Are CDS Auctions Biased and Inefficient?

• DOI: http://doi.org/10.1111/jofi.12541
• Published date: 01 December 2017
• Date: 01 December 2017
• Author: HAOXIANG ZHU,SONGZI DU

THE JOURNAL OF FINANCE •VOL. LXXII, NO. 6 •DECEMBER 2017

Are CDS Auctions Biased and Inefficient?

SONGZI DU and HAOXIANG ZHU∗

ABSTRACT

We study the design of credit default swaps (CDS) auctions, which determine the

payments by CDS sellers to CDS buyers following defaults of bonds. Using a simple

model, we find that the current design of CDS auctions leads to biased prices and

inefficient allocations. This is because various restrictions imposed in CDS auctions

prevent certain investors from participating in the price discovery and allocation

process. The imposition of a price cap or floor also gives dealers large influence on

the final auction price. We propose an alternative double auction design that delivers

more efficient price discovery and allocations.

CREDIT DEFAULT SWAPS (CDS) ARE DEFAULT insurance contracts between buy-

ers of protection (“CDS buyer”) and sellers of protection (“CDS seller”) that

are written against the default of firms or countries. Since the financial cri-

sis, CDS have been among the top financial innovations to receive policy at-

tention and regulatory actions.1As of December 2015, global CDS markets

∗Songzi Du is with Simon Fraser University. Haoxiang Zhu is with Massachusetts Institute of

Technology and NBER. An earlier version of this paper was distributed under the title “Are CDS

Auctions Biased?” For helpful comments, we are grateful to Bruno Biais (Editor), the Associate Ed-

itor,two anonymous referees, Jeremy Bulow, Mikhail Chernov (discussant), Darrell Duffie, Vincent

Fardeau (discussant), Jacob Goldfield, Alexander Gorbenko (discussant), Jean Helwege, YesolHuh,

Ron Kaniel, Ilan Kremer, Yair Livne, Igor Makarov, Andrey Malenko, Ahmad Peivandi, Lisa Pol-

lack, Michael Ostrovsky,Rajdeep Singh (discussant), Ken Singleton, Andy Skrzypacz, Rangarajan

Sundaram, Heather Tookes, Bob Wilson,Anastasia Zakolyukina, Adam Zawadowski (discussant),

and Ruiling Zeng. We also thank seminar participants at Stanford University, University of South

Carolina, Michigan Ross, the ECB-Bank of England Workshop on Asset Pricing, Econometric So-

ciety summer meeting, Financial Intermediary Research Society annual meeting, SIAM, NBER

Summer Institute Asset Pricing meeting, European Finance Association annual meeting, Society

for the Advancement of Economic Theory meeting, and Western Finance Association annual meet-

ing. We are also grateful to a number of senior executives at ISDA for their comments and insights.

Haoxiang Zhu has been affiliated with the CFTC (the U.S. regulator of index CDS) as an unpaid

outside academic economist since July 2016, after the analysis in this paper was completed. The

authors have read the Journal of Finance’s disclosure policy and have no conflict of interest to

disclose.

1For example, the Dodd-Frank Act of the United States mandated and later implemented

mandatory central clearing of standard over-the-counter derivatives, including CDS. Furthermore,

in its Financial Markets Infrastructure Regulation, the European Commission stated that “Deriva-

tives play an important role in the economy but are associated with certain risks. Recent financial

crises have highlighted that these risks are not sufficiently mitigated in the over-the-counter (OTC)

part of the market, especially in regards to credit default swaps (CDS).” See “Financial stability:

New EU rules on central clearing for certain credit derivative contracts,” European Commission

DOI: 10.1111/jofi.12541

2589

2590 The Journal of Finance R

have a notional outstanding of $12.3 trillion and a gross market value of

$421 billion.2

This paper studies the design of CDS auctions, a unique mechanism that

determines postdefault recovery value for the purpose of settling CDS. Since

recovery value is a fundamental parameter for the pricing, trading, and clearing

of CDS contracts, achieving a fair and unbiased auction price is crucial for

the proper functioning of CDS markets. For example, using a sample of U.S.

corporate bond defaults, Gupta and Sundaram (2012) find that “information

generated in CDS auctions is critical for postauction market price formation.”

In addition to price discovery, an auction protocol also has the key benefit of

facilitating cash settlement by producing a transparent price, which overcomes

the difficulty of physical settlement when the outstanding amount of CDS

exceeds the supply of bonds (see Creditex and Markit (2010)).3

The current CDS auction mechanism was initially implemented in 2005,

and in 2009 it became the standard method used to settle CDS contracts after

default (ISDA (2009)). Over the 2005 to May 2016 period, CDS auctions settled

121 defaults of firms (such as Fannie Mae, Lehman Brothers, and General

Motors) and sovereign countries (such as Greece, Argentina, and Ukraine). For

many firms, separate CDS auctions are held for senior and subordinate debt.

As we explain in Section I, the current auction mechanism consists of two

stages. The first stage involves soliciting market orders, called “physical settle-

ment requests,” to buy or sell defaulted bonds. The net market order is called

the “open interest.” At the same time, dealers quote prices on the bonds used to

calculate a price cap or floor. The second stage involves a (one-sided) uniform-

price auction subject to the price cap or floor. If the final auction price is p∗per

$1 of face value, the default payment by CDS sellers to CDS buyers is 1 −p∗.

The primary objective of this paper is to evaluate CDS auctions from a the-

oretical and market design perspective. We show that the current design of

CDS auctions leads to biased prices and inefficient allocations of defaulted

bonds. The primary reason is that various restrictions imposed on the auctions

prevent certain investors from participating in the price discovery and alloca-

tion process. Moreover, the current design leaves ample scope for dealers to

strategically manipulate the price to profit from their existing CDS positions.

We suggest a double auction design that delivers more efficient prices and

allocations.

Our analysis builds on a simple model of divisible auctions that works

roughly as follows. There are two dates, t∈{0,1},and a continuum of infinites-

imal traders. At t=0, everyone has the same probability distribution over the

default of a risky bond. Traders are endowed with i.i.d. private values {bi}for

Press Release, March 1, 2016. In 2012, European regulators banned “naked” sovereign CDS in the

European Union.

2See the latest semiannual OTC derivatives statistics, Bank for International Settlements,

http://www.bis.org/statistics/derstats.htm.

3As noted by Creditex and Markit (2010), in the physical settlement of CDS, if the outstanding

amount of CDS exceeds the supply of defaulted bonds, the bonds need to be recycled multiple times

to settle all CDS contracts, leading to endogenous scarcity of the bonds and artificially high prices.

Are CDS Auctions Biased and Inefficient? 2591

buying or selling CDS on the bond. Each trader incurs a per-unit cost in trad-

ing CDS contracts and a quadratic inventory cost in holding CDS positions.

The CDS positions are determined optimally in a double auction, taking into

account the actions at t=1. At t=1, with some positive probability, the bond

defaults and traders are endowed with high or low valuations {vi}for owning

the defaulted bonds. Immediately afterward, a CDS auction is held. Traders

select the optimal physical requests in the first stage and the optimal demand

schedules in the second stage. Like CDS positions, traders also incur quadratic

costs in buying or selling defaulted bonds in the auction. Optimal strategies in

the CDS auction as well as predefault CDS trading are solved in a subgame-

perfect equilibrium. For simplicity, our model is solved without imposing the

price cap or floor in the second stage.

To see why the price is biased and allocations are inefficient, consider the

strategy of a trader who has a high value for owning the defaulted bonds at

t=1. The trader’s CDS position could be positive (CDS buyer), negative (CDS

seller), or zero. For simplicity, let us consider a high-value trader with zero

CDS position. In practice, this trader could be a specialist in distressed debt

who does not trade CDS. This high-value trader wishes to buy defaulted bonds,

but the current design of the CDS auction stipulates that only CDS sellers can

submit physical requests to buy (i.e., market orders to buy) in the first stage.

Thus, demand to buy bonds from this high-value trader is suppressed in the

first stage of the auction. If the open interest is to buy, the auction protocol also

stipulates that only orders in the opposite direction, that is, sell limit orders,

are allowed in the second stage. Thus, demand to buy bonds from this high-

value trader is suppressed in the second stage of the auction as well. The same

argument applies to a high-value CDS buyer. As a result, if the open interest

is to buy, high-value traders with positive or zero CDS positions are excluded

from the auction, leading to a downward-biased final auction price.4

This result applies symmetrically if the open interest is to sell, but in this case

it is the low-value traders with zero or negative CDS positions who are excluded

from both stages of the auction. Thus, the final auction price is upward-biased

if the open interest is to sell.

Inefficient allocations naturally follow from biased prices. The directions of

the inefficiency are summarized in Table I. In particular, high-value traders’

allocations are almost always too low (with the only exception being high-value

CDS sellers in the high state), and low-value traders’ allocations are almost

always too high (with the only exception being low-value CDS buyers in the

low state).

The restrictions imposed in CDS auctions also have an unintended conse-

quence in CDS markets before default. Because a larger CDS position (in abso-

lute value) relaxes participation constraints in the first stage of CDS auctions,

4We can show that other types of traders—including low-value traders (regardless of CDS

positions) and high-value CDS sellers—either fully or partially participate in either stage of the

auction.