Collateral Rule: Empirical Evidence from the CDS Markets

  • Agostino Capponi, Assistant Professor, Columbia University
Finance, euro, stacks of coins on tables with pen and calculator, panorama, background



Financial markets were at the center of the Great Recession. The interaction between volatility, leverage, and collateralization served to amplify fundamental shocks, and contributed to the creation of self-reinforcing death spirals experienced by major financial institutions. Largely in response to the crisis, a vast literature has analyzed the role of the collateral channel in amplifying financial shocks. The typical mechanism works as follows: after a fundamental shock, an increase in required collateral (due to the increased risks or to the losses incurred by market participants) may force liquidation and deleveraging, and generate additional downward price pressure, thus producing a margin spiral widely studied in the theoretical literature. At the core of models for the collateral feedback channel is the collateral rule that determines how margins are set and how they respond to changes in individual portfolio risks as well as aggregate risks. The literature has employed a variety of models of margining, from an exogenously specified function like Value at Risk to an endogenous collateral requirement. 

Despite the centrality of the margining rule for theoretical models of financial crises and despite the fact that different collateral rules have different implications for the propagation and amplification of shocks through the financial system – empirical evidence on it is scarce: the collateral feedback channel has so far been studied mostly in theoretical work. 

We will fill this gap by providing the first comprehensive study of the collateral rule in the market of cleared credit default swaps: we study how margins depend on portfolio risks and market conditions, and what the implications are for theoretical models of collateral equilibrium.


Clearinghouses have significant discretion over modeling assumptions and parameters used to generate and justify margin requirements. They set them taking into account market conditions, the demand for trading, and collateral quality. In practice, margining rules involve a wide range of scenarios and simulations to arrive at a portfolio loss distribution, requiring the clearinghouses to make various modeling and statistical assumptions. 

Most clearinghouses, including ICC state that their margins are broadly “set to cover five days of adverse price/credit spread movements for the portfolio positions with a confidence level of 99%”, which we refer to as a 5-day 99% Value-at-Risk (VaR) margining rule. Even if clearinghouses were restricted to using VaR based margining rules, the confidence level, margin period risk, and the distributional assumption of losses are inputs that give the clearinghouse significant freedom in setting the actual margin levels.

Overall, clearinghouses employ complex rules to determine the amount of required margins. These rules make it hard to understand what are the main economic determinants of collateral requirements, partly because they are complex and depend on the interactions of a myriad of variables and calibration choices, partly because they do not explicitly take into account variables that, however, may still matter indirectly: for example, aggregate volatility or default risks do not directly enter the calculations, but may still affect the collateral rule empirically because of their effect on the scenarios the clearinghouse uses to simulate portfolio losses, or through the choice of discretionary parameters.


We use statistical hypotheses testing and multivariate linear regression techniques, executed both in the time series and in the cross-section, to estimate the portfolio-level determinants of collateral requirements. We estimate a modified version of the margining model proposed by Duffie, Scheicher and Villeumey (2016), where initial margins are determined as a mix of two specific portfolio variables: maximum shortfall and short notional (the so-called short charge). Their model parameters were calibrated to anecdotal evidence; we instead estimate them from actual data. 

We incorporate both market variables and portfolio variables into our panel regressions: since collateral rules adapt to market conditions, we should expect collateral levels to respond to variables that capture the state of the economy and the markets. We also enlarge the set of predictors in the regressions by incorporating measures of aggregate risk, such as VIX and the average CDS spread of all dealers, and measures of funding opportunity costs, including the CDS spread of each member (that reflects the credit risk of the member, and in turn affects its cost of financing) and the LIBOR-OIS (Overnight Index Swap) spread.


The main findings of this work include the following:

  • Provides direct empirical evidence that margins are much more conservatively set than what a Valueat-Risk (VaR) rule would imply, and are unequally implemented across participants.
  • Shows that more extreme tail risk measures have a higher explanatory power for observed collateral requirements than VaR, consistent with endogenous collateral theories where extreme events dominate in determining collateral. The dependence of collateral requirements on extreme tail risks induces potential nonlinearities in margin spirals, dampening small shocks and amplifying large ones.
  • We confirm empirically and quantify the main channel through which collateral- feedback effects operate in many theoretical models of equilibrium with financial frictions, highlighting the prominent role of aggregate volatility and funding costs.

The author is an independent contributor to the Global Risk Institute and is solely responsible for the content of this article.