) or a signal $S_{B}$ (bad borrower). If the borrower is actually good, then with probability p > g, a good borrower signal ($S_{G}$) is captured, and the bank may lend. With probability (1 – p), the bad borrower signal ($S_{B}$) is received, and no loan will be provided. If the borrower is actually bad, then the bad borrower signal will be received by the bank with probability p, and the good borrower signal with probability (1 – p). With the help of Bayes’ Law (shown below explicitly only for the first of the four cases), this allows the calculation of the probabilities of all combinations of signal and actual quality of entrepreneurs:

$$P\left( {S_{G} \left| G \right.} \right) = \frac{{P\left( {S_{G} \cap G} \right)}}{P\left( G \right)} \Rightarrow P\left( {S_{G} \cap G} \right) = pg$$

$$P\left( {S_{B} \left| B \right.} \right) = p\left( {1 - g} \right)$$

$$P\left( {S_{G} \left| B \right.} \right) = \left( {1 - p} \right)\left( {1 - g} \right)\text{; }P\left( {S_{B} \left| G \right.} \right) = \left( {1 - p} \right)g$$

Banks are assumed to be competitive and to have zero profits, implying that profits from lending to good borrowers must on average compensate the credit losses due to bad borrowers. Let j* be the interest rate that a bank needs to set to have zero expected profits. As lending occurs only if a good signal is obtained, non-zero pay offs occur only in two out of the four possible cases. The zero-profit condition can therefore be formulated as follows for a bank that itself pays zero interest on its liabilities:

$$j^{*} \cdot gp + \left( { - 1} \right)\left( {1 - g} \right)\left( {1 - p} \right) = 0$$

$$\Rightarrow j^{*} = \frac{{\left( {1 - g} \right)\left( {1 - p} \right)}}{gp}$$

Lending, i.e., an active credit market, will take place as long as V_G – 1 ≥ j*. Otherwise, even good entrepreneurs would make losses and will therefore better not launch their projects. For example, writing j* as j*(g,p): j*(0.5, 0.5) = 100%, j*(0.5, 0.75) = 33%; j*(0.5, 0.9) = 11%; j*(0.7, 0.9) = 4.8%, etc.

The model illustrates three causes for a break-down of credit markets: first, a decline of g, the share of Good entrepreneurs; second, a decrease of V_G, the project return of Good entrepreneurs; third, a decrease of p, the power of the banks’ screening technology. All three effects are associated with a negative economic shock. The model thereby explains why economic shocks trigger credit market crises, and moreover why these effects can be so abrupt: according to the model, a minor further parameter deterioration can make the market collapse, because the critical condition no longer holds. But even before such a complete funding market breakdown, economic deterioration is already felt in the form of an increasing equilibrium lending rate j*. This matters at the zero lower bound, when the central bank can no longer compensate such effects through a lowering of its interest rates.

5.3.2 Dealer Markets

Assume a dealer market in which dealers commit bid and ask prices for some standard quantity q. The bid-ask spread, which measures asset liquidity, typically increases in financial crises. The following simple model takes up basic elements of Kyle (1985) to explain why asset liquidity in a dealer market will deteriorate in a financial crisis. The model assumes that:

• the fair value of the asset, A_t, changes every day according to A_t = A_{t – 1} + ε_t with ε_t being a symmetric random variable with expected value 0.

• Every morning, the market maker sets the bid-ask spread z around his estimate of A_t, which is however A_{t – 1}, as he learns about ε_t only with a one day lag. Therefore the bid and ask price set by the market maker are A_{t – 1} – z/2 and A_{t – 1} + z/2, respectively. Market makers are assumed to provide their services under full competition and to not have operating costs.

• Noise traders are uninformed market participants which trade every day an amount W, equally split between demand and supply. The amount W declines when the bid-ask spread increases (dW/dz < 0).

• Only the insider knows ε_t and thus A_t on day t. Whenever A_t is outside the bid-ask spread, i.e. whenever A_t < A_{t – 1} – z/2 or A_t > A_{t – 1} + z/2, the insider exploits the commitment of the market maker and deals with him. The market maker is assumed to become aware of an insider transaction only once he notes the imbalance in demand and supply has reached q. Then, and assuming that the noise-traders have achieved W(z), the market maker stops quoting for that day and only re-opens t + 1 with a bid-ask spread around A_t.

Bid-ask spreads and trading volumes will reflect the existence of insider information, as captured by $\sigma_{\varepsilon }^{2}$. The competitive market maker will set z such that expected profits are zero. Expected profits have two components: the profit of market makers generated by the noise traders is z·W(z). If $f_{\varepsilon } \left( x \right)$ is the density function of asset value innovations ε_t, then the expected losses of the market maker due to insiders are:

$$\left[ {\mathop \smallint \limits_{ - \infty }^{{ - \frac{z}{2}}} \left( {x + \frac{z}{2}} \right)f_{\varepsilon } \left( x \right)dx + \mathop \smallint \limits_{{\frac{z}{2}}}^{\infty } \left( {x - \frac{z}{2}} \right)f_{\varepsilon } \left( x \right)dx} \right]q$$

The competitive equilibrium bid-ask spread z is the one in which the expected profits of the market maker are zero, with the profits extracted from noise traders compensating exactly the expected losses due to insiders. The model illustrates the empirical pattern that information intensity and price volatility of assets reduce asset liquidity, as measured by bid-ask spread, for example. It thereby also explains the strong increases of bid-ask spreads in financial crises, which tend to be characterised by an intensive news flow and high uncertainty.

5.4 Increasing Haircuts and Margin Calls

Financial exposures are often protected by collateral, also called “margin”, such as in particular in the following three cases: (i) Interbank repo operations (i.e. collateralised interbank lending); (ii) Lending of banks to non-banks: e.g. mortgage loans; loans of banks to corporates, to hedge funds, etc. (iii) derivatives transactions, be they via central clearing counterparties (CCPs), or “over-the counter” (OTC), i.e. directly between counterparties.

The unexpected request of large amounts of additional collateral (“margining”) can trigger liquidity crises and depreciate asset values via forced fire sales because margin requirements limit leverage for investors that use as collateral the assets they invest in. Large marginal calls of counterparties were the eventual trigger of the defaults of Lehman brothers in September 2008. Margin calls of CCPs in March 2020 related to the outbreak of the Covid-19 pandemic also created significant stress on banks, but remained manageable (see Huang and Takats 2020).

In financial crises, margin requirements tend to increase substantially. If the cash investor (or a CCP) wants to maintain the probability of a loss conditional on counterparty default at a certain confidence level, then it needs to increase haircuts whenever volatility increases, liquidity decreases, or the desired protection level increases. Haircuts are often set such as to limit the probability of a loss at a certain confidence level in the scenario of collateral liquidation due to borrower default. The haircut is thus calculated on the basis of the following factors:

• The assumed orderly liquidation time T of the asset, i.e. the liquidation time such that liquidation does not negatively affect market prices.

• The asset price volatility σ at a one-day horizon. If daily price changes are independent of each other, then volatility of price changes over T will be $\sigma \sqrt T$.

• The confidence level for not making a loss. For normally distributed price changes, the confidence level β can be translated into a multiplier of volatility using the inverted cumulative standard normal distribution, i.e. Φ⁻¹(β).

We obtain therefore the following adequate haircut h for a daily price volatility σ, a liquidation horizon of T days, and a confidence level β to avoid a loss in case counterparty default and collateral liquidation:

$$h = \varPhi^{ - 1} \left( \beta \right)\sigma \sqrt T$$

In a financial crisis, haircuts will increase because (i) cash providers may seek a higher confidence level of protection as their capital buffers may have suffered due to the crisis, implying a need to reduce overall risk taking: $\beta \uparrow \Rightarrow \varPhi^{ - 1} \left( \beta \right) \uparrow$; (ii) asset price volatility increases: $\sigma \uparrow$; (iii) it takes longer to liquidate assets without the liquidation having additional price effects: T↑. Increases in haircuts will mean less leverage, and thus, for a given level of capital, the need for fire sales with various potential second round effects.

5.5 Interaction Between Crisis Channels

The various crisis channels described in this chapter interact and create vicious circles of deteriorating solvency, liquidity and default leading potentially to an economic meltdown:

• Asset value declines may lead to immediate insolvency, which normally will lead to an inability to roll over funding, and eventual default.

• Even if immediate insolvency can be avoided, still various liquidity channels set in, like loss of funding stability, increasing funding costs due to increased credit risk premia, increase of haircuts, increase of bid-ask spreads, and decline in the ability to monitor and to overcome the adverse selection problem (all illustrated above in simple models).

• Banks’ and corporates’ funding stress possibly forces them to undertake asset fire sales.

• These fire sales may prevent the immediate default of the bank or corporate. Nevertheless, the firm can have generated losses through fire sales that depleted its equity.

• Defaults and fire sales create further asset value declines and a further increase of asset value uncertainty.

Once lending to the real economy tightens, a recession can occur and will lead to additional losses via renewed asset price declines and impairment of banks’ loan portfolios. The resulting dynamics may call for external circuit breakers, including in particular the central bank. Central banks are not subject to liquidity constraints and can in theory provide unlimited liquidity to banks. Thereby, they can suppress the liquidity part of any vicious crisis circle, making the central bank (unwillingly) the key player deciding on the fate of banks and other leveraged entities.