Journal of Financial Market Infrastructures

Risk.net

The recent crises and central counterparty risk practices in the light of procyclicality: empirical evidence

Olga Lewandowska and Florian Glaser

  • This paper focuses on the risk practices of Central Counterparties in the light of their procyclical features.
  • The results reveal only a low average level of conditional correlation between market stress and the total CCP margin requirement, market stress and haircuts respectively.
  • The increases in the CCP haircuts did not make clearing members systematically drop the bonds with increased haircuts from their portfolios.
  • The effectiveness of the regulatory action in the form of macroprudential haircut add-ons is doubtful as systematic overcollateralization of open positions by clearing members, as observed in our data set, may already act as a countercyclical break.

The mandatory central clearing for standardized over-the-counter derivatives, which was introduced by recent financial market reforms, makes central counterparties (CCPs) the most systemically important market participants. However, the theory suggests that, aside from their potential to reduce systemic risk, CCP risk management practices such as margining and collateral haircuts may exacerbate financialcycle fluctuations. Based on almost ten years of empirical data from a leading clearing house, we investigate if and to what extent the procyclical effects suggested by the literature can be statistically confirmed. Our results for the period encompassing the credit crisis and European sovereign debt crisis do not confirm the hypothesis from theoretical research that CCP risk practices are procyclical. Instead, they reveal only a low average level of conditional correlation between market stress and the margin/haircut requirement in the investigated period. Moreover, increases in CCP haircuts do not make clearing members (CMs) systematically drop the bonds with increased haircuts from their portfolios. Our results indicate that the effectiveness of regulatory action in the form of macroprudential haircut add-ons is doubtful, as the systematic overcollateralization of open positions by CMs, as observed in our data set, may already act as a countercyclical break.

The 2007–9 financial crisis exposed weaknesses in the risk management of global financial markets. As a result, policy makers have undertaken a number of initiatives intended to increase the stability of the financial system. One of their main reforms is the obligation to clear all standardized off-exchange traded (over-the-counter, or OTC) derivatives via central counterparties (CCPs). Traditionally, OTC derivatives, like swaps, were negotiated and processed between the counterparties directly involved in a trade. Now, the central clearing service must be applied to OTC derivatives processing. OTC derivatives constitute more than 90% of the global derivatives market. The cleared OTC volumes have been steadily increasing since 2007, so that currently 62% of the volume reported by dealers is cleared centrally (Bank for International Settlements 2016). This clearing obligation is having a major impact on the financial system, as CCPs have become the most systemically important market participants.

On the one hand, CCPs have the potential to decrease systemic risk in OTC markets, when applied comprehensively (Lewandowska 2015). On the other hand, CCPs may themselves concentrate risk or potentially exacerbate the instability of the financial economy in times of downturn.

The aim of this paper is to assess the impact that CCP risk management practices (such as margin requirements and haircuts) had, in the investigated time frame, on procyclicality in the financial markets. In order to achieve this goal, empirical data from one of the leading clearing houses in Europe was analyzed in depth. Broadly speaking, procyclicality is understood as mutually reinforcing interactions between the financial and real sectors of the economy that tend to amplify business-cycle fluctuations and cause or exacerbate financial instability (Committee on the Global Financial System 2010).

A CCP interposes itself between counterparties to a trade, becoming the seller to the buyer and the buyer to the seller. Therefore, a CCP has a neutral position regarding the market risk of the cleared position. However, the CCP becomes exposed to this market risk if any of its members has defaulted. As the CCP guarantees the contract performance to a non-defaulting party, it has to protect itself against any losses from counterparty default. Hence, CCPs collect collateral (margin) from members for their cleared positions. Generally, and regardless of the risk methodology applied, all margining systems contain backward- and forward-looking elements. The initial margin is a forward-looking element, which aims to cover any losses that occurred in the period between member default and liquidation of the respective portfolio. Members have to post the initial margin when the CCP accepts the bilateral trades for clearing. However, the CCP collects variation margin that reflects the daily change in market value of the contracts, ie, the daily gain or loss of a contract due to market movement. CCPs allow members to cover the initial margin requirement via cash or noncash collateral, whereas the variation margin can only be covered using cash. The clearing house applies haircuts on noncash collateral as well as on cash in a foreign currency in order to accommodate for fluctuations in market prices and to protect itself against the envisaged loss in value of the collateral. A haircut is a valuation discount on deposited securities or on cash in a foreign currency. In other words, the deposited collateral is not taken into account at 100% face value.

Haircuts are regularly calibrated to guarantee a high confidence level for extreme market conditions, and they may be an effective microprudential instrument (protecting a CCP and its members from loss). However, in the existing research, a hypothesis has been put forward that margins and haircuts imposed by CCPs may have a procyclical impact (Murphy et al 2014; Committee on the Global Financial System 2010), and may adversely affect the stability of the financial system as a whole. However, there is little or no empirical evidence to back this up.

The procyclicality related to margin requirements and haircuts may be amplified by the reduction in overall margin requirements and haircuts in phases of growth and low-volatility markets as well as the tendency of margin requirements (and haircuts) to rise in periods of market stress, when volatility is also rising. This could impose an additional liquidity pressure on the already stressed markets, leading to defaults of market participants.

