Journal of Financial Market Infrastructures

A balanced approach to central counterparty margining

Sunil Cutinho, Suzanne Sprague and Matt Waldis

  • This paper presents the potential benefits and risks of margin setting frameworks which incorporate expert judgment as a complement to model driven inputs relative to frameworks that rely primarily on automated models. We argue for a balance between these two approaches.  
  • Many global CCPs operate in multiple jurisdictions and, as a result, must adapt to multiple regulatory frameworks. It is important that CCPs be able to tailor their margin models to the unique risks associated with the products they clear.
  • While the margin model framework is important, it is equally, if not more, important to examine margin collection practices and their impact on risk.

This paper is meant to serve as a comparison of the approaches and margin models employed by central clearing counterparties (CCPs). We look at two different approaches: fully automated margin models and margin models that incorporate expert judgment. The latter type of model utilizes value-at-risk (VaR) methodologies, and the outputs of these VaR methodologies are subject to manual review before the final parameters are determined. The former type of model does not involve manual review or any alteration of the parameters. While these approaches may differ greatly, there are components of each that are critical to prudent margin setting. This paper concludes that a combination of data-driven models and applied expert judgment most appropriately captures the risks in centrally cleared markets, emphasizing the importance of expert judgment in margin setting. It is important to outline the goals of initial margin as a risk management tool before selecting any approach to margining. The foremost goal of initial margin is to cover potential future exposures, whether on a product- or portfolio-level basis. This is typically demonstrated through the coverage of losses achieving, at a minimum, the common 99% standard set out in the Principles for Financial Market Infrastructures (PFMIs) by the Committee on Payments and Market Infrastructures (CPMI) and International Organization of Securities Commissions (IOSCO) in April 2012. In addition to achieving coverage, the objective of a margining framework is predictability and stability.

1 The current state of margin and rising trends

Central clearing counterparties (CCPs) typically use two primary tools to manage risk on a day-to-day basis: initial margin and mark-to-market. Initial margin represents good-faith deposits that are posted when any position is taken on. It is designed to cover potential future exposures that may arise from a given portfolio. Fully automated margin models result in initial margin changes on a daily basis, while margin models that incorporate expert judgment are less immediately reactive to daily changes in volatility. Mark-to-market aims to capture current unrealized exposures by collecting an amount equal to the change in value of a portfolio. While margin is not the only risk management tool available to CCPs, it is a significant one, and historically the primary mitigant to manage participant defaults. This makes the margin frameworks used to determine margin levels a critical component of prudent risk management.

Following the 2008 financial crisis, global regulators embraced central clearing as one of the primary risk mitigation tools to combat future crises based on the fundamental risk management standards employed by CCPs, along with daily mark-to-market and initial margin. This prompted policies that strongly encouraged or required market participants in multiple jurisdictions to centrally clear certain derivatives exposures. As a result, banks and regulators increased their focus on margin methodologies; they were most familiar with automated algorithms, given these algorithms’ utilization by banks to quantify the level of risk taken on by the firm. Regardless of the performance of such models, they were perceived to provide transparency due to greater reliance on data as a measure of objectivity. However, the criteria for judging the effectiveness of a margin model is not primarily rooted in the ability to replicate the results of the model, but rather in the ex post performance of the model in capturing the potential future exposures. This ex post performance can be evaluated objectively using daily backtesting.

Many global CCPs operate in multiple jurisdictions and, as a result, must adapt to multiple regulatory frameworks. It is important that CCPs are able to tailor their margin models to the unique risks associated with the products they clear. In addition, prescriptive regulation on margin-setting frameworks often results in unintended negative consequences in terms of model risk. This cultivates model monoculture, which, in turn, increases the likelihood of systemic risk management failures when unanticipated market events occur. Therefore, regulators should avoid a bias toward prescribing specific margin-setting frameworks and focus more on outcomes-based tests, such as the adequacy of portfolio level margin coverage. As such, regulatory authorities should work toward implementing a principles-based regime for evaluating the adequacy of margin frameworks, while using the outcomes of the frameworks as their primary criteria.

Historically, margins for exchange-traded derivatives have been determined using an approach that combines model outputs and expert judgment. This approach provides greater flexibility to address changing market conditions when setting margin levels. However, the current trend in margining has shifted to a greater reliance on automated model-driven margin setting. While there are significant benefits to an automated model-based framework addressing the increasing complexity of certain cleared derivatives, expert judgment has proven invaluable in a number of systemic market events; this has been observed through the use of margin buffers applied ahead of events anticipated to cause extreme volatility, such as the US election. With a significant portion of products shifting to central clearing, the importance of prudent margin-setting practices is now more important than ever. The question we must ask is this: should margin be completely driven by automated models, or should expert judgment play a role (and, if so, how big a role)?

Although it is important to note that, despite the differences between margin-setting frameworks that rely on expert judgment to complement model-driven outputs and those that are fully automated, all margin frameworks rely on human judgment to some degree; even autonomous margin model frameworks require expert judgment in their development as well as in their ongoing maintenance. In reality, the input of expert judgment exists on a spectrum, with some margin-setting frameworks requiring more and others less.

Given CCPs operate in multiple jurisdictions, both regulators and CCPs must take care to strike a balance between the benefits of employing a common margin framework and the risks of model monoculture. Excessive prescription could force CCPs to use the same parameters and inputs in their margining methodologies, leading to identical risk management standards across the industry (despite a diverse set of products and participants) and, thus, risks. This approach may be successful in capturing historically observed volatility, but it increases the likelihood of a systemic risk management failure after an unanticipated market event.

While the margin model framework is important, it is equally, if not more, important to examine margin collection practices and their impact on risk. CCPs in several jurisdictions – the United States, Hong Kong, Singapore and Brazil – have designed their margin methodologies around the gross customer margin collection requirements common under CCP regulations. Gross margining requires that the minimum margin required by the CCP be calculated at the client level and the summed total be passed on to the CCP, prohibiting any netting between client accounts. A net margining regime, however, allows for the positions of all clients at a clearing member to be margined together in one account, resulting in significant netting and margin offsets between client accounts. Gross margin requirements can result in almost two times more margin being required by the CCP. Gross margining results in a significantly larger amount of resources being available to the CCP in order to manage a clearing member default, and it is paramount in ensuring the portability of customer accounts in the event of a clearing member default. Portability means that, in the event of default by a clearing member, the clearing member’s customer positions can be transferred to surviving clearing members, rather than liquidated. In a net margin regime, there is not enough margin to port clients to multiple clearing members, since the positions were netted together. A review of the October 1987 failure of the Hong Kong Guarantee Fund Corporation noted that: “The [clearinghouse] in particular can be better – possibly much better – protected against price swings. Under a gross system it can withstand a higher price shift for the same level of default” (Securities Review Committee 1988).

This paper presents the potential benefits and risks of margin-setting frameworks that incorporate expert judgment as a complement to model-driven inputs, relative to frameworks that rely primarily on automated models. We argue for a balance between these two approaches. On the one hand, model-driven margining frameworks can be used to effectively address large numbers of highly correlated products. On the other hand, expert judgment complements model-driven analysis in cases where products are more standardized but are subject to risk factors that are more challenging to model, or where products have little to no historical data.

2 The benefits of fully automated margin models

As stated previously, fully automated model-based margin-setting frameworks typically rely more on extensive historical data to drive the scenarios that determine margin levels and less on the frequent calibration of parameters by risk managers contributing expert judgment. Often, these frameworks have fewer parameters that are configurable but largely run autonomously day-to-day, resulting in daily margin changes. These margin-setting frameworks can offer great efficiency in margining, including the ability to margin a portfolio with a large number of risk factors in an automated fashion as well as to automatically offer a high degree of portfolio correlation benefits. Developing an automated margin model is a complex undertaking; however, as previously mentioned, once the model is developed, tested and implemented, it can run largely autonomously. As a result, large numbers of portfolios can be run on an ongoing basis, with little manual input being required. This may enable increased granularity in margining as well as increased frequency of adjustment in response to changing market conditions, with ample upfront design and coding. In a similar vein, automated model-based margin-setting frameworks can be structured to automatically offer portfolio correlation benefits, while being agile enough to adjust these benefits based on changes in portfolio makeup and market correlations, assuming good input data.

3 The risks of relying on fully automated margin models

The relatively small amount of ongoing expert judgment in automated model-based margin-setting frameworks may lead to significant challenges when accounting for factors such as seasonality, portfolio concentration, external market dynamics, liquidity and/or event risk. This is because such risks are not readily available through historical data and are hard to model due to their complexity and/or idiosyncrasy. These risk factors, however, can be accounted for and calibrated by risk managers using real-time information, and drawing from extensive market experience that is not bound by a finite data set or algorithm. Thus, the risk management of such factors is typically benefited by a larger degree of expert judgment overlaid on model-driven algorithmic outputs. For instance, although certainly not impossible, it is quite difficult to reliably and accurately model product seasonality and account for changing weather patterns and their impact. Given the unique and complex nature of seasonality factors, designing algorithms that can precisely model the timing and degree of related volatility can be extremely challenging. As such, a larger degree of expert human judgment, typically in the form of additional input parameters, mitigates the shortfalls of an automated model in accounting for such complexities.

In addition, under automated model-driven margin-setting regimes it may be difficult to proactively address stressed market conditions. Known binary events such as political events (eg, the US presidential election or the UK referendum), where certain outcomes can lead to generally unpredictable market volatility, can be especially challenging to models using only a data-driven algorithm. Observations of volatility leading up to and following such events have indicated that, even where risks can be modeled, the results may still lead to adverse or imprudent margin levels. For example, realized volatility in the British pound foreign exchange (FX) futures contract actually decreased heading into the UK referendum, which likely would have resulted in decreasing margins heading into the referendum date for most, if not all, automated margin models. Automated margin models lacking forward-looking inputs and utilizing very little expert judgment input will not proactively raise margins ahead of these types of events, potentially introducing risks into the markets when the unexpected volatility is realized. Unexpected market volatility events, such as a currency de-pegging or a government-driven currency manipulation, may also present challenges. The circumstances and effects of political event risk are unique, leading nearly all margin-setting frameworks to be ill prepared for the associated volatility. However, automated margin models versus margin models that incorporate expert judgment react to such events differently.

When analyzed ex post, automated model-driven margining approaches tend to have larger increases in margin, as models may overreact to shifts in volatility following a period of unexpected volatility. Conversely, during prolonged periods of low volatility, automated margin models may follow volatility below the appropriate margin levels. This possibility often necessitates a margin floor, which CCPs should be cognizant of when developing their margin frameworks. When there is an acute uptick in volatility, the model will overcompensate with significant, and potentially adverse, increases to margins. As previously mentioned, models provide quantitative objective feedback, lacking the sometimes necessary fundamental factors employed by those with risk management expertise. In David Murphy’s discussion of different types of margin models, he states “we want them to ignore ‘temporary’ changes in volatility caused by noise or estimation error, and we want them to react quickly to permanent or structural changes in volatility. These two goals are in opposition to each other” (Murphy et al 2014). He delves further into the predicament and concludes that “if we want both to react fast to changes in volatility and to have stable margin when conditions are stable, the problem cannot be solved by the right choice of decay parameter alone. We have to react fast in order to respond quickly enough to what might turn out to be an important change in the volatility environment, but doing that necessarily makes the model fairly procyclical. Hence, some procyclicality mitigation measures – such as a margin buffer – are required.” Ultimately, even if the model properly captures volatility, additional components are required to prevent it from overreacting to new patterns. Incorporating a higher degree of human judgment can temper a margin framework’s reactions to volatility based on the likelihood of it continuing, allowing it to avoid overreaction. Further, expert real-time judgment on the specific effects of margin increases based on a CCP’s unique portfolios during times of stress can be invaluable in dampening market impact while balancing the need for more margin from increased volatility.

The fall of Long-Term Capital Management (LTCM) provides an excellent example of the issues that come from assuming an infallible risk model has been created. Despite LTCM’s assembled intellectual firepower, it did not predict the Russian default or Asian financial crises, and as a result its model reacted poorly to unanticipated market dislocations. This led to very heavy losses and a US$3.6 billion bailout, organized by the Federal Reserve Bank of New York and footed by Wall Street’s biggest players, who were the firm’s main creditors (Greenspan 2007). Although the LTCM crisis now seems small relative to the losses of the 2008 financial crisis, a greater reliance on data-driven models and its outcomes resulted in an underestimation of the risk of complex structured securities and derivatives on residential mortgages preceding the crisis. Models used by several participants in the market, including investment banks and ratings agencies, to assess the risk of the default of tranches of mortgage-backed securities failed to predict the large changes in value driven by a higher correlation in defaults in the underlying mortgages. This caused enormous losses as mortgage-backed security prices plummeted and credit default swap payouts were triggered, affecting Lehman Brothers, Bear Sterns, AIG and government-sponsored agencies, among others. This ultimately resulted in an over-US$700 billion bailout, this time funded by the American taxpayer (New York Times 2011). Risk models alone were certainly not responsible for the 2008 financial crisis; however, the crisis does demonstrate the possible shortcomings of overreliance and overdependence on data-driven model outputs, and it makes the case for applying expert judgment in complementing the output of an autonomous risk model. Autonomous models cannot always incorporate more nuanced market indicators that require expert judgment for reasonable interpretation; they generally take market data at face value. In the absence of qualitative oversight, this can lead to incomplete or insufficient risk assessment.

4 The benefits of margin models that incorporate expert judgment

There are a variety of benefits to employing expert judgment to complement model-driven analysis in margin setting. As noted previously, certain market factors such as seasonality and event risk are extremely difficult to correctly model. Moreover, certain markets lend themselves more toward margin frameworks that utilize more expert judgment than autonomous modeling, such as natural gas calendar spread markets. High seasonality and unique market conventions, along with large gaps in volatility profiles driven by weather, make this an extremely difficult market to fully model in an automated and consistent way. Margin frameworks incorporating expert judgment give risk managers the flexibility to utilize a variety of data-driven models and control margin changes ahead of binary events, such as those mentioned above. The enhanced flexibility afforded by the forward-looking aspect of these margin frameworks allows risk managers to take steps to be prepared for possible volatility changes due to these events.

In addition to addressing hard-to-model risk factors, margin frameworks that employ expert judgment produce relatively stable margin levels by facilitating limited increases or reductions rather than matching volatility shifts completely. Autonomous margin models that rely on data to determine margin levels produce margins that change daily. That being said, such models often incorporate anti-procyclical features, such as moving averages and volatility floors, to dampen reactiveness. Nonetheless, changing outputs can reduce the predictability of margins, and compensating tools such as real-time margin estimates may be required to give greater visibility to the potential costs associated with a cleared portfolio. In times of interchanging volatility clustering, autonomous model frameworks will typically be very reactive to increases in volatility, as well as decreases in volatility, which could easily lead to hyper-reactive margin requirements. An example of this can be found in the 2014 polar vortex, which enveloped much of the northern United States in arctic temperatures. Unsurprisingly, these frigid temperatures led to huge volatility in the natural gas and power markets, with some markets experiencing an order of magnitude increase. In this type of situation, a largely autonomous margin model framework could react rapidly to compensate for its lack of predictive power; some industry models did just that, resulting in a significant increase in outright margin rates, despite the aforementioned anti-procyclicality tools. The injection of expert judgment and a more real-time and comprehensive assessment of the larger market is critical in these once-in-a-decade-type situations. This is true for all markets exhibiting atypical volatility patterns. A broad array of factors may cause unanticipated price shifts, which, given automated margin models’ heavy reliance on historically observed activity, results in a failure to capture such factors. As was concluded in a Bank of England Working Paper, “different tools have different behaviour in different dimensions. The ten-year value-at-risk (VaR) floor, for instance, is good at mitigating across the cycle procyclicality, but it will do nothing to reduce the impact of large margin calls when margin is already over the floor level” (Murphy et al 2016).

A final benefit of margin frameworks employing a measured amount of expert judgment and human intervention is the ability to review input data for sensibility. Since, to a large extent, a model is only as good as the quality of its input data, bad data can lead autonomous model-based margin-setting frameworks to produce erroneous results. For instance, a bad settlement price in a contract for one day could lead an autonomous model to hike margins significantly, even though the price move was erroneous. Margin models that employ expert judgment on a daily basis are more likely to catch and ignore bad data points such as these, leading to more realistic and stable margin levels.

5 The risks of margin models that incorporate expert judgment

One of the drawbacks of margin frameworks that rely on expert judgment is the need for skilled risk managers to monitor, maintain and manage margins for a large set of products. In addition to the need for a relatively large number of resources, skill and experience are important factors in determining the effectiveness of any margin model. However, even the most qualified of risk managers are prone to human error; this may be exemplified in a number of ways, whether through bias or the misjudgment of anticipated market events. Experts have no more objective insight into future events than quantitative models, and, like a quantitative model, they may react solely on the basis of historical observations, whether right or wrong. Moreover, a large determinant of the effectiveness of an expert judgment-based framework is the risk culture of that institution. It is critical for risk functions to operate openly and independently, and to illicit and share meaningful information within the function itself as well as the broader organization. The absence of these characteristics or appropriate governance can negate the additive value of expert judgment in a margin framework.

The inherent nature of approaches that incorporate expert judgment makes them less objective and, consequently, less transparent. For example, consider the case of the 2014 polar vortex. If margin levels for an effected market are set to be 140% of the model’s output in advance of the winter months, market participants may require greater transparency around the decision to select such a high margin level. Expert judgment does include a subjective component that overrides the results of a purely data-driven model. This may present a challenge for exact replicability; however, market participants often prefer certainty and predictability when it comes to initial margin on a daily basis. In autonomous models, one can derive the margin levels given the input scenarios. However, if the scenarios are calibrated every day, then the margin output changes on a daily basis as well, resulting in loss of predictability. In both cases, CCPs can address the desire for predictability by providing easy-to-use online and real-time tools that allow market participants to anticipate their margins in advance.

6 Identifying the most prudent approach

With prudence and stability being the chief pillars of a given margin framework, CCPs must also consider transparency and predictability when designing a margin framework. Many CCPs achieve transparency and predictability by disseminating margin rates or model parameters to their clearing participants and clients. While less common, some CCPs even offer tools for calculating requirements for future or hypothetical portfolios. These practices increase the transparency and predictability of margin requirements and allow market participants to understand the funding effects of changing positions.

As global standard-setting bodies have undertaken regulatory reform for central clearing, the varying levels of prescriptive regulatory requirements in different jurisdictions have severely constrained CCPs’ ability to manage risk with the same level of historically proven success. For example, in April 2017, ESMA published its opinion, confirming the regulation, which states that “where portfolio margining covers multiple instruments, the amount of margin reductions shall be no greater than 80% of the difference between the sum of the margins for each product calculated on an individual basis and the margin calculated based on a combined estimation of the exposure for the combined portfolio.” (European Securities and Markets Authority (2017)) This appears to call for some type of intervention in the margin model, which may be challenging for automated models. The publication delineates a fairly strict rule with regard to a CCP’s ability to provide offsets to its clearing members and clients, but it offers little analytical evidence that the level is in fact appropriate.

This has resulted in a move away from an expert judgment-based approach to margining, and has driven many CCPs to adopt fully automated model-based margin setting. However, certain models fit some product complexes more appropriately than others. For example, financial products, such as interest rate swaps, tend to have characteristics that are more easily captured by a quantitative model than commodities, such as natural gas, which often have seasonality components that can be more easily captured by models that incorporate expert judgment. All margin frameworks, however, regardless of product class, are exposed to risks that cannot be modeled. As previously mentioned, to appropriately evaluate the models used to account for these complexities, regulations should reflect an outcomes-based approach, rather than a singular focus on the margin framework itself or a bias toward a particular type of model.

Automated models and expert judgment should be viewed as complementary rather than opposing, with each compensating for the other’s shortcomings. CCPs should design results-based frameworks with the following principles in mind: prudence, stability, transparency and predictability. Models provide unbiased, statistical feedback to help inform appropriate margin levels through the use of historical and forward-looking data; as such, past observations, and any volatility changes they may predict, do not account for all risks that margin is intended to cover. By way of example, a VaR model with a ten-year lookback period and a 99% confidence interval would have triggered a margin decrease for the British pound FX futures contract during the period of volatility that resulted from the UK referendum. Given the significant volatility in multiple products observed following the vote, relying solely on model-driven outputs would have resulted in substantial margin undercoverages and exposed CCPs to uncollateralized risks. Expert judgment helps to mitigate risks posed by these scenarios by allowing CCPs to incorporate an additional margin buffer to account for volatility that may not be predicted by an automated model. Events similar to the UK referendum, such as the US election or Organization of the Petroleum Exporting Countries (OPEC) meetings, cannot be considered solely from a quantitative perspective, as models cannot accurately predict all outcomes and subsequent volatility in the market.

7 Challenges in implementing the right framework

Notwithstanding the significant efforts by global CCPs to enhance their margining practices, designing a framework that is simultaneously prudent, stable, transparent and predictable is not a simple task. As the prominence of central clearing has increased, so too has the breadth of products available. A one-size-fits-all framework cannot adequately capture the risk factors associated with these vastly different product complexes. Commodities behave differently than financials, often in ways that cannot be reflected in a purely quantitative margin model; but they also present too many complexities for us to rely solely on expert judgment. Heightened regulatory prescriptiveness often amplifies these challenges. The requirement to apply a multitude of rigid parameters makes it difficult to appropriately incorporate expert judgment, even where that approach is most prudent. CCPs must be allowed the flexibility to determine the most appropriate balance between expert judgment- and autonomous model-based frameworks, given that no single approach captures every risk evident in all clearing services. The role and level of expert judgment in a given margin model is market and CCP specific. Regulators should take an outcomes-based, principles approach to evaluating the sufficiency of CCPs’ margin frameworks in order to ensure that CCPs are not prohibited from exercising the necessary flexibility to properly manage their risk.

Declaration of interest

The authors are operators of a CCP. The authors alone are responsible for the content of the paper and do not represent the views of CME Group.


Special thanks to Lee Betsill, Sean Downey, Udesh Jha, Jeff Krugler and Abby Perry for their contribution to this paper.


European Securities and Markets Authority (2017). Portfolio margining requirements under Article 27 of Commission Delegated Regulation (EU) No 153/2013. “Opinion” Report, April 10, ESMA, p. 2. URL:

Greenspan, A. (2007). The Age of Turbulence: Adventures in a New World, pp. 193–195. Penguin.

Murphy, D., Vasios, M., and Vause, N. (2014). An investigation into the procyclicality of frisk-based initial margin models. Financial Stability Paper 29, Bank of England, p. 13. URL:

Murphy, D., Vasios, M., and Vause, N. (2016). A comparative analysis of tools to limit the procyclicality of initial margin requirements. Staff Working Paper 597, Bank of England, p. 22. URL:

New York Times (2011). Adding up the government’s total bailout tab. Article, July 24. URL:

Securities Review Committee (1988). The operation and regulation of the Hong Kong Securities Industry. Report, SRC.

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:

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 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: