Author Archives: Haoxiang Zhu

Designing More Efficient CDS Auctions

Professor Haoxiang Zhu

Credit default swaps (CDS) are an important derivative class. According to the Bank for International Settlements, as of June 2013, CDS contracts have $24 trillion notional amount outstanding and $725 billion market value globally.

A CDS contract is a default insurance contract written on a firm, loan or sovereign country. Buyers of protection (CDS buyers) pay periodic premiums on a notional amount of debt to sellers of protection (CDS sellers), until the contract expires or default occurs, whichever is earlier. For example, if a bond investor wishes to insure against the default of $100 million corporate bonds, and a five-year CDS contract on that bond is quoted at 500 basis points (5%) per year, then the CDS buyer pays $5 million per year to the CDS seller for five years. If the bond defaults within five years, the CDS seller pays the CDS buyer the loss given default. The question is: What are the recovery value and default compensation?

The market uses CDS auctions to determine the recovery value of a defaulted bond. For example, the auction-determined recovery rate of Greece debt is 21.5 cents per euro. So, CDS sellers pay CDS buyers 78.5 (=100-21.5) cents per euro of debt insured. In addition to settling CDS contracts, CDS auctions also give investors an opportunity to trade defaulted bonds at zero bid-ask spread. Given the sheer size of CDS markets and high-profile defaults in the recent recession, it is important that CDS auctions deliver unbiased prices and efficient allocations. But do they?

In a recent research paper with Professor Songzi Du of Simon Fraser University, we find that the current design of CDS auctions leads to systematically biased prices and inefficient allocations.

To understand why the auction design is biased and inefficient, we need to understand the auction procedure itself. A CDS auction consists of two stages. In the first stage, investors who have CDS positions submit market orders (called “physical requests”) to buy or sell the defaulted bonds. An investor’s market order on the bond must be in the opposite direction of his CDS position, and no larger in magnitude. The sum of these market orders, called “open interest,” is sold in the second stage, which is a uniform-price auction. Importantly, only one-sided limit orders are allowed in the second stage. If the open interest is to buy, only limit sell orders are allowed; if the open interest is to sell, only limit buy orders are allowed. The market-clearing price in the second stage is the “official” recovery rate of the defaulted bond for settling CDS.

From 2006 to November 2013, this auction procedure has settled more than 140 defaults, including those of Lehman Brothers, Fannie Mae, General Motors, and Greece, among others.

Through a formal auction model, we show that the biased design comes from the restrictions imposed on the two stages of CDS auctions. To get the intuition, consider a CDS buyer. A CDS buyer naturally wishes to sell the defaulted bonds to minimize the uncertainty in the auction final price; he can eliminate this price risk by selling an amount equal to his CDS position. If this CDS buyer also has a low value for owning the bonds (for information or hedging motives), he would want to sell more. But the auction procedure forbids him from selling these additional quantities. A supply to sell is therefore suppressed in the first stage. In the second stage, this supply is suppressed again if the open interest is to sell (as only buy limit orders are allowed). Information from a suppressed demand cannot come into the price. Therefore, price is systematically biased, and allocations of bonds are inefficient.

This is not the end of the story. In an earlier version of the same research paper, we show that, if CDS traders are large, they also have a strong incentive to manipulate the final price auction price—to get favorable settlement payments on their CDS positions. Manipulation also leads to price biases.

The administrators of CDS auctions are aware of the biased prices; as a remedy, they impose a price cap or a price floor, depending on dealers’ quotes and the direction of the open interests. But this measure is imperfect and can backfire. We find that although a price cap or floor can correct price biases, it can also make bond allocations even less efficient.

What is a better solution then? We show that a simple, unconstrained double auction delivers better price discovery and allocative efficiency. A double auction for settling CDS is similar to the open and close auctions on stock exchanges. Since double auctions have done well in equity markets, why not consider it for CDS auctions?

For details of this research, see Songzi Du and Haoxiang Zhu, “Are CDS Auctions Biased and Inefficient?”. An earlier version of this paper is summarized by FT Alphaville (part 1, part 2).

Haoxiang  Zhu is Assistant Professor of Finance at the MIT Sloan School of Management

“Dark Pools” can improve price discovery in open exchanges

When big investors want to execute trades but fear the size of the transaction could move the market, they often go to dark pools—alternative trading systems where orders are not publicly displayed. These opaque trading venues, now accounting for about 12 percent of equity trading volume in the United States, have sparked concern among regulators and in the financial press. With so many transactions occurring out of public view, critics warn that price discovery, the accurate determination of asset prices, will become more difficult.

Professor Haoxiang Zhu

But despite their sinister sounding name, dark pools can actually improve price discovery in open exchanges, I have found in my research, “Do dark pools harm price discovery?” Open a dark pool alongside a ‘lit’ pool, such as the New York Stock Exchange or Nasdaq, where orders are visible to all, and pricing can become more accurate on the open exchange.

To understand why this happens, it helps to consider investors to be divided loosely into two groups: informed and uninformed. Informed investors examine balance sheets, study analyst reports, read company press releases, and use advanced technology to monitor the market. When informed investors decide to execute a trade, they hope to profit from the information they gathered–and quickly.

Uninformed investors trade mainly for liquidity reasons. They may need to get cash to meet an obligation, or they may have cash they need to invest. They want to make their transaction regardless of a company’s fundamentals.

A dark pool presents different execution risks to informed and uninformed investors. This risk arises because a dark pool relies on matching, rather market makers, to execute trades. For example, if a dark pool has 300 shares to buy and 200 shares to sell, then only two thirds of each buy order is executed. Failure of execution is costly. (On an open exchange, those extra 100 shares to buy would be executed by market makers or liquidity providers). Informed investors have a high execution risk in the dark pool because they tend to trade in the same direction. Uninformed traders have a lower execution risk because their orders tend to be balanced on either side of the market.

This difference in execution risk tends to drive a greater proportion of informed investors to the open exchange and a larger share of uninformed investors to the dark pool. Having more informed investors trading on an exchange improves price discovery in the exchange and yield more accurate asset prices—precisely the opposite of what critics of dark pools fear.

While having non-displayed orders can help price discovery, dark pools are opaque in other, potentially harmful, ways. For example, dark pools typically do not disclose how they operate, and investors and the public often don’t know how the venues set execution prices or process orders. Increased oversight and closer scrutiny of dark pools could well be a very good thing for financial markets. But not displaying orders in itself needs not threaten price discovery in transparent venues. Instead, it could help.

Haoxiang  Zhu is Assistant Professor of Finance at the MIT Sloan School of Management