The procyclicality resulting from risk management practices can be an issue for both bilateral and cleared markets, as in both cleared and uncleared trading regimes volatility-sensitive risk models are used to calculate margin requirements. As shown in Murphy et al (2014), all commonly used risk models demonstrate some degree of procyclicality. However, as recent financial market reform has made the clearing of OTC markets via CCPs mandatory, the procyclicality of margins imposed by CCPs is of particular concern. One reason for this is that CCP margin calls are unilateral. A CCP makes a call, and all of its members must meet it; otherwise, they are declared to be in default.

In CCP risk management systems, the increasing asset prices and increasing market value of the cleared portfolio typically have no direct influence on the margin requirements posed by the CCP. However, price volatility is a major factor in the margining models. During economic growth, increasing prices are typically associated with lower price volatility. When low volatility is factored into the margining models, CCPs will typically reduce the margin requirement. This leads to the freeing up of resources of clearing members (CMs) for further investments. Under these circumstances, and if market participants’ risk appetites spur further expansion, procyclicality may occur. Two additional effects can be observed in the upcycle. First, a CCP typically reacts to growth phases by reducing haircuts as the volatility of prices of noncash collateral decreases. Second, the increasing asset prices have an impact on the value of noncash collateral posted at CCPs. The value of collateral increases, making it possible for CMs to use the excess collateral for further investment or the collateralization of further transactions. Both effects may incentivize further expansion in growth phases.

In times of high volatility for asset prices, CCPs will typically increase the margin requirement and haircuts. Clearing members have to post additional collateral for their cleared portfolios. In stressed markets, this may lead to liquidity shocks. The resulting liquidity crunch may cause fire sales (ie, banks will try to sell a large amount of financial assets in a short period of time), which will exacerbate the crisis further. Nevertheless, haircuts depend not only on price volatility but also on the credit risk of a security issuer and the liquidity risk of the posted collateral.

The concerns of the regulators with regard to the procyclical features of CCP practices regarding margins and haircuts are reflected in the recent financial market reforms introduced via the European Market Infrastructure Regulation (EMIR) and the Dodd–Frank Act of the US Senate. Apart from the obligation imposed by EMIR to clear all standardized OTC derivatives via a CCP, the regulatory technical standards (RTSs) for CCPs detailed in EMIR, as stated in Articles 28 and 41 (EUR-Lex 2012), require CCPs to select a margin and haircut policy that limits procyclicality. CCPs should avoid any disruptive step changes in margin requirements and, at the same time, establish a transparent and predictable procedure for adjusting margin requirements in response to changing market conditions.

The aim of this paper is to investigate the risk management practice of a leading European CCP in the context of its potential procyclical impact. We do not aim to test the CCP margin model based on its mathematical construction or regarding its potential procyclical features. Instead, our aim is to test the hypothesis of procyclicality based on unique empirical data that reflects both portfolio effects and the changes in the cleared volume. The credit crisis and European sovereign debt crisis are our reference periods.

This paper is structured as follows. Section 2 gives an overview of the related literature. In Section 3, CCP risk management practice is described in detail. Section 4 describes the available data set, research design and results. Section 5 provides the conclusion, research limitations and outlook.

2 Related literature

The existing theoretical studies propose stylized models of how haircuts and initial margin may contribute to a procyclical expansion of leverage and liquidity during the growth phase as well as accelerate de-leveraging and the drying up of liquidity during downturns.

Brunnermeier and Pedersen (2009) show that margins and haircuts have an impact on asset prices via the loss spiral and margin spiral mechanisms. In the loss spiral, the traders’ initial losses lead to funding constraints. As a response, the investors de-leverage their positions by selling assets, which causes a drop in asset prices. As prices move away from fundamentals, margin levels increase, thus worsening the funding conditions for the investors and so on. In the margin spiral, a fall in asset prices induces lenders to increase haircuts and initial margins as a risk management measure. When borrowers face capital constraints and drying liquidity, additional collateral postings may cause fire sales of assets into already falling markets.

What is being recommended to policy makers by the existing theoretical research is, however, unclear, as the explored models are highly simplified and of a stylized nature. While the models focus on haircuts, many other lending terms are also relevant in determining the effective supply of leverage to market participants. For example, Ashcraft et al (2010) show that central banks’ lending facilities mitigate leverage constraints during the crisis. Thus, while, in the models, credit supply invariably responds to adjustments in collateral haircuts, such effects may be less clear in the presence of other credit terms that are simultaneously adjusting to the dynamically changing situation. This caveat is particularly important to keep in mind when evaluating the implications of policies that target the level of haircuts and initial margins.

In one of the few empirical studies on margin setting by CCPs, Abruzzo and Park (2016) analyze the margining method used by the US clearing house of the Chicago Mercantile Exchange (CME). They provide evidence that the margin setting of the CME Group is indeed sensitive to volatility, and that it has a more procyclical impact in times of higher market volatility than in calm periods (as the CME does not immediately decrease margins when volatility drops). This implies that the most disruptive effects of increased margin can be observed when a shock appears after a long period of low volatility markets, as margins are increased from a lower level. Acharya and Viswanathan (2011) also support this thesis.

