Tải bản đầy đủ
2 Risk mitigation techniques – the Eurosystem approach

2 Risk mitigation techniques – the Eurosystem approach

Tải bản đầy đủ

278

Bindseil, U. and Papadia, F.

market practice, values collateral daily and has set an symmetric trigger
level of 0.5 per cent, i.e. when the collateral value, after haircuts (see
below), falls below 99.5 per cent of the cash leg, a margin call is triggered.
 Haircuts: in case of counterparty default, the collateral needs to be sold.
This takes some time and, for less liquid markets, a sale in the shortest
possible time may have a negative impact on prices. To ensure that there
are no losses at liquidation, a certain percentage of the collateral value
needs to be deducted when accepting the collateral. This percentage
depends on the price volatility of the relevant asset and on the prospective
liquidation time. The higher the haircuts, the better the protection, but the
higher also the collateral needed for a given amount of liquidity. This
trade-off needs to be addressed by setting a certain confidence level against
losses. The Eurosystem, for instance, sets haircuts to cover 99 per cent of
price changes within the assumed orderly liquidation time of the
respective asset class. Chapter 8 provides the Eurosystem haircuts for
marketable tier one assets. Haircuts increase with maturity, because so
does the volatility of asset prices. In addition, haircuts increase as liquidity
decreases.
 Limits: to avoid concentration, limits may be imposed, which can take
the following form: (i) Limits for exposures to individual counterparties
(e.g. limits to the volume of refinancing provided to a single counterparty). (ii) Limits to the use of specific collateral by single counterparties:
e.g. percentage or absolute limits per issuer or per asset type can be
imposed. For instance, counterparties could be requested to provide not
more than 20 per cent in the form of unsecured bank bonds. (iii) Limits
to the total submitted collateral from one issuer, aggregated over all
counterparties. This is the most demanding limit specification in terms
of implementation, as it requires that the aggregate use of collateral
from any issuer is aggregated and, when testing collateral submission,
counterparties are warned if the relevant issuer is already at its limit. This
specification is also problematic as it makes it impossible for counterparties to know in advance whether a given security will be usable as
collateral.
As the usage of limits always creates some implementation and monitoring
costs and constrains counterparties, it is preferable, when possible, to try to
set the other parameters of the framework to avoid the need for limits. This
is what the Eurosystem has done so far, including the application of different haircuts to different assets. The differentiation of haircuts should also
contribute to reduce concentration risk, avoiding that counterparties have

279

Risk management and market impact of credit operations

Table 7.1 Shares of different types of collateral received by 113 institutions responding to the 2006
ISDA margin survey

Type of collateral

Per cent
of total

Per cent of
total non-cash

Cash
Bonds – total
 Government securities
 Government agency securities
 Supranational bonds
 Covered bonds
Letters of credit
Equities
Metals
Others

72.9%
16.4%
11.8%
4.2%
0.4%
0.0%
2.2%
4.2%
0.2%
1.7%


66.4%
47.8%
17.0%
1.6%
0.0%
8.9%
17.0%
0.8%
6.9%

Source: ISDA. 2006. ‘ISDA Margin Survey 2006’, Memorandum, Table 3.1.

incentives to provide disproportionately one particular type of collateral.
This could happen, in particular, if the central bank would set too lax risk
control measures, thereby making the use of a given collateral type too
attractive (in particular if compared to the conditions in which that asset is
used in private sector transactions).

2.3 Collateral eligibility and risk control measures in inter-bank transactions
As noted by the Basel Committee on the Global Financial System (CGFS
2001), collateral has become one of the most important and widespread risk
mitigation techniques in wholesale financial markets. Collateral is in particular used in: (i) secured lending; (ii) to secure derivatives positions, and
(iii) for payment and settlement purposes (e.g. to create liquidity in a RTGS
system). Regular updates about the use of collateral in inter-bank markets are
provided by ISDA (International Swap and Derivatives Association) documents, like the 2006 ISDA Margin Survey. According to this survey (Table
7.1.), the total estimated collateral received and delivered in 2006 would have
had a value of USD 1,329 trillion. The collateral received by the 113 firms
responding to the survey (which are estimated to cover around 70 per cent of
the market) would have had a composition as indicated in Table 7.1.
Obviously, cash collateral is not suitable for inter-bank (or central bank)
secured lending operations, in which the purpose is just to get cash. The

280

Bindseil, U. and Papadia, F.

high share of cash collateral therefore indicates that secured lending is not
the predominant reason for collateralization. Amongst bonds, Government
securities and, to a lesser extent, Government agencies dominate. Also the
use of equities is not negligible. The 113 respondents also reported in total
109,733 collateral agreements being in place (see ISDA Marginal Survey, 9),
of which 21,889 were bilateral, i.e. created collateralization obligations for
both parties, the rest being unilateral (often reflecting the higher credit
quality of one of the counterparties). The most commonly used collateralization agreements are ISDA Credit Support Annexes, which can be
customized according to the needs of the counterparties. Furthermore, the
report notes that, in 2006, 63 per cent of all exposures created by OTC
derivatives were collateralized.
The ISDA’s Guidelines for Collateral Practitioners7 describe in detail
principles and best practices for collateralization, which are not fundamentally different from those applied by the Eurosystem (see above).
Table 7.2 summarizes a few advices from this document, and checks
whether or not, or in which sense, the Eurosystem practices are consistent
with these advices.
It should also be noted that haircuts in the inter-bank markets may
change over time, in particular they are increased in case of financial market
tensions which are felt to affect the riskiness of certain asset types. For
instance Citigroup estimated that, due to the tensions in the sub-prime US
markets, haircuts applied to CDOs of ABSs have more than doubled in the
period from January to June 2007. In particular, haircuts on AAA rated
CDOs of ABSs would have increased from 2–4 per cent to 8–10 per cent, on
A rated ones from 8–15 per cent to 30 per cent and on BBB rated ones even
from 10–20 per cent to 50 per cent. (Citigroup Global Markets Ltd., Matt
King, ‘Short back and sides’, July 3, 2007). The same analysis notes that ‘the
level of haircuts varies from broker to broker: too high, and the hedge funds
will take their business elsewhere; too low, and the broker could face a nasty
loss if the fund is wound up’. Changing risks, combined with this competitive pressure, thus lead to changes of haircuts across times; such
changes, however, will be more limited for the standard types of collateral
used in the inter-bank market, in particular for Government bonds. In
contrast, central banks will be careful in raising haircuts in case of financial
tensions, as they should not add to potentially contagious dynamics, possibly leading to financial instability.
7

International Swaps and Derivatives Association. 2007. ‘Guidelines for Collateral practitioners’, Memorandum.

281

Risk management and market impact of credit operations

Table 7.2 Comparison of the key recommendations of ISDA Guideline for Collateral Practitioners with the
Eurosystem collateralization framework
Recommendation according to ISDA
Guidelines for Collateral Practitioners

Eurosystem approach

Importance of netting and cross-product
collateralization for efficiency (pp. 16–9).

Netting is normally not relevant as all exposures
are one-sided. Cross-product pooling is ensured
in a majority of countries (one collateral pool
for all types of Eurosystem credit operations
with one counterparty).

Collateral should preferably be liquid, and risk
control measures should depend on liquidity.
Liquidity can be assumed to depend on the
credit rating, currency, issue size, and pricing
frequency (pp. 19–25).

Eurosystem accepts collateral of different
liquidity, but has defined haircuts which
differentiate between four liquidity categories.

Instruments with low price volatility are
preferred. Higher volatility should be reflected
in higher haircuts and lower concentration
limits (p. 20).

Low volatility is not an eligibility criterion
and also not relevant for any limit. However,
volatilities impact on haircuts.

A minimum credit quality should be stipulated
for bonds, such as measured e.g. by rating
agencies (p. 20).

For securities at least one A - rating by one
recognized rating agency (for credit claims an
equivalent 10 basis point probability of default).

Collateral with longer duration should have higher
haircuts due to higher price volatility (p. 20).

Maturities are mapped into price volatilities and
therefore on haircuts (see above).

Avoid negative correlation of collateral value
with exposure value (in OTC derivatives) (p. 21).

Not relevant (exposure is given by cash leg).

Avoid positive correlation between collateral
value and credit quality of the issuer (p. 21).

Not specifically addressed – with the exception
of the prohibition of close links (of a control
type). Potential weaknesses: large amounts of
unsecured bank bonds submitted (sector
correlation), Pfandbriefe and ABSs originated
by the counterparty itself.

Haircuts should be designed to cover losses
of value due to the worst expected price move
(e.g. at a 99 per cent confidence level) over the
holding period, as well as costs likely to be
incurred in liquidating the assets, such as
commissions and taxes (pp. 21–5).

99 per cent confidence level over holding
period, but nothing for commissions or taxes.

The holding period should span the maximum
time lapse possible between the last valuation and
possibility of a margin call, and actually being able
to liquidate collateral holding in the event of default.
Traditionally, the assumed holding period was one
month, but practice seems to have been moving
to 10 business days (p. 24).

For Government bonds, Eurosystem assumes
one week (five business days) holding period,
for the other three liquidity categories 2, 3 and 4
weeks, respectively.

282

Bindseil, U. and Papadia, F.

Table 7.2 (cont.)
Recommendation according to ISDA
Guidelines for Collateral Practitioners

Eurosystem approach

Low rated debt, such as that rated below
investment grade, might warrant an additional
haircut (p. 23).

Eurosystem does not accept BBB rated (i.e. still
investment grade) collateral, so no need for
additional credit haircut – see also Chapter 8

Concentration of collateral should be avoided;
maximum single issuer concentration limits are
best expressed as a percentage of the market
capitalization of the issuer. There should be
haircut implications if diversification is
compromised (p. 26).

Not applied by Eurosystem.

Collateral and exposures should be
marked-to-market daily (p. 38).

Yes.

2.4 Monitoring the use of the collateral framework and related risk taking
Even if thorough analytical work underlies a given collateral framework, the
actual use of collateral and the resulting concentration of risks cannot be
fully anticipated. This is particularly important because, in practice, an
appropriate point in the flexibility/precision trade-off must be chosen when
building a framework. Indeed, to remain flexible, as well as simple, transparent and efficient, a collateral framework has to accept a certain degree of
approximation. But the degree of approximation which is thought acceptable ex ante may appear excessive in practice, for instance because a specific
collateral type is used in a much higher proportion than anticipated.
The point can be better made with an example: the Eurosystem has defined,
as mentioned above, four liquidity categories and has classified assets in these
categories on the basis of institutional criteria, as shown in Chapter 8.
Obviously liquidity also differs within these categories, as Table 7.3, which
takes bid–ask spreads as an indicator of liquidity, shows.
For instance, while government bonds are normally very liquid, eurodenominated government bonds of e.g. Slovenia and of new EU Member
States are less so – mainly due to their small size. The Eurosystem’s classification of all government bonds under the most liquid category is thus a
simplification. The justification for this simplification is that it does not
imply substantial additional risks: even if there would be larger than
expected use of such bonds, this could not create really large risks, as their

283

Risk management and market impact of credit operations

Table 7.3 Bid–ask spreads as an indicator of liquidity for selected assets (2005 data)

Liquidity
category

Issuers (ratings in parentheses)

Liquidity indicator:
bid–ask spread
(in cent)a

1
1
1
1
1
1
1

Germany, France, the Netherlands and Spain (AAA)
Austria, Finland, Ireland (AAA)
Italy and Belgium (AA)
Portugal (AA/A)
Greece (A)
Slovenia (AA)
Non-euro area new EU Member States
(mostly A rated)

0.5–1
1
0.5–1
1
1
20
15–20

2
2

German La¨nder (AAA-A)
Agencies/supranationals and Jumbo
Pfandbriefe (mostly AAA)

3–5
3–5

3

Non-Jumbo Pfandbriefe (mostly AAA)

3–5

a

Bid–offer spreads observed in normal times on five-year euro-denominated bonds in Trade
Web (when available) in basis points of prices (so-called cents or ticks). Indicative averages
for relatively small tickets (less than EUR 10 million). Bid–offer spreads very much depend
on the size of the issue and how old it is. The difference in bid–offer spreads between the
various issuers tends to increase rapidly with the traded size.

maximum use is still extremely small compared to total outstanding
operations. The table also reports, for information, the ratings of the different bonds, revealing that the effect of ratings on bid–offer spreads is
rather small. This is another justification (see also Chapter 8) for not
introducing credit risk related haircuts into the collateral framework, as the
value added of doing this would be more than compensated by the costs in
terms of added complication. At the end, the approximation deriving from
classifying the assets in four liquidity categories is acceptable provided that:
(i) the average liquidity of each class is correctly estimated; (ii) the heterogeneity within each asset class is not too high; and (iii) the prevailing
heterogeneity does not lead to severe distortions and concentration risk.
In general, the central bank, as any institution offering a collateral
framework, should monitor the actual use of collateral, not only on
aggregate, but also on a counterparty-by-counterparty basis, to determine
whether an adjustment of the framework may be needed. It should also aim
at calculating aggregate risk measures, such as a portfolio VaR figure
reflecting both credit and liquidity risk. A methodology for doing so is

284

Bindseil, U. and Papadia, F.

presented and applied to the Eurosystem in detail in Chapter 10 of this
book. In practice defining an efficient collateral and risk mitigation
framework has to be seen as a continuous interactive process.

3. A cost–benefit analysis of a central bank collateral framework
A central bank should aim at economic efficiency and base its decisions on
a comprehensive cost–benefit analysis. In the case of the Eurosystem, this
principle is enshrined in article 2 of the ESCB/ECB Statute, which states
that ‘the ESCB shall act in accordance with the principle of an open market
economy with free competition, favouring an efficient allocation of
resources’. The cost–benefit analysis should start from the condition,
established in Section 2, that risk mitigation measures make the residual
risk of each collateral type equal and consistent with the risk tolerance of
the central bank. Based on this premise, the basic idea of an economic
cost–benefit analysis is that all collateral types can be ranked in terms of
the cost of their use. This will in turn depend on the five characteristics
listed in Section 2.1. Somewhere on the cost schedule between the least and
the most costly collateral types, the increasing marginal cost of adding one
more collateral type will be equal to its declining marginal value. Of
course, estimating the ‘cost’ and ‘benefit’ curves is challenging, and will
probably rarely be done explicitly in practice. Still, such an approach
establishes a logical framework to examine the eligibility decisions. The
next sub-section provides an example of such a framework in the context
of a simple model.

3.1 A simple model
The following model simplifies drastically in one dimension, namely by
assuming homogeneity of banks, both in terms of needs for central bank
refinancing and in terms of holdings of the different asset types. Even with
this simplification, the estimation of the model appears difficult. Still, it
illustrates certain aspects that might escape attention if eligibility decisions
were not dealt with in a comprehensive model. For instance, if a central
bank underestimated the handling costs of a specific asset type, and thus
overestimated its use by counterparties in central bank operations, then it
may take a socially sub-optimal decision when making it eligible.

285

Risk management and market impact of credit operations

A = {1 . . . n}
E&A
Wj

Vj
D

Kj
kjVj
cjVj

Set of all asset types that may potentially be eligible as
collateral.
Set of eligible assets, as decided by the central bank.
Ineligible assets are (A\E) (i.e. set A excluding set E).
Available amount of asset j in the banking system which can
be potentially used as collateral. This is, where relevant, after
application of the relevant risk mitigation measures needed
to achieve the desired low residual risk; obviously j 2 E.
Amount of collateral j that is actually submitted to the
central bank (again, after haircuts).
Aggregate refinancing needs of banking system vis-a`-vis the
central bank (‘liquidity needs’). Exogenously given in our
model.
Fixed cost for central bank to include asset j for one year in
the list of eligible assets.
Total variable cost for central bank of handling asset j. The
costs include the costs of risk mitigation measures.
Total variable cost for banks of handling asset j. Again, this
includes all handling and assessment costs. If haircuts are
high, obviously costs are increased proportionally. Moreover, this includes opportunity costs: for some collateral,
there may be use in the inter-bank repurchase market, and
the associated value is lost if the collateral is used for central
bank refinancing.

When deciding which collateral to make eligible, the central bank has first to
take note of the banking system’s refinancing needs vis-a`-vis the central
bank (D) and it should in any case ensure that
X
Wj ! D
ð7:1Þ
j2E

Inequality (7.1) is a precondition for a smooth monetary policy implementation. A failure of monetary policy implementation due to collateral
scarcity would generate very high social costs. For the sake of simplicity, we
assume that D is exogenous and fixed; in a more general model, it could be a
stochastic variable and the constraint above would be transformed into a
confidence level constraint. In addition, collateral provides utility as a buffer
against inter-bank intraday and end-of-day liquidity shocks. We assume
that one has to ‘use’ the collateral to protect against liquidity shocks, i.e. one

286

Bindseil, U. and Papadia, F.

has to bear the related fixed and variable costs (one can imagine that
the collateral has to be pre-deposited with the central bank). For the sake of
simplicity, we also assume that, as long as sufficient collateral is available,
liquidity-consuming shocks do not create costs. If however the bank runs
out of collateral, costs arise.
We look at one representative bank, which is taken to represent the entire
P
banking system, thus avoiding aggregation issues. Let r ¼ ÀD þ Vj be
j2E

the collateral reserves of the representative bank to address liquidity shocks.
Let e be the liquidity shock with expected value zero and variance r2 and let
F be a continuous cumulative density function and f be the corresponding
symmetric density function. The costs of a liquidity shortage are p per euro.
Assume that the bank orders collateral according to variable costs in an
optimal way, such that C(r) is the continuous, monotonously increasing
and convex cost function for pre-depositing collateral with the central bank
for liquidity purposes. The risk-neutral representative bank will choose
P
r 2 ½0; Wi Š that minimizes expected costs G of collateral holdings and
i2E

liquidity shocks:

0

1
Z1
EðGðrÞÞ ¼ EðCðrÞ þ pmaxðÀr þ e; 0ÞÞ ¼ @CðrÞ þ p fx ðx À rÞdx A
r

ð7:2Þ
The first-order condition of this problem is (see e.g. Freixas and Rochet
1997, 228)
@C=@r À pFðÀrÞ ¼ 0

ð7:3Þ

The cost function @C/@r increases in steps as r grows, since the collateral is
ordered from the cheapest to the most expensive. The function pF(Àr)
represents the gain from holding collateral, in terms of avoidance of costs
deriving from insufficient liquidity, and is continuously decreasing in r,
starting from p/2. While the first-order condition (7.3) reflects the optimum
from the commercial bank’s point of view, it obviously does not reflect the
optimum from a social point of view, as it does not include the costs borne
by the central bank. If social costs of collateral use are C(r) þ K(r), then the
first-order condition describing the social optimum is simply
@C=@r þ @K =@r À pFðÀrÞ ¼ 0

ð7:4Þ

287

Risk management and market impact of credit operations

Table 7.4 Example of parameters underlying a cost–benefit analysis of collateral eligibility

Category (j)

Available
amounta (W)

Fixed costs
for central
banka (V)

Variable unitary
cost for central
bankb (k)

Variable unitary
cost for
banksb (c)

a (e.g. government securities)
b (e.g. Pfandbriefe)
c
d
e (e.g. bank loans)
f (e.g. commodities)

1,000,000
1,000,000
500,000
500,000
500,000
500,000

0
5
5
5
20
50

0.5
0.5
1
1
1
10

0.5
0.5
1
1
2
5

a
b

in EUR billions.
in basis points per year.

Consider now a simple numerical example (Table 7.4) that illustrates the
decision-making problem of both the commercial and the central bank and
its welfare effects. Note that we assume, in line with actual central bank
practice, that no fees are imposed on the banking system for the posting of
collateral. Obviously, fees, like any price, would play a key role in ensuring
efficiency in the allocation of resources. In the example, we assume that
liquidity shocks are normally distributed and have a standard deviation of
EUR 1,000 billion and that the cost of running out of collateral in case of a
liquidity shock is five basis points in annualized terms. We also assume that
the banking system has either a zero, a EUR 1,500 billion or a EUR 3,000
billion structural refinancing need towards the central bank. The first-order
condition for the representative bank (3) is illustrated in Figure 7.1. The
intersection between the bank’s marginal costs and benefits will determine
the amount of collateral posted, provided the respective collateral type is
eligible.
It can be seen from the chart that if D ¼ 0, 1,500 or 3,000, the bank (the
banking system) will post EUR 1,280, 2,340 and 3,250 billion as collateral,
respectively, moving from less to more costly collateral. In particular, where
D ¼ 3,000, it will use collateral up to type e – provided this collateral and all
the cheaper ones are eligible. How does the social optimality condition on
eligibility (equation (7.4)) compare with that of the commercial bank (7.3)?
First, the central bank should make assets eligible as collateral to respect
constraint (7.1), e.g. when D ¼ 1,500 it needs to make eligible all category
a and b assets. Beyond this, it should decide on eligibility on the basis of
a social cost–benefit analysis. Considering (unlike the commercial bank
that does not internalize the central bank costs) all costs and benefits,

288

Bindseil, U. and Papadia, F.

Table 7.5 Social welfare under different sets of eligible collateral and refinancing needs of the
banking system, excluding costs and benefits of the provision of collateral for refinancing needs
(in EUR billions)
Eligible assets

D¼0

D ¼ 1,500

D ¼ 3,000

A
aþb
aþbþc
aþbþcþd
aþbþcþdþe
aþbþcþdþeþf

30.2
42.8
37.8
32.8
12.8
À37.2

Mon. pol. failure
45.0
15.2
10.2
À9.8
À59.8

Mon. pol. failure
Mon. pol. failure
Mon. pol. failure
0
À40.0
À89.0

5

Marginal costs to banks
Marginal value if D is 0
Marginal value if D is 1500
Marginal value if D is 3000

Marginal value and cost of collateral

4.5
4
3.5
3
2.5
2
1.5
1
0.5

900
108
0
126
0
144
0
162
0
180
0
198
0
216
0
234
0
252
0
270
0
288
0
306
0
324
0
342
0
360
0
378
0
396
0

540
720

360

0
180

0

Collateral posted (billions of EUR)

Figure 7.1.

Marginal costs and benefits for banks of posting collateral with the central bank, assuming structural
refinancing needs of zero, EUR 1,500 billion and EUR 3,000 billion.

Table 7.5 provides, for the three cases, the total costs and benefits for society
of various eligibility decisions.
The highest figure in each column, highlighted in bold, indicates the
socially optimal set of eligible collateral. It is interesting that while in the
first scenario (D ¼ 0) the social optimum allows the representative bank to
post as much collateral as it wishes, taking into account its private benefits
and costs, this is not the case in the second and third scenarios (D ¼ 1,500
and 3,000 respectively). Here, the social optimum corresponds to a smaller