© 2010 Interactive Data Pricing and Reference Data, Inc. 1 WHITE PAPER Corporate Bond Trading Costs During the Financial Crisis Peter Ciampi, Interactive Data Corporation Eric Zitzewitz, Dartmouth College June 2010 Abstract We exploit the addition of a direction of trade variable to FINRA’s TRACE™ system to measure effective bid- ask spreads for corporate bonds. We find significantly larger spreads for smaller trades, lower quality bonds, longer duration bonds, and during the “crisis” period of late 2008 and early 2009. We find higher spreads than reported by other studies for earlier time periods. We also document a pairing of dealer-client and interdealer trades, suggesting that many client trades, especially smaller trades, are intermediated by a dealer who earns a spread for matching the client with the dealer who is the ultimate source of supply or demand. Introduction Whereas the study of the execution quality of equity trades has been facilitated by datasets such as the New York Stock Exchange’s Trades and Quotes data, no comparable dataset exists for corporate bonds, which trade largely in a dealer market rather than on an exchange. From 2002-6, the TRACE reporting system began disclosing transaction prices, but no comparable publicly available dataset exists for quotes. In November 2008, TRACE added a “reporting party side” variable that discloses whether a trade involved a dealer that was purchasing from a non-dealer client (B), selling to a non-dealer client (S), or trading with another dealer (D). In this note, we exploit this variable to estimate effective bid-ask spreads. We find spreads that are significantly larger than those estimated in past work, particularly during the last two months of 2008 and the first half of 2009. During our sample period, the median retail-sized trade (defined as under $100,000 in face value) pays a two-way effective bid-ask spread of 152 basis points (bp), while the average institutionally-sized trade (greater than or equal to $500,000 in face value) pays 31 bp. Mean spreads are 1.5-2 times larger than median spreads: 231 bp for retail-sized and 53 bp for institutionally-sized trades. Larger trading costs for smaller trades are also found for municipal and Treasury bonds (Chakravarty and Sarkar, 2003; Harris and Piwowar, 2006) but not equities (e.g., Lin, Sanger, and Booth, 1995). The closest prior study to ours is Edwards, Harris, and Piwowar (2006), which estimates mean and median spreads for the TRACE sample as a parametric function of trade size. 1 Their estimated median two-way spreads from January 2003 to January 2005 range from 66-120 bp for retail-sized and 2-20 bp for institutionally-sized trades; for mean spreads the analogous ranges are 92-150 bp and 8-28 bp (see their Table IV). Our spreads for the time period November 2008 to April 2010, estimated using a similar method, are around twice as large. Even if we restrict attention to the second half of our sample, which had lower spreads than the crisis period at the beginning, we still find substantially higher trading costs than in 2003-5. 1 Edwards, Harris, and Piwowar (2006) had access to a reporting party side variable in TRACE for their earlier time period. Goldstein, Hotchkiss, and Sirri (2006) also had access to this variable for their study of the BBB-rated corporate bonds that were added to TRACE in 2003 as part of a controlled experiment. Other work on corporate bond trading costs includes Hong and Warga (2000), Schultz (2001), Chakravarty and Sarkar (2003), and Bessembinder, Maxwell, and Venkataraman (2006). These papers use the National Association of Insurance Commissioners (NAIC) data on bond transactions by insurance companies. Bessembinder, Maxwell, and Venkataraman report that the NAIC data accounted for 12.5% of the dollar trading volume on TRACE in the second half of 2002 (p. 263). The studies on NAIC find smaller trading costs than reported by Edwards, Harris, and Piwowar, likely reflecting the exclusively institutional nature of their sample.