Murphy et al (2014) propose two measures for assessment of the initial margin models regarding their contribution to procyclicality. The “peak-to-trough measure” is a ratio of the maximum initial margin required for a constant portfolio to the minimum margin required over a fixed observation period across a business cycle. The second metric is defined as the largest increase in margin over an n-day period for a constant portfolio over a fixed observation period. This measure captures the amount of extra margin that market participants would need to fund on a short-term basis. Murphy et al (2014) conclude that all of the most common initial margin models are to some extent procyclical, as the margin requirement increases in times of higher asset price volatility, and reduces when the volatility is decreasing. A Bank for International Settlements report titled “The role of margin requirements and haircuts in procyclicality” identifies collateral haircuts and margining practices in OTC derivatives as one source of procyclicality in the financial system (Committee on the Global Financial System 2010). This report and other papers recently issued by regulators (see European Systemic Risk Board 2015) reflect the consideration that macroprudential haircuts and margin instruments should be introduced by CCPs, like minimum constant through-the-cycle margins and haircuts or countercyclical add-ons. A good overview of the regulatory requirements for clearing houses with regard to limiting procyclicality can be found in Mai (2016).

Brumm et al (2015) argue that the macroprudential tools aimed at reducing procyclicality should be broadly applied across products in order to be effective, as their broad application would stop any leakages to nonregulated products.

However, Goodhart et al (2012) show, based on a simulation approach, that a countercyclical change of macroprudential haircuts results in a minor and ambiguous impact on welfare.

In addition, one should not forget that the purpose of the margin method used as a microprudential tool is to protect a CCP from loss in case of member default. Moreover, any macroprudential margin floors and haircut add-ons that lead to over-margining represent an additional cost for CMs, and may have an adverse impact when considered from a macroeconomic perspective.

This research aims to close the gap in the existing academic literature on CCPs’ margining and haircut practices as well as deliver empirical evidence for a policy recommendation.

3 Risk management by central counterparties

3.1 Margining process

Ordinarily, a CCP calls for margin once a day. The calculation of the required amount of collateral is based on the end-of-day security prices. Apart from this regular margin call, a CCP, if necessary, may call for intraday margin (Wendt 2006). Using real-time prices, a CCP calculates the position’s value and evaluates the intraday risk. In case of a margin shortfall, the CCP issues a margin call against the CM. The CM has an option to enter a risk-reducing trade, deliver security collateral (from, eg, a bond market) or provide additional cash (eventually entering the repurchase agreement (repo) market). When none of the above measures is taken within a predefined time frame, the CM is declared to be in default.

When the haircut applied to the price of a certain security increases dramatically, CMs who have pledged that security as collateral face a decline in the values of their portfolios. Given that the exposure to cleared products remains unchanged, the CMs’ collateral portfolios might become insufficient to cover the margin requirements of the CCP. A CM has several options when it comes to adjusting its collateral portfolio in the case of rising haircuts on some securities (see Figure 1). First, CMs significantly overfund their collateral accounts; they can therefore rely on this buffer to dampen the impact of haircut changes. If a CM is in this comfortable position, it can simply leave the structure of its collateral portfolio unchanged. Second, a CM could choose to substitute the security, subject to an increase in haircut, and either sell it or use it for repo contracts. Hence, if haircuts increase, we should observe a reduction in the quantity (share) of that respective security. Third, the CM could supplement their portfolio by adding more securities to meet the increased margin requirements. Finally, a CM could provide additional cash as collateral.

The relationship between a CCP margin call/haircut increase and a CM’s collateral allocation decisions
Figure 1: The relationship between a CCP margin call/haircut increase and a CM’s collateral allocation decisions.

3.2 Collateral eligibility and calculation of haircuts

The assets accepted as collateral by clearing houses are typically high-quality, fixed-income securities (see European Central Bank (2013) for a comparison of eligible collateral between clearing houses). A few clearing houses also accept equities. The level of the haircut for a given asset is driven mostly by the liquidity of the market for this asset, the volatility of its price (market risk) and the credit risk of a security issuer.

CCPs can use different risk models to calculate haircuts and margining requirements.

4 Research design

4.1 Hypotheses

According to the definition of procyclicality, a link between higher margin requirements, diminishing collateral portfolio via higher haircuts and market stress should be empirically observable. Therefore, we hypothesize the following.

H1:

total margin requirement (TMR) imposed by CCP and market stress are correlated.

H2:

CCP haircuts and market stress are correlated.

The CCP risk practices are said to be procyclical when the conditional correlation is positive and exceeds 50%.

During the recent sovereign debt crisis, the haircuts on peripheral government bonds sharply increased, reducing their liquidity and amplifying the rise in yields of these securities. Higher initial margins diminished the ability of leveraged investors to borrow, and tightened their funding constraints. Consequently, leveraged investors reduced their positions on the bonds with higher haircuts and shifted their portfolios toward securities with lower margins in order to relax their funding constraints (Molteni 2015). Based on this observation, we hypothesize that if haircuts increase, we should observe a reduction in the quantity of that respective security in the collateral portfolio. We develop a test for the resulting scenario of the security dropout event, as stated above.

H3:

securities with increased haircuts are dropped systematically from collateral portfolios.

As high-haircut securities are crowded out from the collateral portfolio, CMs may substitute them with a range of lower-haircut bonds, leading to the diversification of their collateral portfolio. Therefore, we hypothesize the following.

H4:

members diversify their collateral portfolio more in times of market stress, as new securities are added to the collateral pool.

Based on our exploratory data analysis, we hypothesize that CMs systematically overfund their collateral accounts so they have a buffer to rely on, which dampens the impact of haircut changes. The collateral buffer could act as a countercyclical break when CCPs increase margins under stressed market conditions.

H5:

clearing members systematically overcollateralize their positions at a CCP.

4.2 Data description

The available data set, which is unique when compared with the existing research, encompasses the anonymized TMR for 263 CMs of a CCP between November 2005 and February 2015.11The CCP clears both listed and OTC derivatives. The collateral portfolio composition is given for each CM (International Securities Identification Number (ISIN), bond or stock, cash), along with the last available price of collateral (including haircuts). In addition, the margin call amount for each CM is given for each day within the specified time frame. Generally, up to 263 CMs and up to 1 617 000 unique bond and security ISINs (see Figure 2) per day are included in our data set.22The analysis of the membership basis of the CCP shows that the number of new members and leaving CMs is balanced in our data set.

The data for the investigation of any procyclical effects should cover the full credit and business cycle. Our data set fulfils both criteria regarding the length of the analyzed time period (almost ten years) and the interval between two data points (daily data).

Number of unique ISINs (in thousands) in member collateral portfolios and CISS values.
Figure 2: Number of unique ISINs (in thousands) in member collateral portfolios and CISS values.

On a regular basis, the European Central Bank (ECB) publishes an indicator of contemporaneous stress in the financial system: the Composite Indicator of Systemic Stress (CISS). This index, based on euro-area data, specifically captures the effects of flight-to-quality and flight-to-liquidity by incorporating factors from securities markets (such as volatilities, risk spreads and cumulative valuation losses). The CISS involves a large part of the financial system: bank and nonbank financial intermediaries, money markets, equities and bonds markets as well as foreign exchange markets (Hollo et al 2012). We use CISS weekly data for the period 2005–15, downloaded from the ECB’s website (European Central Bank 2016).

We interpolate the weekly CISS values, using the cubic spline interpolation method, in order to account for the daily data from the TMR aggregated through all CMs. Figure 3 shows the daily (interpolated) CISS index values and daily TMR in trillion euros.

During the investigated period (November 2005–February 2015), the CCP applied a uniform margining methodology. The major change to this risk methodology came into force at the end of 2015.

MR in trillion euros and daily CISS
Figure 3: TMR in trillion euros and daily CISS.
Haircut index versus CISS
Figure 4: Haircut index versus CISS.

Figure 4 shows the haircut index and daily CISS. The haircut index encompasses Italian, Spanish, Portuguese, Irish and Greek government bonds, as they were affected the most in the recent European sovereign debt crisis. The index is equally weighted and includes all maturities.

In 2004, the CCP changed the haircut methodology from a static to a dynamic calculation method. Since then, no major change in bond haircut methodology has been applied. Only the minimum haircut levels changed on a regular basis to reflect changes in risk. Figure 5 shows the margin calls aggregated through all CMs per day.

Margin call in billion euros
Figure 5: Margin call in billion euros.

In our analysis, we use daily returns, as opposed to end-of-day values, of TMR and haircuts. We do this for two main reasons. First, just as the returns of an asset are a complete and scale-free summary of the investment opportunity, the returns of the TMR represent the scale-free size of the potential margin call. Second, return series are easier to handle than series of nominal values, because the former have more attractive statistical properties.

Let Pt denote the value of the TMRs or haircut index at the end of trading day t. The log return is defined as rt=ln(Pt/Pt-1). The analysis of the plots of the daily log returns, squared returns and absolute value of returns for, respectively, daily TMR and CISS, weekly TMR and CISS, and the haircut index and CISS delivers evidence of volatility clustering (the plots are available on request). The volatility clustering characteristic for financial data refers to the observation that “large changes tend to be followed by large changes, of either sign, and small changes tend to be followed by small changes” (Mandelbrot 1963). In other words, volatility depends more on recent-past values than distant-past values.

Table 1 provides some standard summary statistics of the returns, along with the Jarque–Bera (JB) test for normality (Bera and Jarque 1980).33Under the null hypothesis, both the skewness and the excess kurtosis are zero. Any deviation from this increases the JB statistic. The distribution of daily TMR and haircut index returns is clearly nonnormal, exhibiting negative skewness, pronounced excess kurtosis and high JB values. Part of this nonnormality is caused by the strong leptokurtosis during the subprime crisis and the European sovereign crisis. Weekly TMR and CISS returns have a distribution that is much closer to normal than daily returns.

Table 1: Summary statistics for daily and weekly TMR and CISS data. [Sample period is July 28, 2005–February 23, 2015 (giving 2451 daily observations) and August 5, 2005–February 20, 2015 for weekly data (giving 484 weekly observations). SD denotes standard deviation.]
                 
(a) Daily returns
Time                
series Mean Median Min Max SD Skew Kurt JB
TMR 0.0005 0.0005 -0.1699 0.1059 0.0174 -0.1810 07.1754 05 300
CISS -0.0001 -0.0003 -0.4004 0.6769 0.0705 1.0745 15.1693 24 000
                 
(b) Weekly returns
Time                
series Mean Median Min Max SD Skew Kurt JB
TMR 0.0024 0.0032 -0.1484 0.1426 0.0329 0.0148 2.4866 120
CISS -0.0009 -0.0062 -0.9251 1.2211 0.2638 0.4525 2.3381 130
Table 2: Summary statistics for daily haircut index and CISS. [Sample period is November 8, 2005–February 23, 2015, giving 2336 daily observations.]
Time                
series Mean Median Min Max SD Skew Kurt JB
Haircut 0.0019 0.0000 -0.5219 0.5840 0.0777 -0.1768 14.2011 20 000
index                
CISS -0.0003 -0.0005 -0.4004 0.6769 0.0711 1.1144 15.2993 23 000

Weekly TMR returns are calculated as the sum of daily returns from a given week. They display less volatility clustering than the daily returns. Further tests confirm this observation. Table 3 shows values of multivariate Ljung–Box Q-statistics (MQ), computed from daily and weekly squared TMR returns as well as squared daily haircut index returns (Tsay 2014). Also, the results of the Lagrange multiplier (LM) test for autoregressive conditional heteroscedasticity (ARCH) are presented in Table 3, for various values of p, ie, lags (Engle 1982), calculated as in Tsay (2005, pp. 101–102). There is clear evidence of volatility clustering in the daily and weekly returns. The LM test shows strong ARCH effects (especially for daily returns). Moreover, based on the results of the Ljung–Box statistics, the null hypothesis that the autocorrelations for lags 1 through p are all jointly zero can be rejected.

The augmented Dickey–Fuller test returns large negative numbers, so the hypothesis that there is a unit root can be rejected (Banerjee et al 1993). Further, the results of the Kwiatkowski–Phillips–Schmidt–Shin (KPSS) test let us assume that the investigated time series are trend stationary (Kwiatkowski et al 1992).

Table 3: Results of MQ and LM test for ARCH. [*CISS daily is shown twice, based on different sample sizes.]
             
(a) Daily returns
  Ljung–Box LM test
  statistics for ARCH
     
? 1 5 10 1 5 10
TMR 46.7 098.8 143.7 47 074 093
  (0) (0) (0) (0) (0) (0)
CISS 59.4 222.4 298.3 59 170 210
  (0) (0) (0) (0) (0) (0)
Haircut 51.4 069 098.5 51 071 094
index (0) (0) (0) (0) (0) (0)
CISS* 56.2 210.8 283.5 56 160 200
  (0) (0) (0) (0) (0) (0)
             
(b) Weekly returns
  Ljung–Box LM test
  statistics for ARCH
     
? 1 5 10 1 5 10
TMR 08.8 24.53 039.86 08.8 18 35
  (0.003) (0) (0) (0.003) (0.003) (0)
CISS 18.3 65.2 109.3 18.0 42 55
  (0) (0) (0) (0) (0) (0)

4.3 Methodology

Financial time series that exhibit time-varying volatility clustering are often evaluated with generalized autoregressive conditional heteroscedasticity (GARCH) models. GARCH models allow us to estimate current and future levels of volatilities based on historical data. These models capture the nonlinear dynamics in volatility and recognize that volatilities are not constant in time; for instance, a particular volatility may be high or low, depending on the time period in question.

Engle (2002) proposed a dynamic conditional correlation (DCC GARCH) model, which can be estimated using a two-step method based on the likelihood function. This model allows the simultaneous modeling of variances and conditional correlations of several time series. The estimation consists of two steps. First, the conditional variance of each univariate time series is estimated. Second, the standardized regression residuals obtained in the first step are used to model those conditional correlations that vary through time.

In the first step of the analysis, we apply a DCC GARCH model to the interpolated CISS values and aggregated margin requirements of all CMs. We do this in order to test our hypothesis of correlation between the TMR and systemic stress. The margin requirement is aggregated per day among all members. Ten members were removed from the sample, as they are state institutions exempted from posting margins. The weekly CISS values were interpolated (using the cubic spline interpolation method) to obtain the daily values of the market stress index. For comparison, we estimate DCC GARCH parameters also based on the weekly CISS and weekly sums of TMR. The parameters of the applied DCC model are specified in Table 4.

Table 4: Parameters of the applied DCC(1,1) models. [“MVT” denotes Student t distribution.]
  Daily TMR–CISS Weekly TMR–CISS Haircut–CISS
Estimation Two step Two step Two step
Distribution MVT Norm MVT
Number of parameters 16 16 16
Number of series 02 02 02

Analogously to an analysis of the correlation between market stress and the TMR, we perform an analysis of collateral haircuts and the CISS index. Here, we also use the interpolated CISS data.

It is a well-known fact that returns from financial market variables such as exchange rates or asset prices, measured over short time intervals, are much better fitted by non-Gaussian probability distributions. The same applies to margin requirements and haircuts that refer to the risks of financial assets. The empirical distribution of such returns is more peaked and has heavier tails than the normal distribution, which implies that significant changes in returns occur with a higher frequency than under normality. Therefore, it may be more appropriate to use a distribution that has fatter tails than the normal distribution. One of the most common fat-tailed error distributions for fitting GARCH models is the Student t distribution. Based on the results presented in Tables 1 and 2 (showing negative skewness, excess kurtosis and large JB values), we adopt the Student distribution for fitting the model of daily TMR–CISS and haircuts–CISS.

Different numbers of lags can be used as parameters of the GARCH models; however, Hansen and Lunde (2004) provide compelling evidence that it is difficult to find a volatility model that outperforms the simple GARCH(1,1). Therefore, we apply the first lags in the ARCH and GARCH processes.

Another stylized fact of financial volatility is that bad news tends to have a larger impact on volatility than positive news. That is, volatility tends to be higher in a falling market than in a rising market (see Black (1976) for an explanation of this “leverage effect”). Nelson (1991) proposed the exponential GARCH (EGARCH) model to allow for leverage effects.

Next, we take a closer look at the different effects of the collateral portfolio composition. We treat a dropout event as a trigger of a forthcoming fire sale. A dropout is the reduction of a given bond share in the collateral portfolio to zero, ie, the bond is dropped completely from the collateral portfolio. In the first step of our analysis, we identify all dropout events for each ISIN and member. In our estimation window of thirty days preceding the dropout, we calculate the average haircut value for the given bond. We then test if the difference between the level of haircut at the date of a dropout event and the average from the preceding thirty days is larger than zero.

To test H4 regarding the diversification of collateral in stressed market conditions, we first calculate a number of distinct ISINs per day (both for bonds and equities). In a second step, we calculate the correlation between the number of ISINs and CISS. As we expect the correlation to be equally strong in both calm and stressed markets, we calculate a Pearson’s product–moment correlation coefficient.

For the analysis of potential overcollateralization, we compare the daily TMR for each member with the collateral portfolio value for that day (for the first time in the literature, including cash posted as collateral and taking collateral haircuts and foreign exchange haircuts on foreign currencies into account). In order to assess the extent of the collateralization, we calculate the daily ratios R of over/undercollateralization for each member:

  RCMi=j=1l((CVi,j-TMRi,j)/TMRi,j)N,  

where CMi is CM i; j represents day; CVij is collateral value including haircuts of member i on day j; TMRij is the total margin requirement of member i on day j; N is the number of days in the data set; and N=1,,l.

4.4 Results

Dynamic conditional correlation between CISS and TMR (daily data)
Figure 6: Dynamic conditional correlation between CISS and TMR (daily data).

Figures 6 and 7 show the conditional correlation of CISS and aggregated TMR (based on daily and, subsequently, weekly data). The obtained results reveal only a low average correlation level, which leads to the rejection of H1 based on the definition of procyclicality in this research. In the investigated period, the procyclical impact of the TMR imposed on CMs by the CCP could not be confirmed.

Interestingly, the obtained results seem to point to an amplification of dynamic conditional correlations during the period of crisis, which stretched from August 7, 2007 to November 18, 2012 (with short recovery periods in the first half of 2010 and 2011). The phases of market stress are identified by above-average values of the daily interpolated CISS index (ie, above 0.262).

The second interesting result is that the correlation between TMR and CISS becomes negative in 2013. This can be explained by the fact that the CCP incorporates historical values of high volatility in the margin model. The resulting margin requirement stays at a high level, whereas the current market volatility is actually falling.

Dynamic conditional correlation between CISS and TMR (weekly data)
Figure 7: Dynamic conditional correlation between CISS and TMR (weekly data).

Figure 8 shows the conditional correlation between CISS and the haircut index. The overall low level of dynamic conditional correlation leads to the rejection of H2, according to the definition applied in this research. The haircut policy of the investigated CCP cannot be confirmed to cause a procyclical effect in the investigated time period. However, the obtained results point to an amplification of the correlation during the periods of crisis. They suggest the correlations between haircuts and CISS have risen (by a different magnitude) in times of market stress, with characteristic correlation spikes. The sudden haircut increases may have a severe impact on CMs in stressed markets and spur a liquidity crunch when the external financing conditions deteriorate.

Conditional correlation between bond haircut index and CISS
Figure 8: Conditional correlation between bond haircut index and CISS.

Table 5 shows the estimated dynamic correlation coefficients for TMR and CISS as well as for CISS and the haircut index. The results indicate that the correlations are significant at the 10% level (for haircuts–CISS) or higher, with dynamic conditional correlation parameters noted by dcca1 and dccb1. Notably, the sum of those coefficients is, in all cases, very close to one, which indicates a persistence of volatility in time.

Table 5: Results of the DCC GARCH(1,1) models for the dynamic conditional correlation between TMR and CISS, and between CISS and the haircut index. [*, ** and *** denote statistical significance at the 1%, 5% and 10% levels. SE denotes standard error.]
Coefficient TMR–CISS (daily) TMR–CISS (weekly) Haircuts–CISS
dcca1 0.004211** 0.017323 0.008208***
  SE: 0.001847 SE: 0.005342 SE: 0.005009
  t-value: 2.27969 t-value: 3.24248 t-value: 1.6385e+00
dccb1 0.994967 0.966710 0.989131
  SE: 0.001637 SE: 0.065318 SE: 0.004737
  t-value: 607.96260 t-value:14.80009 t-value: 2.0883e+02

Figure 9 shows dropout events per date. An extremely high number of these events (in which members completely removed a given bond from their collateral portfolio) were observed starting in 2012, which can be associated with the European sovereign debt crisis.

Number of bond dropout events per day
Figure 9: Number of bond dropout events per day.

The haircut policy of the CCP was proven to be prone (only to a limited extent) to fire sales. Only 21% of the dropout events were driven by an increase of collateral haircuts during the thirty days preceding the event. H3 has been rejected.

However, the collateral portfolios become more diversified under the stressed market conditions and less diversified in calm market times. The correlation between the CISS values and the number of unique ISINs (0.7682) was significant at the 95% confidence interval. The empirical data supports H4.

Over/undercollateralization of member accounts
Figure 10: Over/undercollateralization of member accounts.

H5, regarding systemic overcollateralization, was confirmed based on empirical data. As shown in Figure 10, only five members systematically undercollateralized their collateral portfolios (by a few percent of the TMR).44Undercollateralization is only short term in nature. When a member does not have sufficient collateral in their account to cover the margin requirement, the CCP submits a margin call. The member then has a few hours to provide the required collateral, without being declared in default. The remaining members maintained more collateral than required to cover the current TMR. In some cases, the average overcollateralization was very high, exceeding a thousand times the current margin requirement. Figure 11 shows the median of the under/overcollateralization ratios per day. The results may be explained by the fact that some members do not actively manage their collateral accounts at the CCP to keep the operational effort low, or they keep more collateral than required by the CCP to reflect their internal assessment of the risk exposure. Figure 5 showing margin call makes it clear that a margin call may exceed the previous day TMR by several times.

Median of the under/overcollateralization ratios per day
Figure 11: Median of the under/overcollateralization ratios per day.

5 Conclusion, research limitations and outlook

5.1 Conclusion

A major policy concern in the centrally cleared derivatives market is that margin requirements and collateral haircuts, imposed by CCPs, can sharply rise in times of stress, inhibiting trading and causing liquidity problems that exacerbate the crisis. This opinion is widely shared both in theoretical research and in the industry.

This research gives a first insight into the impact of CCPs’ risk management practices, seen from a systemic risk perspective. The comprehensive data set we use spans almost ten years. Based on this data set, it cannot be statistically confirmed that, in the investigated period, CCP risk practices caused a significant procyclical effect.

The historical data shows only a low average level of correlation between price volatility and the level of margins or haircuts in the investigated time frame. One possible explanation for the average low correlation levels is that the CCP already applies margin floors. However, the floors were imbedded in the applied CCP risk methodology, and we do not possess the data required to extract them. Therefore, due to a lack of possibility for further testing, we cannot definitely confirm the margin floors as a valid explanation.

The correlation between market stress and the level of margins or haircuts increases suddenly during periods of market stress. Nevertheless, whether the observed spikes of correlation could really spur a liquidity crunch for CMs depends on the external financing conditions.

The dynamic conditional correlation between the TMR and the stress index of the eurozone, published by the ECB, proved to be a good metric of procyclicality, but it requires further research in order to obtain a better understanding of how CCP risk management practices could amplify business-cycle fluctuations. Addressing procyclicality in the financial system is an essential component of strengthening the regulatory framework. Therefore, regulators are considering the introduction of macroprudential tools, such as higher minimum haircuts and margins, or countercyclical add-ons. Having said that, we found that the systematic overcollateralization of open positions by CMs (as observed in our data set) may already act as a countercyclical break. This result indicates that the burden of a reasonably low, nonvoluntary collateral floor on the liquidity of CMs would probably not be high. However, there might be some challenges in the implementation of countercyclical breaks, eg, the timing of the introduction of such a regulatory tool, the determination of the appropriate level and, finally, the question of whether different levels are feasible for different financial instruments. Another interesting result of this research is that the negative correlation between the margin requirement and market stress appears after the period of crisis. The negative correlation can be explained by the fact that the CISS index was sinking after the crisis period, when markets recovered, but the CCP margin was not decreasing, as the previous events (shocks from the crisis) were incorporated via stress scenarios into the margin model. Abruzzo and Park (2016) come to a similar conclusion. Finally, the haircut policy of the CCP was not proven to be prone to fire sales, as suggested by the literature. Only some of the dropout events were motivated by an increase in haircuts. On the other hand, the collateral portfolios of the CMs became more diversified in times of market stress than in the calm market. This finding suggests that new types of securities were added to those already present in the collateral portfolio when the CMs needed to cover the increase in margin requirement under the stressed market conditions.

5.2 Research limitations and outlook

As different CCPs use proprietary risk margining models, the transferability of our results on other CCPs is limited.

In the presented research, we abstract from the assessment how quickly and how strongly the CCP should react to a significant change in risk regime, and what is the acceptable level of margin/haircut increase in a given time frame. The existing measures of procyclicality would allow us to compare risk management models between different clearing houses when the required data is available.

It was also beyond the scope of this paper to analyze the direct impact of the minimum haircuts applied by the CCP. Such research would require a simulation of the level of haircut, both with and without the minimum haircut restriction, as well as a simulation of the corresponding change in the collateral practices of the CMs.

When formulating our hypotheses, we focused mostly on bond haircuts, leaving foreign exchange and equity haircuts out of our study. The limited number of events in our data set prevented us from a deeper analysis.

Moreover, for dropout events, we have not taken the potential time-lagged manner of haircut settings and collateral dropouts into account. It may be the case that the members who possess private information react more quickly (ie, drop the bond from their portfolio faster) than the CCP increases the haircut. The dropout events are, therefore, only to a small extent explained by an increase in the CCP haircut. Further analysis of haircut setting by CCPs and member collateral practices could shed some light on this aspect.

Declaration of interest

The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.

References

Abruzzo, N., and Park, Y. (2016). An empirical analysis of futures margin changes: determinants and policy implications. Journal of Financial Services Research 49(1), 65–100 (http://doi.org/b2cw).

Acharya, V. V., and Viswanathan, S. (2011). Leverage, moral hazard, and liquidity. Journal of Finance 66(1), 99–138 (http://doi.org/b2cv).

Ashcraft, A., Gârleanu, N., and Pedersen, L. H. (2010). Two monetary tools: interest rates and haircuts. NBER Macroeconomic Annual 25, 143–180 (http://doi.org/b6qv5c).

Banerjee, A. J., Dolado, J., Galbraith, J. W., and Hendry, D. F. (1993). Cointegration, Error Correction, and the Econometric Analysis of Non-Stationary Data. Oxford University Press (http://doi.org/cvfpjj).

Bank for International Settlements (2016). OTC derivatives statistics at end-June 2016. Derivatives Publication, BIS. URL: http://www.bis.org/publ/otc_hy1611.htm.

Bera, A. K., and Jarque, C. M. (1980). Efficient tests for normality, homoscedasticity and serial independence of regression residuals. Economics Letters 6(3), 255–259 (http://doi.org/dm8gjx).

Black, F. (1976). Studies in stock price volatility changes. In Proceedings of the 1976 Business Meeting of the Business and Economics Statistics Section, pp. 177–181. American Statistical Association.

Brumm, J., Grill, M., Kubler, F., and Schmedders, K. (2015). Margin regulation and volatility. Journal of Monetary Economics 75(C), 54–68.

Brunnermeier, M. K., and Pedersen, L. H. (2009). Market liquidity and funding liquidity. Review of Financial Studies 22(6), 2201–2238 (http://doi.org/drd3sm).

Committee on the Global Financial System (2010). The role of margin requirements and haircuts in procyclicality. CGFS Paper 36, Bank for International Settlements. URL: http://www.bis.org/publ/cgfs36.pdf.

Engle, R. (1982). Autoregressive conditional heteroscedasticity with estimates of the variance of United Kingdom inflation. Econometrica 50, 987–1007.

Engle, R. (2002). Dynamic conditional correlation: a simple class of multivariate generalized autoregressive conditional heteroskedasticity models. Journal of Business and Economic Statistics 20(3), 339–350 (http://doi.org/fb46f2).

EUR-Lex (2012). Commission Delegated Regulation (EU) No 153/2013 of 19 December 2012 supplementing Regulation (EU) No 648/2012 of the European Parliament and of the Council with regard to regulatory technical standards on requirements for central counterparties. Official Journal of the European Union L 52/41. URL: http://bit.ly/2lMCNmV.

European Central Bank (2013). Collateral eligibility requirements. a comparative study across specific frameworks. Report, ECB. URL: http://bit.ly/1wK49Ih.

European Central Bank (2016). Statistical data warehouse. Website. URL: http://sdw.ecb.europa.eu.

European Systemic Risk Board (2015). The efficiency of margining requirements to limit pro-cyclicality and the need to define additional intervention capacity in this area. Report, European System of Financial Supervision. URL: http://bit.ly/2lMMMc3.

Goodhart, C., Kashyap, A., Tsomocos, D., and Vardoulakis, A. (2012). Financial regulation in general equilibrium. Chicago Booth Research Paper 12-11, Fama–Miller Working Paper, University of Chicago (http://doi.org/b2cx).

Hansen, P., and Lunde, A. (2004). A forecast comparison of volatility models: does anything beat a GARCH(1,1) model? Journal of Applied Econometrics 20, 873–889 (http://doi.org/cqjnjj).

Hollo, D., Kremer, M., and Lo Duca, M. (2012). CISS: a composite indicator of systemic stress in the financial system. Working Paper 1426, European Central Bank.

Kwiatkowski, D., Phillips, P. C. B., Schmidt, P., and Shin, Y. (1992). Testing the null hypothesis of stationarity against the alternative of a unit root. Journal of Econometrics 54(1–3), 159–178 (http://doi.org/ddvc8r).

Lewandowska, O. (2015). OTC clearing arrangements for bank systemic risk regulation: a simulation approach. Journal of Money, Credit and Banking 47, 1177–1203 (http://doi.org/b2cz).

Mai, E. (2016). Regulatory requirements for limiting the procyclicality of the CCPs. Presentation, Mainstay. URL: http://bit.ly/2mwwsKJ.

Mandelbrot, B. B. (1963). The variation of certain speculative prices. Journal of Business 36, 392–417.

Molteni, F. (2015). Liquidity, government bonds and sovereign debt crises. Working Paper 2015-32, CEPII.

Murphy, D., Vasios, M., and Vaus, N. (2014). An investigation into the pro-cyclicality of risk-based initial margin models. Financial Stability Paper 29, Bank of England. URL: http://bit.ly/2lh541L.

Nelson, D. B. (1991). Conditional heteroskedasticity in asset returns: a new approach. Econometrica 59, 347–370 (http://doi.org/bz6t2m).

Tsay, R. S. (2005). Analysis of Financial Time Series, 2nd edn. Wiley (http://doi.org/cnr7z2).

Tsay, R. S. (2014). Multivariate Time Series Analysis with R and Financial Applications. Wiley.

Wendt, F. (2006). Intraday margining of central counterparties: EU practice and a theoretical evaluation of benefits and costs. Working Paper, De Nederlandsche Bank. URL: http://bit.ly/2lYIEpW.

Only users who have a paid subscription or are part of a corporate subscription are able to print or copy content.

To access these options, along with all other subscription benefits, please contact [email protected] or view our subscription options here: http://subscriptions.risk.net/subscribe

You are currently unable to copy this content. Please contact [email protected] to find out more.

You need to sign in to use this feature. If you don’t have a Risk.net account, please register for a trial.

Sign in
You are currently on corporate access.

To use this feature you will need an individual account. If you have one already please sign in.

Sign in.

Alternatively you can request an individual account here: