Three Essays on Corporate Bond Market Liquidity

Dick-Nielsen, Jens Doctoral Thesis Three Essays on Corporate Bond Market Liquidity PhD Series, No. 33.2010 Provided in Cooperation with: Copenhagen Business School (CBS) Suggested Citation: Dick-Nielsen, Jens (2010) : Three Essays on Corporate Bond Market Liquidity, PhD Series, No. 33.2010, ISBN 9788759384473, Copenhagen Business School (CBS), Frederiksberg, https://hdl.handle.net/10398/8198 This Version is available at: https://hdl.handle.net/10419/208766 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-nd/3.0/ The PhD School of Economics and Management PhD Series 33.2010 PhD Series 33.2010 Three Essays on Corporate Bond Market Liquidity copenhagen business school handelshøjskolen solbjerg plads 3 dk-2000 frederiksberg denmark www.cbs.dk ISSN 0906-6934 ISBN 978-87-593-8447-3 Three Essays on Corporate Bond Market Liquidity Jens Dick-Nielsen CBS PhD nr 33-2010 Jens Dick-Nielsen · A4 OMSLAG.indd 1 01/11/10 12.15 Three Essays on Corporate Bond Market Liquidity Jens Dick-Nielsen Three Essays on Corporate Bond Market Liquidity 1st edition 2010 PhD Series 33.2010 © The Author ISBN: 978-87-593-8447-3 ISSN: 0906-6934 “The Doctoral School of Economics and Management is an active national and international research environment at CBS for research degree students who deal with economics and management at business, industry and country level in a theoretical and empirical manner”. All rights reserved. No parts of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or by any information storage or retrieval system, without permission in writing from the publisher. Jens Dick-Nielsen Three Essays on Corporate Bond Market Liquidity 1st edition 2010 PhD Series 33.2010 © The Author ISBN: 978-87-593-8447-3 ISSN: 0906-6934 “The Doctoral School of Economics and Management is an active national and international research environment at CBS for research degree students who deal with economics and management at business, industry and country level in a theoretical and empirical manner”. All rights reserved. No parts of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or by any information storage or retrieval system, without permission in writing from the publisher. Three Essays on Corporate Bond Market Liquidity Jens Dick-Nielsen Ph.D. Dissertation Department of Finance Copenhagen Business School August, 2010 ii ii Contents Preface 1 1 Liquidity Biases in TRACE 3 1.1 Introduction............................ 4 1.2 ReportingErrors ......................... 5 1.3 ErrorFilter ............................ 9 1.4 ErrorImpact ........................... 16 1.5 RemainingIssues......................... 19 1.6 Conclusion ............................ 23 2 Corporate Bond Liquidity Before and After the Onset of the Subprime Crisis 25 2.1 Introduction............................ 26 2.2 Literature review . . . . . . . . . . . . . . . . . . . . . . . . . 28 2.3 Datadescription ......................... 30 2.4 Empirical methodology . . . . . . . . . . . . . . . . . . . . . . 31 2.5 Liquiditypremia ......................... 34 2.6 Determinants of bond illiquidity . . . . . . . . . . . . . . . . 54 2.7 Conclusion ............................ 60 2.8 Appendix: Robustness checks . . . . . . . . . . . . . . . . . . 62 3 Index Driven Price Pressure for Corporate Bonds 67 3.1 PricePressure........................... 68 3.2 IndexTracking .......................... 69 3.3 The Lehman/Barclay Corporate Bond Index . . . . . . . . . 76 3.4 Measuring Abnormal Corporate Bond Returns . . . . . . . . 80 3.5 Maturity Less Than 1 year . . . . . . . . . . . . . . . . . . . 81 3.6 Newly Issued Bonds . . . . . . . . . . . . . . . . . . . . . . . 87 3.7 Downgraded Bonds . . . . . . . . . . . . . . . . . . . . . . . . 94 3.8 UpgradedBonds .........................100 3.9 Conclusion ............................106 iii iv Contents Summary 109 EnglishSummary............................109 DanskResum´e .............................113 Bibliography 117 iv Contents Summary 109 EnglishSummary............................109 DanskResum´e .............................113 Bibliography 117 Preface This thesis marks the end of my Ph.D studies in Finance at Copenhagen Business School. The thesis consists of three empirical studies on the liquidity of the corporate bond market. Each of the three essays in the thesis are self-contained with included literature reviews and can be read independently. Structure of the thesis The three essays study the US corporate bond market with special attention to bond liquidity. All essays are empirical studies which rely heavily on the availability of transactions data. Earlier studies had to use quoted bond prices for empirical studies, but with the introduction of the TRACE system and with the following dissemination of transaction prices the data quality on corporate bonds has improved immensely. In the years after 2000 a range of studies assessed the performance of structural credit risk models and found that they were not able to fully explain the size of the average credit spread for corporate bonds. Huang and Huang (2003) suggested (among others) that the remaining non-default-component of the credit spread was an illiquidity premium. Using transaction data this thesis studies the impact of illiquidity and trading frictions on corporate bonds. The first essay forms the basis of the following two essays and describes in detail how the data from TRACE should be handled and cleaned up before usage. Most other papers using the disseminated data from TRACE lack such a description, are doing an insufficient clean up procedure or have misunderstood how the errors accumulate in TRACE. Both Bloomberg and WRDS which both provide TRACE data fail to remove the majority of the errors. In the essay I present an error filter and show that it is able to clean out almost all errors. I also show how commonly used liquidity measures will be severely biased if nothing is done. The second essay (co-authored with Peter Feldh¨utter and David Lando) use the transaction data from TRACE to estimate various corporate bond liquidity measures and asses their performance before and after the onset of the subprime crisis. A linear combination of the Amihud price impact measure, a measure for roundtrip costs and the standard deviation of these 1 8Essay 1 the disseminated data as an as-of trade. This plays a role for reporting errors detected on a later date. If a broker has to cancel a report filed on an earlier date, this is done by filing a report identical to the original report but marked as a reversal7. A correction on a later date is done by first filing the reversal, thereby canceling the report containing the error, and then the broker has to file the right report but now marked as an as-of trade. In the disseminated data a reversal implies that we should delete both the original report and the reversal report. Table 1.2 shows an example of a report that has been canceled by a reversal and then followed up by an as- of trade report containing the correct information. In this case we cannot identify the original report by its message sequence number, because it is an intraday number and the reversal and the original are not filed on the same day. Instead the original report is identified as a report that is identical to the reversal (except that the original is not marked as a reversal). Note how small the error is in the original trade report compared to the correction both for the same-day correction and for the reversal. It is not clear that we could have identified any of these trades as being errors based on the prices or yields alone. In this sense most errors in TRACE are not outliers deviating from the surrounding reports. Bond ID Date Time Price Par volume Yield As-Of CSCO.GB 20060221 15:21:42 100.348 1000000 5.17050 CSCO.GB 20060221 15:21:42 100.348 1000000 5.17050 R CSCO.GB 20060221 15:21:42 100.348 1000000 5.17042 A Table 1.2: Disseminated Trade Reports with non-identical reversals. This table contains a typical example of how a reversal is disseminated in TRACE. The reversal report is an identical copy of the original report but with a ’R’ in the as-of indicator. Some reversals are followed up by an as-of report indicated by the ’A’ in the as-of indicator. The follow up report then contains the correct trade report information. The table only displays a selection of the disseminated information about each report. These are bond ID, Date, Time, Price, Par Volume, Yield and As-Of indicator. Failing to delete the error reports will yield a substantial amount of double counting8. We show in section 1.4 that this double counting will severely overstate the trading activity and depth of the market. 7See FINRA (2008) page 31. 8In a strategic analysis of the transaction prices it is debatable which of the reports to use but not without taking a stand on whether or not the market found the given report credible or not at the time of the filing of the original report. 8Essay 1 the disseminated data as an as-of trade. This plays a role for reporting errors detected on a later date. If a broker has to cancel a report filed on an earlier date, this is done by filing a report identical to the original report but marked as a reversal7. A correction on a later date is done by first filing the reversal, thereby canceling the report containing the error, and then the broker has to file the right report but now marked as an as-of trade. In the disseminated data a reversal implies that we should delete both the original report and the reversal report. Table 1.2 shows an example of a report that has been canceled by a reversal and then followed up by an as- of trade report containing the correct information. In this case we cannot identify the original report by its message sequence number, because it is an intraday number and the reversal and the original are not filed on the same day. Instead the original report is identified as a report that is identical to the reversal (except that the original is not marked as a reversal). Note how small the error is in the original trade report compared to the correction both for the same-day correction and for the reversal. It is not clear that we could have identified any of these trades as being errors based on the prices or yields alone. In this sense most errors in TRACE are not outliers deviating from the surrounding reports. Bond ID Date Time Price Par volume Yield As-Of CSCO.GB 20060221 15:21:42 100.348 1000000 5.17050 CSCO.GB 20060221 15:21:42 100.348 1000000 5.17050 R CSCO.GB 20060221 15:21:42 100.348 1000000 5.17042 A Table 1.2: Disseminated Trade Reports with non-identical reversals. This table contains a typical example of how a reversal is disseminated in TRACE. The reversal report is an identical copy of the original report but with a ’R’ in the as-of indicator. Some reversals are followed up by an as-of report indicated by the ’A’ in the as-of indicator. The follow up report then contains the correct trade report information. The table only displays a selection of the disseminated information about each report. These are bond ID, Date, Time, Price, Par Volume, Yield and As-Of indicator. Failing to delete the error reports will yield a substantial amount of double counting8. We show in section 1.4 that this double counting will severely overstate the trading activity and depth of the market. 7See FINRA (2008) page 31. 8In a strategic analysis of the transaction prices it is debatable which of the reports to use but not without taking a stand on whether or not the market found the given report credible or not at the time of the filing of the original report. Liquidity Biases in TRACE 9 1.3 Error Filter This section describes a simple algorithm that detects and deletes the reporting errors. Based on the disseminated transaction information it is not possible to set up a perfect filter, so we test the performance of the proposed error filter by comparing with the statistics from the official TRACE fact book. 1.3.1 Description The filtering of the reporting errors in the disseminated data takes place in three steps. 1Deleting true duplicates. In the disseminated data each report has an intra-day unique message sequence number. We delete any duplicates identified by the message sequence number9. 2Deleting reversals. Since reversals are typed in later than same-day corrections we start by deleting those which are newest in a chronological sense. All reports marked as a reversal are deleted and for each reversal we also delete the original report. Each reversal should exactly match one original report10. 3Deleting same-day corrections. There are two types of same-day corrections in the disseminated data. These can be identified by the trade status of the report. If the correction is a cancelation, both reports should be deleted and if it is a correction only the original should be deleted. Contrary to reversals the original can be identified through the original message sequence number which is given as part of the correcting report. 1.3.2 Stylized facts Even though the outline of the filter is quite simple in principle, it is not possible to actually implement it in that exact form. Particularly the second step of the algorithm can be problematic. When a broker files a reversal she 9A duplicate in this step indicates that two interdealer reports have been disseminated, even though only one should have been according to the description of the disseminated data. In the TRACE data disseminated though WRDS, this is no longer a big issue. However, it used to be. At some point in 2007 the database was altered and almost all these duplicates removed. Before that point in time almost 25% of all reports were duplicates of this type. 10It is not possible to partially reverse a trade report or to have a reversal canceling more than one original report. See FINRA (2008) page 31. 10 Essay 1 has to supply a 10-digit TRACE-assigned control number from the original trade allowing the SEC to keep a linkage between the reversal and the original report. The only way to make the same linkage within the disseminated data, between an original trade report and a reversal, is to rely on brokers filing the reversal as an exact replica of the original (as they should do according to guidelines11 and as we saw it in table 1.2). But not all reversals can be matched with an identical original report. In panel A of table 1.3 we have a reversal without any identical report but with a possible original report. These two reports differ on two variables. The differences may just be a matter of rounding, but more than one report may match if we were to round the numbers. In panel B we have another reversal with a possible original report. Again the two reports are not identical12. In these two examples the reversals differ on three variables all together and in the second example the original cannot be matched by rounding either the reversal or the original. This makes it hard to set up a rule for identifying an original report when the reversal does not match any reports exactly. Bond ID Date Time Price Par volume Yield As-Of Panel A CSCO.GB 20060412 11:39:03 99.250 3000000 5.42600 CSCO.GB 20060412 11:39:00 99.250 3000000 5.42635 R Panel B CSCO.GB 20060215 13:45:33 99.809 1MM+ 5.29398 CSCO.GB 20060215 13:45:33 99.809 2000000 5.29398 R Table 1.3: Disseminated Trade Reports with non-identical reversals. This table contains two examples of reversals for which the original trade report is not identical to the reversal. This makes the identification of the original report hard or even impossible. The table only displays a selection of the disseminated information about each report. These are bond ID, Date, Time, Price, Par Volume, Yield and As-Of indicator. In the TRACE User Guide it is stressed that one reversal report can only cancel one original report. This may be the rule, but in some cases as in table 1.4 it could be, that the reversal is meant to cancel more than just one report, since the reversal matches four other reports. It is rather unusual in the data to have more trades in the same second with matching prices 11According to FINRA (2008) page 31 a reversal should be exactly identical to the report it is reversing. 12The second example in table 1.3 is due to a rating change between the date of the first report and the date of the reversal report. Trading volumes are censored for investment grade bonds at $5 millions and for speculative grade at $1 million. 10 Essay 1 has to supply a 10-digit TRACE-assigned control number from the original trade allowing the SEC to keep a linkage between the reversal and the original report. The only way to make the same linkage within the disseminated data, between an original trade report and a reversal, is to rely on brokers filing the reversal as an exact replica of the original (as they should do according to guidelines11 and as we saw it in table 1.2). But not all reversals can be matched with an identical original report. In panel A of table 1.3 we have a reversal without any identical report but with a possible original report. These two reports differ on two variables. The differences may just be a matter of rounding, but more than one report may match if we were to round the numbers. In panel B we have another reversal with a possible original report. Again the two reports are not identical12. In these two examples the reversals differ on three variables all together and in the second example the original cannot be matched by rounding either the reversal or the original. This makes it hard to set up a rule for identifying an original report when the reversal does not match any reports exactly. Bond ID Date Time Price Par volume Yield As-Of Panel A CSCO.GB 20060412 11:39:03 99.250 3000000 5.42600 CSCO.GB 20060412 11:39:00 99.250 3000000 5.42635 R Panel B CSCO.GB 20060215 13:45:33 99.809 1MM+ 5.29398 CSCO.GB 20060215 13:45:33 99.809 2000000 5.29398 R Table 1.3: Disseminated Trade Reports with non-identical reversals. This table contains two examples of reversals for which the original trade report is not identical to the reversal. This makes the identification of the original report hard or even impossible. The table only displays a selection of the disseminated information about each report. These are bond ID, Date, Time, Price, Par Volume, Yield and As-Of indicator. In the TRACE User Guide it is stressed that one reversal report can only cancel one original report. This may be the rule, but in some cases as in table 1.4 it could be, that the reversal is meant to cancel more than just one report, since the reversal matches four other reports. It is rather unusual in the data to have more trades in the same second with matching prices 11According to FINRA (2008) page 31 a reversal should be exactly identical to the report it is reversing. 12The second example in table 1.3 is due to a rating change between the date of the first report and the date of the reversal report. Trading volumes are censored for investment grade bonds at $5 millions and for speculative grade at $1 million. Liquidity Biases in TRACE 11 but different quantities. This makes the trade sequence seems suspicious. On the other hand one could argue that a broker might be chopping up a deal into smaller pieces but actually selling it all to just one customer for some reason. If this is the case the price sequence is misleading since the negotiated price is not for separate small trades but for a larger package i.e. a larger volume, in which case we do not have 9 consecutive trades with the same the price but one larger trade. Bond ID Date Time Price Par volume Yield As-Of CSCO.GB 20060217 14:25:00 101.000 10000 5.02100 CSCO.GB 20060217 14:25:00 101.000 10000 5.02100 CSCO.GB 20060217 14:25:00 101.000 20000 5.02100 CSCO.GB 20060217 14:25:00 101.000 20000 5.02100 CSCO.GB 20060217 14:25:00 101.000 10000 5.02100 CSCO.GB 20060217 14:25:00 101.000 20000 5.02100 CSCO.GB 20060217 14:25:00 101.000 10000 5.02100 CSCO.GB 20060217 14:25:00 101.000 10000 5.02100 R CSCO.GB 20060217 16:01:09 101.000 10000 5.02100 Table 1.4: Disseminated Trade Reports with non-identical reversals. This table shows an example where the reversal matches more than just one trade report. According to TRACE guidelines one reversal is only meant to cancel one original report. The table only displays a selection of the disseminated information about each report. These are bond ID, Date, Time, Price, Par Volume, Yield and As-Of indicator. On a more technical note the disseminated TRACE data from WRDS do not include the filing date of the reports, which would be helpful when we want to match the reversal with the original report13. In panel A of table 1.5 the reversal matches the as-of trade report and not what is likely to be the original report. In this case the bond has gone from a speculative grade rating to an investment grade rating in the time between the original report filing and the filing of the reversal report. At the time of the original filing the volume then was censored at $1 million dollars whereas at the time of the reversal report the censoring was at $5 million. Looking at the message sequence numbers (not shown in the table) the reversal and the as-of trade have consecutive numbers with the reversal having the lower number. This indicates that the reversal is filed before the as-of trade report and therefore meant to cancel the third report in panel A14. In panel B the as-of trade matching the reversal was filed with a wrong time record. Then the reversal 13The filing date is actually public information in the way that it is disseminated for example as part of the FISD time sales data. 14There is a possibility of the two reports being filed at different days and still getting consecutive number, but it is a rather small possibility. This question could be resolved with the information about filing date as for example FISD contains. If the reports are 12 Essay 1 is canceling the as-of trade that it matches and the last as-of trade is the one with the correct time record. In this case it is safe to assume that the reversal is referring to the as-of trade. In panel C of table 1.5 it is again not clear which report the reversal is meant to cancel. The reversal does not actually match any of the other reports but comes closest to matching the as-of report. As in panel A the message sequence numbers indicate that the reversal is filed before the as-of trade in which case it could not be canceling the as-of trade. It then seems most likely that the reversal is meant to cancel one (or both) of the other reports. In conclusion, since we are not allowed to see the direct link between the reversal and the original report in the disseminated data, the filling date (which is public information but not provided by WRDS) would be helpful when making the indirect link. Bond ID Date Time Price Par volume Yield As-Of Panel A CSCO.GB 20060215 15:01:40 100.028 5000000 5.24356 R CSCO.GB 20060215 15:01:40 100.028 5000000 5.24356 A CSCO.GB 20060215 15:01:40 100.028 1MM+ 5.24356 Panel B CSCO.GB 20060330 12:54:00 99.370 25000 5.39700 R CSCO.GB 20060330 12:54:00 99.370 25000 5.39700 A CSCO.GB 20060330 13:12:00 99.370 25000 5.39700 A Panel C CSCO.GB 20060828 16:53:00 99.656 15000 5.33700 CSCO.GB 20060828 16:53:00 99.656 15000 5.33700 CSCO.GB 20060828 17:13:57 99.656 15000 5.33702 R CSCO.GB 20060828 17:13:57 99.656 15000 5.33700 A Table 1.5: Disseminated Trade Reports with reversals of as-of trades. This table shows three examples where it is not clear which reports the reversals are meant to cancel. It could look like the reversals are canceling the as-of reports, which is only true in panel B. The table only displays a selection of the disseminated information about each report. These are bond ID, Date, Time, Price, Par Volume, Yield and As-Of indicator. When using the disseminated TRACE data from WRDS we make the following choices when implementing the filter. If there is no identical match for a reversal based on the parameters shown in table 1.3-1.5 we only delete the reversal because the original report cannot be identified with certainty. Second, if there is more than one trade report, that matches the reversal filed on the same day and the reversal has the lower message sequence number then the reversal was filed before the as-of report for sure. In this case the reversal could not be canceling the as-of trade, since the as-of trade was not yet filed. 12 Essay 1 is canceling the as-of trade that it matches and the last as-of trade is the one with the correct time record. In this case it is safe to assume that the reversal is referring to the as-of trade. In panel C of table 1.5 it is again not clear which report the reversal is meant to cancel. The reversal does not actually match any of the other reports but comes closest to matching the as-of report. As in panel A the message sequence numbers indicate that the reversal is filed before the as-of trade in which case it could not be canceling the as-of trade. It then seems most likely that the reversal is meant to cancel one (or both) of the other reports. In conclusion, since we are not allowed to see the direct link between the reversal and the original report in the disseminated data, the filling date (which is public information but not provided by WRDS) would be helpful when making the indirect link. Bond ID Date Time Price Par volume Yield As-Of Panel A CSCO.GB 20060215 15:01:40 100.028 5000000 5.24356 R CSCO.GB 20060215 15:01:40 100.028 5000000 5.24356 A CSCO.GB 20060215 15:01:40 100.028 1MM+ 5.24356 Panel B CSCO.GB 20060330 12:54:00 99.370 25000 5.39700 R CSCO.GB 20060330 12:54:00 99.370 25000 5.39700 A CSCO.GB 20060330 13:12:00 99.370 25000 5.39700 A Panel C CSCO.GB 20060828 16:53:00 99.656 15000 5.33700 CSCO.GB 20060828 16:53:00 99.656 15000 5.33700 CSCO.GB 20060828 17:13:57 99.656 15000 5.33702 R CSCO.GB 20060828 17:13:57 99.656 15000 5.33700 A Table 1.5: Disseminated Trade Reports with reversals of as-of trades. This table shows three examples where it is not clear which reports the reversals are meant to cancel. It could look like the reversals are canceling the as-of reports, which is only true in panel B. The table only displays a selection of the disseminated information about each report. These are bond ID, Date, Time, Price, Par Volume, Yield and As-Of indicator. When using the disseminated TRACE data from WRDS we make the following choices when implementing the filter. If there is no identical match for a reversal based on the parameters shown in table 1.3-1.5 we only delete the reversal because the original report cannot be identified with certainty. Second, if there is more than one trade report, that matches the reversal filed on the same day and the reversal has the lower message sequence number then the reversal was filed before the as-of report for sure. In this case the reversal could not be canceling the as-of trade, since the as-of trade was not yet filed. Liquidity Biases in TRACE 13 we delete all of them not including any as-of trades that might match the reversal. That is, we do not delete any as-of reports that match the reversal on the chosen set of parameters. Then we would not make a mistake in panel A and C of table 1.5 but we do make an error in panel B of the same table. We delete too few reports when we cannot find an original report and when we do not delete as-of trades that are later reversed. On the other hand we delete to many reports when we delete all ordinary reports that match a reversal. 1.3.3 Performance Each year FINRA publishes a TRACE Fact Book with summary statistics of trading over the past year. We test the performance of our error filter by matching the number of trades post filtering for the 10 most frequently traded investment grade bonds in 2007 with the numbers for the same bonds from the official fact book. In table 1.6 we can see that the algorithm performs fairly well. The most traded bond that year was a General Electric bond with an official number of 12,857 trades. In the raw data file the same bond has 13,479 trade reports, but after the filtering the number of trade reports is 12,856, i.e. only 1 report short of the official number. Apparently, there is some discrepancy between the fact book of 2007 and the number of disseminated reports in that the second most traded bond according to the fact book was a Morgan Stanley bond. However for unknown reasons, the raw TRACE file only displays one third of the transactions reported for the bond in the TRACE fact book. All of the other bonds are close to the official number of trades. The largest deviation is 1.5%. Looking at the sign of the deviations we can see that the filter most commonly deletes too few observations. This happens because the dominating problem is that we cannot match a reversal with an original report. For the few bonds where we delete too many observations some reversals have matched more than just one original report. When we compare this with the number of unmatched reversals the latter problem seems most important. If we only deleted one report each time we had an identical match to a reversal and refrained from deleting more than one if there were more matches, we would still get a small error percentage for the filter. The first step of the algorithm deletes 2,532 reports15. These are the reports which match in pairs on all parameters including the unique intraday message sequence number. When two reports filed on the same day have matching message sequence numbers it means that they are refering to the same trade and one of them has to be deleted if we want to avoid double counting. These reports are interdealer trades where both reports are accidentally disseminated for some reason. In the second step of the algorithm we delete reversals and matching original reports. There are a 15Our data sample covers transactions up to and including 2008Q3. 14 Essay 1 Symbol Issuer Name Coupon Maturity Actual Raw Reports Post Filter Rev. Missing Dev. Pct. GE.ADF GE Company 5.000 2/1/13 12,857 13,479 12,856 12 -0.01 MS.QP Morgan Stanley 4.750 4/1/14 12,333 3940 3788 11 - GS.OU Goldman Sachs Group 5.700 9/1/12 11,573 12,061 11,577 16 0.03 C.HEF Citygroup 5.000 9/15/14 11,212 11,693 11,217 19 0.04 GE.AAD GE Capital Corp. 6.000 6/15/12 11,085 11,511 11,086 21 0.01 BLS.HW Bellsouth Corp. 6.000 11/15/34 10,450 11,072 10,594 148 1.38 WMT.HN Wal-Mart Stores 4.550 5/1/13 9,681 10,052 9,678 18 -0.03 GE.WB GE Corp. 5.875 2/15/12 9,468 9,957 9,470 16 0.02 GS.WL Goldman Sachs Group 5.625 1/15/17 8,108 8,655 8,115 15 0.09 JPM.QP J. P. Morgan Chase 5.750 1/2/13 8,051 8,365 8,048 10 -0.04 Table 1.6: Performance of the error filter. This table shows the error filter performance on a selection of bonds for which the actual number of trades are known from the Trace Fact Year Book 2007. The selected bonds are the top 10 most traded investment grade issues in 2007. The differences between the filtered disseminated data and the actual number of trades arise because a perfect filtering is not possible using only the disseminated data. In the table ’actual’ refers to the actual number of trades, ’raw reports’ refers to the number of reports disseminated in Trace, ’post filter’ refers to the number of reports or trades left after applying the filter, ’rev. missing’ refers to the number of reversals where it was not possible to find an identical original report and ’the deviation in percentage’ is between the actual number of trades and the post filtered amount. total of 418,626 reversal reports and we end up deleting 763,013 reports in this second step. As in table 1.6 we are not able to match all of the reversals with an original report. Table 1.7 shows a summary of the filtering process. Note that a total of 120,420 reversals remains unmatched in the filtering. Finally, in the third step we delete same-day cancelations and corrections. We end up with a total of 26,943,152 trade reports having deleted 1,404,978 reports in the third step. The database contains a total of 29,113,675 raw reports of which we have dropped 7.5% in our filtering. The official error rate is 7.7% which is slightly higher than ours since we still lack to match some reversals to their original reports. When introducing a new reporting system such as the TRACE system it is natural for the users to make errors simply because they are not familiar with the system yet. Figure 1.1 shows a time series plot of the monthly reporting error rate. There is a clear downward trend in the error rate over time, but it is still far from zero. So even if we decided only to look at a new data sample from TRACE, we would still need to filter the data before use. 14 Essay 1 Symbol Issuer Name Coupon Maturity Actual Raw Reports Post Filter Rev. Missing Dev. Pct. GE.ADF GE Company 5.000 2/1/13 12,857 13,479 12,856 12 -0.01 MS.QP Morgan Stanley 4.750 4/1/14 12,333 3940 3788 11 - GS.OU Goldman Sachs Group 5.700 9/1/12 11,573 12,061 11,577 16 0.03 C.HEF Citygroup 5.000 9/15/14 11,212 11,693 11,217 19 0.04 GE.AAD GE Capital Corp. 6.000 6/15/12 11,085 11,511 11,086 21 0.01 BLS.HW Bellsouth Corp. 6.000 11/15/34 10,450 11,072 10,594 148 1.38 WMT.HN Wal-Mart Stores 4.550 5/1/13 9,681 10,052 9,678 18 -0.03 GE.WB GE Corp. 5.875 2/15/12 9,468 9,957 9,470 16 0.02 GS.WL Goldman Sachs Group 5.625 1/15/17 8,108 8,655 8,115 15 0.09 JPM.QP J. P. Morgan Chase 5.750 1/2/13 8,051 8,365 8,048 10 -0.04 Table 1.6: Performance of the error filter. This table shows the error filter performance on a selection of bonds for which the actual number of trades are known from the Trace Fact Year Book 2007. The selected bonds are the top 10 most traded investment grade issues in 2007. The differences between the filtered disseminated data and the actual number of trades arise because a perfect filtering is not possible using only the disseminated data. In the table ’actual’ refers to the actual number of trades, ’raw reports’ refers to the number of reports disseminated in Trace, ’post filter’ refers to the number of reports or trades left after applying the filter, ’rev. missing’ refers to the number of reversals where it was not possible to find an identical original report and ’the deviation in percentage’ is between the actual number of trades and the post filtered amount. total of 418,626 reversal reports and we end up deleting 763,013 reports in this second step. As in table 1.6 we are not able to match all of the reversals with an original report. Table 1.7 shows a summary of the filtering process. Note that a total of 120,420 reversals remains unmatched in the filtering. Finally, in the third step we delete same-day cancelations and corrections. We end up with a total of 26,943,152 trade reports having deleted 1,404,978 reports in the third step. The database contains a total of 29,113,675 raw reports of which we have dropped 7.5% in our filtering. The official error rate is 7.7% which is slightly higher than ours since we still lack to match some reversals to their original reports. When introducing a new reporting system such as the TRACE system it is natural for the users to make errors simply because they are not familiar with the system yet. Figure 1.1 shows a time series plot of the monthly reporting error rate. There is a clear downward trend in the error rate over time, but it is still far from zero. So even if we decided only to look at a new data sample from TRACE, we would still need to filter the data before use. Liquidity Biases in TRACE 15 Date Pct Error Reports (Monthly) 5 6 7 8 9 10 2002M07 2004M02 2005M09 2007M04 2008M09 Figure 1.1: Monthly Percentage of Error Reports. This graph shows the monthly percentage of error reports in the disseminated TRACE data. There is a small decline in errors over time. The error rate peaks after the start of the last phase of the dissemination in 2004Q4. 16 Essay 1 Description Trade reports Raw reports 29,113,675 Step 1 Deleted 2,532 Post step 1 29,111,143 Step 2 Deleted 763,013 Reversals 418,626 Unmatched Reversals 120,421 Post step 2 28,348,130 Step 3 Deleted 1,404,978 Post step 3 26,943,152 Table 1.7: Filtering Summary. This table shows how many reports that are deleted in each step of the error filter and for step 2 the table lists the number of unmatched reversals. The unmatched reversals are the main problem for the filter as seen in table 1.6 1.4 Error Impact In this section we first show that ignoring the reporting errors will bias some of the most commonly used liquidity measures. Secondly, we show that replacing the error filter with a standard stock market filter is not a viable alternative. In stock market research there are well tested filters which screen the price sequences for outliers. However, applying a similar approach to the TRACE data will have almost no effect on the biases from section 1.3. 1.4.1 Market Liquidity If the TRACE data are not cleaned up before use, the number of transactions will be too high. For many research applications this will lead to a bias in the results16. Furthermore, the reporting error percentage is increasing in the trade size of the bonds. In table 1.8 the error percentage is listed for different trade sizes. For trades of par value $1,000,000 and above the amount of error reports is 13.2%. Most studies find larger trades more interesting than smaller trades, because these trades are likely to have been carried out by well informed institutional traders with high bargaining power and 90%- 16This section is in part inspired by the SAS-programs that WRDS has on their website. These are very helpful for first time users but they are ignoring the problems with reporting errors. Using their code for research will result in the biases of this section. 16 Essay 1 Description Trade reports Raw reports 29,113,675 Step 1 Deleted 2,532 Post step 1 29,111,143 Step 2 Deleted 763,013 Reversals 418,626 Unmatched Reversals 120,421 Post step 2 28,348,130 Step 3 Deleted 1,404,978 Post step 3 26,943,152 Table 1.7: Filtering Summary. This table shows how many reports that are deleted in each step of the error filter and for step 2 the table lists the number of unmatched reversals. The unmatched reversals are the main problem for the filter as seen in table 1.6 1.4 Error Impact In this section we first show that ignoring the reporting errors will bias some of the most commonly used liquidity measures. Secondly, we show that replacing the error filter with a standard stock market filter is not a viable alternative. In stock market research there are well tested filters which screen the price sequences for outliers. However, applying a similar approach to the TRACE data will have almost no effect on the biases from section 1.3. 1.4.1 Market Liquidity If the TRACE data are not cleaned up before use, the number of transactions will be too high. For many research applications this will lead to a bias in the results16. Furthermore, the reporting error percentage is increasing in the trade size of the bonds. In table 1.8 the error percentage is listed for different trade sizes. For trades of par value $1,000,000 and above the amount of error reports is 13.2%. Most studies find larger trades more interesting than smaller trades, because these trades are likely to have been carried out by well informed institutional traders with high bargaining power and 90%- 16This section is in part inspired by the SAS-programs that WRDS has on their website. These are very helpful for first time users but they are ignoring the problems with reporting errors. Using their code for research will result in the biases of this section. Liquidity Biases in TRACE 17 95% of all trading measured by volume takes place in institutional trade sizes (typically a institutional trade is defined as a trade with par value size above $100,000). Trade Size Error Pct -5,000 5.27 5,000-10,000 4.60 10,000-25,000 4.71 25,000-250,000 6.86 250,000-1,000,000 11.68 1,000,000- 13.24 Table 1.8: Errors as a function of trade size. This table shows the error percentage as a function of the trading size. As the trading size increases so does the number of corrections and cancelations. A typical measure of liquidity is the turnover of an asset. We define the turnover for a bond as the daily average of total trading volume taken over days with at least one trade17. When the error reports are deleted from the raw TRACE data the turnover will be lower. That is, if we do not delete the errors the trading activity on the market will appear higher than it actually is. From table 1.9 we can see that the median turnover deviation is 7.2%. A large part of the bonds have the same turnover measure before and after the filtering simply because these bonds have very little or no reports deleted in the filtering. But at the other end of the scale a quarter of the bonds will have the turnover overestimated by more than 14%. Another popular measure of liquidity is the Amihud price impact measure. We calculate a quarterly measure for each bond by first calculating a daily Amihud measure and then taking the median across days with a non-zero measure. The daily measure is given on the form: Amihudbond,t =1 N N |rij| Qj where N is the number of trades on day t for the bond, rij is the return between consecutive trades jand iand Qiis the dollar par volume for trade i. Note that taking the median rather than the mean across the quarter makes the measure far more robust. But even with this robust definition of the Amihud measure a large part of the bonds have a too low measure pre filtering. In table 1.9 we can see that the median bias is at 0.0% but that for more than a quarter of the quarterly measures the bias is as high 17This definition is taken from one of the example programs for TRACE on the WRDS website. Since the majority of the bonds trade infrequently a more proper definition would be to take the average over all days. 24 Essay 1 is important. The agency trades are part of the official turnover statistics from the TRACE fact book. Deleting them will give a bias when comparing turnover with the official FINRA statistics. With the disseminated data none of the filters can be constructed to eliminate all errors because the information linking the transactions together is not disseminated. Our filter relies on a number of assumptions that are needed in order to indirectly identify the reports with errors. All of the assumptions we use are fairly conservative and since we are able to closely replicate the official trade statistics, the assumptions seem appropriate24. 24SAS programs, with both kinds of filters implemented and easy to use, are available from the author upon request. 24 Essay 1 is important. The agency trades are part of the official turnover statistics from the TRACE fact book. Deleting them will give a bias when comparing turnover with the official FINRA statistics. With the disseminated data none of the filters can be constructed to eliminate all errors because the information linking the transactions together is not disseminated. Our filter relies on a number of assumptions that are needed in order to indirectly identify the reports with errors. All of the assumptions we use are fairly conservative and since we are able to closely replicate the official trade statistics, the assumptions seem appropriate24. 24SAS programs, with both kinds of filters implemented and easy to use, are available from the author upon request. Essay 2 Corporate Bond Liquidity Before and After the Onset of the Subprime Crisis 1 Co-authored with Peter Feldh¨utter and David Lando, Copenhagen Business School. Abstract We analyze liquidity components of corporate bond spreads by combining the superior data quality of transaction-level corporate bond prices from TRACE with the increase in bond spreads caused by the crisis. A single linear combination of four liquidity proxies captures most of the liquidity-related variation of spreads before and during the crisis. The contribution to spreads from illiquidity increases dramatically with the crisis. We use our measure to shed new light on flight-to- quality, liquidity risk, the impact of trading frequency, the role of funding shocks to lead underwriters, and the liquidity of corporate bonds issued by financial firms. 1We thank Yakov Amihud, Sreedhar Bharath, Michael Brennan, Tom Engsted, Edie Hotchkiss, Marco Pagano, Lasse Pedersen, Ilya Strebulaev and seminar participants at seminars at the Goethe University in Frankfurt, Deutsche Bundesbank, ECB, Oesterreichische Nationalbank, CBS, NYU, and at conferences in Bergen (EFA), Konstanz, Florence, London and Venice for helpful comments. 25 26 Essay 2 2.1 Introduction The onset of the subprime crisis caused a dramatic widening of corporate bond spreads. In light of the strong evidence that illiquidity in addition to credit risk contributes to corporate bond spreads, it is reasonable to believe that at least part of the spread widening can be attributed to a decrease in bond liquidity, and perhaps to an increase in liquidity risk as well. To show this we need robust measures of liquidity and liquidity risk which enable us to disentangle the credit risk component and the liquidity component of corporate bond spreads. Ideally, a robust measure should be significant before and after the crisis, and we would expect it to reveal a strong decrease in liquidity around the onset of the crisis. We show in this paper that a sum of four liquidity proxies has been a consistent contributor to corporate bond spreads both before and after the onset of the crisis and across rating categories. The four variables are Amihud’s measure of price impact, a measure of roundtrip cost of trading, and the variability of each of these two measures. We can think of the Amihud measure and the roundtrip cost measure as measuring liquidity, and the two variability measures as representing liquidity risk. We arrive at our liquidity measure through a principal component analysis which reveals that the first principal component among eight liquidity variables is almost the same before and after the onset of the crisis and it is close to being an equally weighted sum of the four variables mentioned above. When we regress corporate bond spreads on the principal components, and control for credit risk, only the first component contributes to corporate bond spreads consistently across ratings and regime. In this sense our liquidity measure dominates trading frequency of bonds used in Chen, Lesmond, and Wei (2007) and Roll’s bid-ask measure used by Bao, Pan, and Wang (2009). This consistency is important for drawing conclusions when we split the sample by industry and lead underwriter as explained below. We use our liquidity measure to identify the contribution of liquidity to corporate bond spreads before and after the onset of the crisis, across different rating categories and across maturity. The procedure we use is to first compute our liquidity measure for each bond in the sample. Within a rating category, we then order the bonds according to this measure. Higher values correspond to lower liquidity. We then compute the difference between the 5% and the 50% quantiles and multiply by the regression coefficient for that rating category. The result is the liquidity-related difference between the spread of bonds with median and with high liquidity. How large is then the effect of liquidity on spreads? Before the crisis, it was small for investment grade bonds both as a fraction of the yield spread and measured in basis points. The contribution to spreads from lack of liquidity rose through both an increase in our liquidity measure and in the sensitivity to this measure across all rating categories at the onset of 26 Essay 2 2.1 Introduction The onset of the subprime crisis caused a dramatic widening of corporate bond spreads. In light of the strong evidence that illiquidity in addition to credit risk contributes to corporate bond spreads, it is reasonable to believe that at least part of the spread widening can be attributed to a decrease in bond liquidity, and perhaps to an increase in liquidity risk as well. To show this we need robust measures of liquidity and liquidity risk which enable us to disentangle the credit risk component and the liquidity component of corporate bond spreads. Ideally, a robust measure should be significant before and after the crisis, and we would expect it to reveal a strong decrease in liquidity around the onset of the crisis. We show in this paper that a sum of four liquidity proxies has been a consistent contributor to corporate bond spreads both before and after the onset of the crisis and across rating categories. The four variables are Amihud’s measure of price impact, a measure of roundtrip cost of trading, and the variability of each of these two measures. We can think of the Amihud measure and the roundtrip cost measure as measuring liquidity, and the two variability measures as representing liquidity risk. We arrive at our liquidity measure through a principal component analysis which reveals that the first principal component among eight liquidity variables is almost the same before and after the onset of the crisis and it is close to being an equally weighted sum of the four variables mentioned above. When we regress corporate bond spreads on the principal components, and control for credit risk, only the first component contributes to corporate bond spreads consistently across ratings and regime. In this sense our liquidity measure dominates trading frequency of bonds used in Chen, Lesmond, and Wei (2007) and Roll’s bid-ask measure used by Bao, Pan, and Wang (2009). This consistency is important for drawing conclusions when we split the sample by industry and lead underwriter as explained below. We use our liquidity measure to identify the contribution of liquidity to corporate bond spreads before and after the onset of the crisis, across different rating categories and across maturity. The procedure we use is to first compute our liquidity measure for each bond in the sample. Within a rating category, we then order the bonds according to this measure. Higher values correspond to lower liquidity. We then compute the difference between the 5% and the 50% quantiles and multiply by the regression coefficient for that rating category. The result is the liquidity-related difference between the spread of bonds with median and with high liquidity. How large is then the effect of liquidity on spreads? Before the crisis, it was small for investment grade bonds both as a fraction of the yield spread and measured in basis points. The contribution to spreads from lack of liquidity rose through both an increase in our liquidity measure and in the sensitivity to this measure across all rating categories at the onset of Corporate Bond Liquidity Before and After the Onset of the Subprime Crisis 27 the crisis, although the AAA contribution remains small during the crisis. Our finding that liquidity components in AAA-rated bonds are small even after the onset of the crisis is consistent with a flight-to-quality into those bonds. Measured as a fraction of spreads, there was almost no change in the liquidity component for speculative grade bonds. When we zoom in on the time series behavior of liquidity premia, we find that they persistently increase for investment grade bonds during the crisis and peak around the rapid stock market decline in the first quarter of 2009. For speculative grade bonds, premia are less persistent, peak around the Lehman default in the fall of 2008, and returned almost to pre-crisis levels in the summer of 2009. Our measure is also useful for analyzing other aspects of corporate bond illiquidity. We construct a liquidity beta, i.e. a measure for the covariation of an individual bond’s liquidity with that of the entire corporate bond market. We show that this liquidity beta is not a significant contributor to spreads before the onset of the crisis but it does contribute to spreads for bonds except for AAA-rated bonds after the onset of the crisis. This indicates that the flight-to-quality effect in investment grade bonds found in Acharya, Amihud, and Bharath (2010) is confined to AAA-rated bonds. We also ask whether financial distress of a lead underwriter of a corporate bond issue affects the liquidity of the bond in the secondary market. If lead underwriters are providers of liquidity of the bond in secondary market trading, it is conceivable that if a lead underwriter is in financial distress, the liquidity of the bond decreases relative to other bonds. We show that bonds which had Bear Stearns as lead underwriter had lower liquidity during the takeover of Bear Stearns and bonds with Lehman as lead underwriter had lower liquidity around the bankruptcy of Lehman. Finally, we investigate whether the time series variation of liquidity of corporate bonds issued by financial firms is different from the variation for bonds issued by industrial firms. There is conflicting empirical evidence on this issue: Longstaff, Mithal, and Neis (2005) find that bonds issued by financial firms are more illiquid than other bonds, while Friewald, Jankowitsch, and Subrahmanyam (2009) find this not to be the case. Our time series study reveals that bonds issued by financial firms have similar liquidity as bonds issued by industrial firms, except in extreme stress periods, where bonds of financial firms become very illiquid, overall and compared to bonds issued by industrial firms. The detailed trading data for corporate bonds available from the TRACE database are critical for our ability to measure liquidity proxies properly, and they help us shed new light on previous results on liquidity in corporate bonds. We show that Datastream’s record of zero return days for a bond, which in Chen, Lesmond, and Wei (2007) is used to proxy for days when the bond does not trade, has little connection to the actual trades recorded in TRACE. With actual trades, the LOT measure employed in Chen, Lesmond, and Wei (2007) becomes unrealistically large. We also show that the Amihud measure is strongly influenced by restricting the universe of trades to large 28 Essay 2 trades, as we do in this paper. Using large trades only, the median price impact of a 300.000 dollar trade is roughly 0.1%, whereas Han and Zhou (2008) using all trades obtain an impact of 10.2%. To support the claim that our measure is not measuring credit risk, we run regressions on a matched sample of corporate bonds using pairs of bonds issued by the same firm with maturity close to each other. Instead of credit controls, we use a dummy variable for each matched pair and estimate the response of spreads to our liquidity measure. The measure remains significant. In an appendix, we also show that our regression results change only slightly if we choose Treasury instead of swap rates as our riskless rates, and we test for simultaneous equation bias arising from joint determination of credit and liquidity risk and for omitted variables. The flow of our paper is as follows. We describe our data set and how we define the eight liquidity variables that enter into the regressions. After providing summary statistics of our liquidity proxies, we run regressions on the eight liquidity variables one at a time while controlling for credit risk. We see that four variables stand out as significant predictors of spreads. Remarkably, these four variables also form the first component in a principal component decomposition of the standardized liquidity variables - and this decomposition is stable before and after the onset of the crisis. We then perform the same regressions as above - using one principal component at a time instead of the liquidity proxies. The first principal component is the only consistently significant regressor variable. Since the principal component is close to being an equally weighted sum of the four liquidity variables, we define an operational measure of liquidity as the sum of the four variables. This measure is then used to measure the contribution of illiquidity to corporate bond spreads across ratings and maturities, before and after the onset of the crisis. Furthermore, we use our measure to examine how the covariance between bond-specific liquidity and market-wide liquidity affects bond spreads, and how financial distress of a lead underwriter and the type of firm issuing the bond affect bond liquidity. 2.2 Literature review It has been recognized for a long time that the ease with which a security is traded influences its price. A comprehensive survey of both the different notions of and the empirical evidence on liquidity can be found in Amihud, Mendelson, and Pedersen (2005). Here, we will focus on the growing literature that deals specifically with corporate bonds. In recent years, the illiquidity of corporate bonds has been seen as a possible explanation for the ’credit risk puzzle’, i.e. the claim that yield spreads on corporate bonds are larger than what can be explained by default risk - even after adjusting for recovery risk and compensation for bearing default risk. Huang and Huang (2003) calibrate structural default risk models 28 Essay 2 trades, as we do in this paper. Using large trades only, the median price impact of a 300.000 dollar trade is roughly 0.1%, whereas Han and Zhou (2008) using all trades obtain an impact of 10.2%. To support the claim that our measure is not measuring credit risk, we run regressions on a matched sample of corporate bonds using pairs of bonds issued by the same firm with maturity close to each other. Instead of credit controls, we use a dummy variable for each matched pair and estimate the response of spreads to our liquidity measure. The measure remains significant. In an appendix, we also show that our regression results change only slightly if we choose Treasury instead of swap rates as our riskless rates, and we test for simultaneous equation bias arising from joint determination of credit and liquidity risk and for omitted variables. The flow of our paper is as follows. We describe our data set and how we define the eight liquidity variables that enter into the regressions. After providing summary statistics of our liquidity proxies, we run regressions on the eight liquidity variables one at a time while controlling for credit risk. We see that four variables stand out as significant predictors of spreads. Remarkably, these four variables also form the first component in a principal component decomposition of the standardized liquidity variables - and this decomposition is stable before and after the onset of the crisis. We then perform the same regressions as above - using one principal component at a time instead of the liquidity proxies. The first principal component is the only consistently significant regressor variable. Since the principal component is close to being an equally weighted sum of the four liquidity variables, we define an operational measure of liquidity as the sum of the four variables. This measure is then used to measure the contribution of illiquidity to corporate bond spreads across ratings and maturities, before and after the onset of the crisis. Furthermore, we use our measure to examine how the covariance between bond-specific liquidity and market-wide liquidity affects bond spreads, and how financial distress of a lead underwriter and the type of firm issuing the bond affect bond liquidity. 2.2 Literature review It has been recognized for a long time that the ease with which a security is traded influences its price. A comprehensive survey of both the different notions of and the empirical evidence on liquidity can be found in Amihud, Mendelson, and Pedersen (2005). Here, we will focus on the growing literature that deals specifically with corporate bonds. In recent years, the illiquidity of corporate bonds has been seen as a possible explanation for the ’credit risk puzzle’, i.e. the claim that yield spreads on corporate bonds are larger than what can be explained by default risk - even after adjusting for recovery risk and compensation for bearing default risk. Huang and Huang (2003) calibrate structural default risk models Corporate Bond Liquidity Before and After the Onset of the Subprime Crisis 29 to match the default probabilities and recoveries of corporate bonds. They use a specification of the risk premium - learned from equity markets - to price the default risk in corporate bonds and show that the resulting credit spreads are smaller than the observed spreads. Other works supporting the idea that there are components of credit spreads that are unrelated to default risk include Elton, Gruber, Agrawal, and Mann (2001) who show that yield spreads cannot entirely be explained by credit risk and tax effects, and Collin-Dufresne, Goldstein, and Martin (2001) who show that changes in credit spreads cannot be explained by credit risk alone. Covitz and Downing (2007) study credit spread components in the short maturity commercial paper market and while they do find evidence of a contribution to spreads from illiquidity, they find credit risk to be the main determinant of spreads for short maturities. Longstaff, Mithal, and Neis (2005) subtract CDS premia from bond spreads to extract a non-default component of a corporate bond spread. They show that this component is correlated with proxies for liquidity both in the cross section of corporate bond spreads and in the time series evolution of spreads. In our paper, we cover a larger segment of the corporate bond market than those for which CDS premia exist. Also, it is frequently the case that the CDS spread is larger than the comparable bond spread indicating that the CDS market may also be prone to buying and selling pressures. This suggests that there are also liquidity components in CDS spreads as confirmed by Bongaerts, Driessen, and de Jong (2009). Earlier papers which show that liquidity proxies are significant explanatory variables for corporate bond spreads and bond returns are Houweling, Mentink, and Vorst (2005)), Downing, Underwood, and Xing (2005), and de Jong and Driessen (2006). An early contribution, which also stresses the importance of matrix pricing for empirical studies of bond liquidity, is Sarig and Warga (1989). TRACE transactions data became available only recently, and therefore few studies have used the data set. Bao, Pan, and Wang (2009) use TRACE data to study liquidity effects focusing in particular on a transformation of the Roll measure. There are several studies on the effects of the introduction of TRACE. These show that the enhanced price transparency following the dissemination of prices has lowered transaction costs for investors, see Edwards, Harris, and Piwowar (2007), Goldstein, Hotchkiss, and Sirri (2007), and Bessembinder, Maxwell, and Venkaraman (2006). This would suggest that liquidity has increased. However, as shown in Goldstein, Hotchkiss, and Sirri (2007) trading volume and trading frequency have not increased as a consequence of bond price dissemination, and it is still the case that a large number of bonds trade very infrequently. This is also confirmed by Mahanti, Nashikkar, Subramanyam, Chacko, and Mallik (2008). They combine data on holdings on corporate bonds by different investors with turnover measures of these investor’s portfolios to infer a turnover measure for bonds, called latent liquidity. This measure is shown to have predictive power for 30 Essay 2 other measures of liquidity. However, since we are interested in yield spread effects of illiquidity, we must confine ourselves to the more liquid segment of the corporate bond market for which we can actually observe some trading and therefore some prices and price changes. Friewald, Jankowitsch, and Subrahmanyam (2009) and Han and Zhou (2008) are other papers using the TRACE data set. 2.3 Data description Since January 2001 members of the Financial Industry Regulatory Authority (FINRA) have been required to report their secondary over-the-counter corporate bond transactions through TRACE (Trade Reporting and Compliance Engine). Because of the uncertain benefit to investors of price transparency not all trades reported to TRACE were initially disseminated at the launch of TRACE July 1, 2002. Beginning October 1, 2004 trades in almost all bonds except some lightly traded bonds are disseminated (see Goldstein and Hotchkiss (2008) for details). Therefore our sample starts on this date. We use a sample of straight coupon bullet bonds with trade reports from October 1, 2004 to June 30, 2009. That is, we require that bonds are fixed rate bullet bonds that are not callable, convertible, putable, or have sinking fund provisions. We obtain bond information from Bloomberg, and this provides us initially with 10.785 bond issues. We use rating from Datastream and bonds with missing rating are excluded.2This reduces the sample to 5.376 bonds. For these bonds we collect the trading history from TRACE covering the period from October 1, 2004 to June 30, 2009 and after filtering out erroneous trades, as described in Dick-Nielsen (2009), we are left with 8.212.990 trades. Finally we collect analysts’ forecast dispersion from IBES, share prices for the issuing firms and firm accounting figures from Bloomberg, swap rates from Datastream, Treasury yields consisting of the most recently auctioned issues adjusted to constant maturities published by the Federal Reserve in the H-15 release3and LIBOR rates from British Bankers’ Association. If forecast dispersion, share prices, or firm accounting figures are not available, we drop the corresponding observations from the sample. 2We use the rating from S&P. If this rating is missing we use the rating from Moody’s and if this is missing the rating from Fitch. If we still do not have a rating we use the company rating. 3Further information about the Treasury yield curve methodology can be found on the United States Department of Treasury’s web page http://www.treas.gov/offices/domesticfinance/debt-management/interest-rate/yieldmethod.html. 30 Essay 2 other measures of liquidity. However, since we are interested in yield spread effects of illiquidity, we must confine ourselves to the more liquid segment of the corporate bond market for which we can actually observe some trading and therefore some prices and price changes. Friewald, Jankowitsch, and Subrahmanyam (2009) and Han and Zhou (2008) are other papers using the TRACE data set. 2.3 Data description Since January 2001 members of the Financial Industry Regulatory Authority (FINRA) have been required to report their secondary over-the-counter corporate bond transactions through TRACE (Trade Reporting and Compliance Engine). Because of the uncertain benefit to investors of price transparency not all trades reported to TRACE were initially disseminated at the launch of TRACE July 1, 2002. Beginning October 1, 2004 trades in almost all bonds except some lightly traded bonds are disseminated (see Goldstein and Hotchkiss (2008) for details). Therefore our sample starts on this date. We use a sample of straight coupon bullet bonds with trade reports from October 1, 2004 to June 30, 2009. That is, we require that bonds are fixed rate bullet bonds that are not callable, convertible, putable, or have sinking fund provisions. We obtain bond information from Bloomberg, and this provides us initially with 10.785 bond issues. We use rating from Datastream and bonds with missing rating are excluded.2This reduces the sample to 5.376 bonds. For these bonds we collect the trading history from TRACE covering the period from October 1, 2004 to June 30, 2009 and after filtering out erroneous trades, as described in Dick-Nielsen (2009), we are left with 8.212.990 trades. Finally we collect analysts’ forecast dispersion from IBES, share prices for the issuing firms and firm accounting figures from Bloomberg, swap rates from Datastream, Treasury yields consisting of the most recently auctioned issues adjusted to constant maturities published by the Federal Reserve in the H-15 release3and LIBOR rates from British Bankers’ Association. If forecast dispersion, share prices, or firm accounting figures are not available, we drop the corresponding observations from the sample. 2We use the rating from S&P. If this rating is missing we use the rating from Moody’s and if this is missing the rating from Fitch. If we still do not have a rating we use the company rating. 3Further information about the Treasury yield curve methodology can be found on the United States Department of Treasury’s web page http://www.treas.gov/offices/domesticfinance/debt-management/interest-rate/yieldmethod.html. Corporate Bond Liquidity Before and After the Onset of the Subprime Crisis 31 2.4 Empirical methodology This section provides details on the regression analysis conducted in the next section and defines the set of liquidity variables we use. 2.4.1 Regression As dependent variable we use the yield spread for every bond at the end of each quarter in the regressions. We calculate the quarter-end yield as the average yield for all trades on the last day in the quarter where the bond traded. If a bond did not trade during the last month of the quarter, it is excluded from that quarter. Retail-sized trades (trade below $100,000 in volume) are discarded. Yield spreads are calculated as the difference between the quarter-end yield and the interpolated maturity-matched swap rate calculated on the same day as the yield is measured. We exclude yield spreads for bonds that have less than one month to maturity or have a time to maturity when issued of more than 30 years. To control for credit risk, we follow Blume, Lim, and MacKinlay (1998) and others and add the ratio of operating income to sales, ratio of long term debt to assets, leverage ratio, equity volatility and four pretax interest coverage dummies to the regressions.4In order to capture effects of the general economic environment on the credit risk of firms we include the level and slope of the swap curve, defined as the 10-year swap rate and the difference between the 10-year and 1-year swap rate. Duffie and Lando (2001) show that credit spreads may increase when there is incomplete information on the firm’s true credit quality. To proxy for this effect, we follow G¨untay and Hackbarth (2006) and use dispersion in earnings forecasts as a measure of incomplete information. Finally we add bond age, time-to-maturity, and size of coupon to the regressions - see for example Sarig and Warga (1989), Houweling, Mentink, and Vorst (2005) and Longstaff, Mithal, and Neis (2005). For each rating class we run separate regressions using quarterly obser- 4The pretax interest coverage dummies are defined as follows. We define the pretax interest rate coverage (IRC) ratio as EBIT divided by interest expenses. It expresses how easily the company can cover it’s interest rate expenses. However, the distribution is highly skewed. As in Blume, Lim, and MacKinlay (1998) we control for this skewness by creating four dummies (pretax dummies) which allows for a non-linear relationship with the spread. The first dummy is set to the IRC ratio if it is less than 5 and 5 if it is above. The second dummy is set to 0 if IRC is below 5, to the IRC ratio minus 5 if it lies between 5 and 10 and 5 if it lies above. The third dummy is set to 0 if IRC is below 10, to the IRC ratio minus 10 if it lies between 10 and 20 and 10 if it lies above. The fourth dummy is set to 0 if IRC is below 20 and is set to IRC minus 20 if it lies above 20 (truncating the dummy value at 80). 32 Essay 2 vations. The regressions are Spreadit =α+γLiquidityit +β1Bond Ageit +β2Amount Issuedit +β3Couponit +β4Time-to-Maturityit +β5Eq.Volit +β6Operatingit +β7Leverageit +β8Long Debtit +β9,pretax Pretax dummiesit +β10 10y Swapt +β11 10y-2y Swapt+β12 forecast dispersionit +it (2.1) where iis bond issue, tis quarter, and Liquidityit contains one of the liquidity proxies defined below. Since we have panel data set of yield spreads with each issuer potentially having more than one bond outstanding at any point in time we calculate two-dimensional cluster robust standard errors (see Petersen (2009)). This corrects for time series effects, firm fixed effects and heteroscedasticity in the residuals. 2.4.2 Liquidity Measures Since there is no single measure that adequately describes the liquidity of an asset, we define several liquidity-related measures for corporate bonds in this section. We winsorize the 0.5% highest values of every liquidity variable, meaning that all values above the 99.5% percentile are set to the 99.5% percentile. Amihud measure (price impact of trades) Amihud (2002) constructs an illiquidity measure that is based on the theoretical model of Kyle (1985). It measures the price impact of a trade per unit traded and we use a slightly modified version of this measure. For each corporate bond the measure is the daily average of absolute returns rjdivided by trading volume Qj(in million $) of consecutive transactions: Amihudt=1 Nt Nt  j=1 |rj| Qj =1 Nt Nt  j=1 |Pj−Pj−1 Pj−1| Qj where Ntis the number of returns on day t. At least two transactions are required on a given day in order to calculate the measure, and we define a quarterly Amihud measure by taking the median of daily measures within the quarter. Roll measure (bid-ask spread) A liquid asset can be bought or sold close to the fundamental price of the asset, implying that roundtrip costs are small. A proxy for roundtrip costs 32 Essay 2 vations. The regressions are Spreadit =α+γLiquidityit +β1Bond Ageit +β2Amount Issuedit +β3Couponit +β4Time-to-Maturityit +β5Eq.Volit +β6Operatingit +β7Leverageit +β8Long Debtit +β9,pretax Pretax dummiesit +β10 10y Swapt +β11 10y-2y Swapt+β12 forecast dispersionit +it (2.1) where iis bond issue, tis quarter, and Liquidityit contains one of the liquidity proxies defined below. Since we have panel data set of yield spreads with each issuer potentially having more than one bond outstanding at any point in time we calculate two-dimensional cluster robust standard errors (see Petersen (2009)). This corrects for time series effects, firm fixed effects and heteroscedasticity in the residuals. 2.4.2 Liquidity Measures Since there is no single measure that adequately describes the liquidity of an asset, we define several liquidity-related measures for corporate bonds in this section. We winsorize the 0.5% highest values of every liquidity variable, meaning that all values above the 99.5% percentile are set to the 99.5% percentile. Amihud measure (price impact of trades) Amihud (2002) constructs an illiquidity measure that is based on the theoretical model of Kyle (1985). It measures the price impact of a trade per unit traded and we use a slightly modified version of this measure. For each corporate bond the measure is the daily average of absolute returns rjdivided by trading volume Qj(in million $) of consecutive transactions: Amihudt=1 Nt Nt  j=1 |rj| Qj =1 Nt Nt  j=1 |Pj−Pj−1 Pj−1| Qj where Ntis the number of returns on day t. At least two transactions are required on a given day in order to calculate the measure, and we define a quarterly Amihud measure by taking the median of daily measures within the quarter. Roll measure (bid-ask spread) A liquid asset can be bought or sold close to the fundamental price of the asset, implying that roundtrip costs are small. A proxy for roundtrip costs Corporate Bond Liquidity Before and After the Onset of the Subprime Crisis 33 is the bid-ask spread, but bid-ask spreads are not available in TRACE. Since November 2008, buy-sell indicators are available, but this covers only a fraction of our sample. Roll (1984) finds that under certain assumptions the effective bid-ask spread equals two times the square root of minus the covariance between adjacent price changes: Rollt= 2−cov(∆Pi,∆Pi−1) where tis the time period for which the measure is calculated. The intuition is that the bond price bounces back and forth within the bid-ask band, and higher bid-ask bands lead to higher negative covariance between adjacent price changes. We define a daily Roll measure using a rolling window of 21 trading days, and the measure is only well-defined if there are at least four transactions in the window. We define a quarterly Roll measure by taking the median of daily measures within the quarter. Unique roundtrip cost (bid-ask spread) An alternative measure of transaction costs, proposed in Feldh¨utter (2009), is calculated using unique roundtrip trades (URT). Often, we see a corporate bond trading two or three times within a very short period of time after a longer period with no trades. This is likely to occur because a dealer matches a buyer and a seller and collects the bid-ask spread as a fee. When a dealer has found a match, a trade between seller and dealer along with a trade between buyer and dealer are carried out. Possibly, the matching occurs through a second dealer in which case there is also a transaction between the two dealers. If two or three trades in a given bond with the same volume take place on the same day, and there are no other trades with the same volume on that day, we define the transactions as part of a URT. For a URT we define the unique roundtrip cost (URC) as Pmax −Pmin Pmax where Pmax is the largest price in the URT and Pmin is the smallest price in the URT. A daily estimate of roundtrip costs is the average of roundtrip costs on this day for different volumes, and we estimate quarterly roundtrip costs by averaging over daily estimates. URC overcomes the problem that we only have information on trading volume and not, as in Green, Hollifield, and Sch¨urhoff (2007b), on bid and ask prices or dealer identity. Feldh¨utter (2009) examines the properties of URTs in detail, including how much of total trading volume is captured and for a subsample of TRACE data with buy-sell indicators available, to what extent URTs capture full roundtrip costs. 40 Essay 2 is approximately a factor of 4 for bid-ask spreads and a factor of 8 for lack of market depth. That is, not only have bid-ask spreads widened strongly during the crisis but the ability to sell large notional amounts of bonds without a sizeable discount has disappeared. We see that liquidity in the second quarter of 2009 slowly returns to the market since Amihud and URC are finally decreasing after the increase in previous years. Volume has traditionally been regarded as a proxy for liquidity, since it should be easier to trade when markets are more active. However, Johnson (2008) finds in a simple frictionless model that volume is unrelated to the level of liquidity but related to liquidity risk as measured by the variance of liquidity. Table 2.2 shows that 9 out of 10 regression coefficients for volume are negative indicating that large volumes tend to reduce credit spreads. The significance of the coefficients is modest though, so the evidence is not conclusive. Liquidity risk is clearly priced since Amihud and URC risk have significantly positive regression coefficents in 19 out of 20 cases. Interestingly, all coefficients increase strongly in size post-subprime. Thus, investors require a larger compensation post-subprime for investing in bonds with a high uncertainty about the liquidity discount when selling the bond. Since liquidity risk has increased strongly as Figure 2.1 shows, the impact of liquidity risk is twofold; through a larger level of liquidity risk and through a higher risk premium on liquidity risk. Turning to zero trading days Table 2.2 shows surprisingly that there is no consistent relationship between the number of zero trading days and spreads. If anything, the relationship tends to be negative since 14 out of 20 bond and firm zero regression coefficients are negative. Constantinides (1986) finds theoretically that in the presence of transaction costs, investors will trade infrequently, and consistent with this line of reasoning Chen, Lesmond, and Wei (2007) find that corporate bond spreads - when controlling for credit risk - depend positively on the number of zero trading days. The difference between our results and those of Chen, Lesmond, and Wei (2007) is likely to be the data source. While we use actual transaction data and can directly detect when a trade occurs, Chen, Lesmond, and Wei (2007) use data from Datastream and define a zero trading day as a day where the price does not change. We find that Datastream corporate bond data can differ substantially from actual transaction data in non-predictable ways. To illustrate this, we calculate for each bond quarter the percentage zero trading days using Datastream, and Figure 2.2 plots all pairs of TRACE and Datastream percentage zero trading days. The figure shows that there is very little relation between actual and Datastream zero trading days, and while Datastream often understates the number of zero trading days, they are also overstated for some observations. Although zero-trading days are not correctly identified in Datastream, the LOT measure of Chen, Lesmond, and Wei (2007) could be a relevant measure to include in our analysis. Therefore we have calculated a yearly LOT measure as in Chen, Lesmond, and Wei 40 Essay 2 is approximately a factor of 4 for bid-ask spreads and a factor of 8 for lack of market depth. That is, not only have bid-ask spreads widened strongly during the crisis but the ability to sell large notional amounts of bonds without a sizeable discount has disappeared. We see that liquidity in the second quarter of 2009 slowly returns to the market since Amihud and URC are finally decreasing after the increase in previous years. Volume has traditionally been regarded as a proxy for liquidity, since it should be easier to trade when markets are more active. However, Johnson (2008) finds in a simple frictionless model that volume is unrelated to the level of liquidity but related to liquidity risk as measured by the variance of liquidity. Table 2.2 shows that 9 out of 10 regression coefficients for volume are negative indicating that large volumes tend to reduce credit spreads. The significance of the coefficients is modest though, so the evidence is not conclusive. Liquidity risk is clearly priced since Amihud and URC risk have significantly positive regression coefficents in 19 out of 20 cases. Interestingly, all coefficients increase strongly in size post-subprime. Thus, investors require a larger compensation post-subprime for investing in bonds with a high uncertainty about the liquidity discount when selling the bond. Since liquidity risk has increased strongly as Figure 2.1 shows, the impact of liquidity risk is twofold; through a larger level of liquidity risk and through a higher risk premium on liquidity risk. Turning to zero trading days Table 2.2 shows surprisingly that there is no consistent relationship between the number of zero trading days and spreads. If anything, the relationship tends to be negative since 14 out of 20 bond and firm zero regression coefficients are negative. Constantinides (1986) finds theoretically that in the presence of transaction costs, investors will trade infrequently, and consistent with this line of reasoning Chen, Lesmond, and Wei (2007) find that corporate bond spreads - when controlling for credit risk - depend positively on the number of zero trading days. The difference between our results and those of Chen, Lesmond, and Wei (2007) is likely to be the data source. While we use actual transaction data and can directly detect when a trade occurs, Chen, Lesmond, and Wei (2007) use data from Datastream and define a zero trading day as a day where the price does not change. We find that Datastream corporate bond data can differ substantially from actual transaction data in non-predictable ways. To illustrate this, we calculate for each bond quarter the percentage zero trading days using Datastream, and Figure 2.2 plots all pairs of TRACE and Datastream percentage zero trading days. The figure shows that there is very little relation between actual and Datastream zero trading days, and while Datastream often understates the number of zero trading days, they are also overstated for some observations. Although zero-trading days are not correctly identified in Datastream, the LOT measure of Chen, Lesmond, and Wei (2007) could be a relevant measure to include in our analysis. Therefore we have calculated a yearly LOT measure as in Chen, Lesmond, and Wei Corporate Bond Liquidity Before and After the Onset of the Subprime Crisis 41 0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 Actual percentage zero−trading days Percentage zero−trading days using Datastream Figure 2: Zero-trading days using Datastream. This graph plots for every bond in the sample and every quarter from 2005:Q1 to 2007:Q4 the percentage zero-trading days using Datastream on the x-axis and actual percentage zero-trading days (based on all trades in TRACE) on the y-axis. The thickness of a point depends on the number of observations in that point. The total number of observations is 60,680. 60 Figure 2.2: Zero-trading days using Datastream. This graph plots for every bond in the sample and every quarter from 2005:Q1 to 2007:Q4 the percentage zero-trading days using Datastream on the x-axis and actual percentage zero-trading days (based on all trades in TRACE) on the y-axis. The thickness of a point depends on the number of observations in that point. The total number of observations is 60,680. Corporate Bond Liquidity Before and After the Onset of the Subprime Crisis 41 0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 Actual percentage zero−trading days Percentage zero−trading days using Datastream Figure 2: Zero-trading days using Datastream. This graph plots for every bond in the sample and every quarter from 2005:Q1 to 2007:Q4 the percentage zero-trading days using Datastream on the x-axis and actual percentage zero-trading days (based on all trades in TRACE) on the y-axis. The thickness of a point depends on the number of observations in that point. The total number of observations is 60,680. 60 Figure 2.2: Zero-trading days using Datastream. This graph plots for every bond in the sample and every quarter from 2005:Q1 to 2007:Q4 the percentage zero-trading days using Datastream on the x-axis and actual percentage zero-trading days (based on all trades in TRACE) on the y-axis. The thickness of a point depends on the number of observations in that point. The total number of observations is 60,680. 42 Essay 2 (2007) for all TRACE bonds for the years 2005, 2006, and 2007 based on all TRACE trades. The median roundtrip cost is 237 basis points, which appears too high compared to findings in Edwards, Harris, and Piwowar (2007), Goldstein, Hotchkiss, and Sirri (2007), and Bessembinder, Maxwell, and Venkaraman (2006). From a theoretical point of view the mixed results regarding the impact of zero trading days on spreads can be explained by results in Huberman and Stanzl (2005). They show that an investor trades more often when price impact of trades is high, because he attempts to reduce the total price impact by submitting more but smaller orders. All else equal more trades therefore occur in illiquid bonds since it is necessary to split a sell order in many small trades, while it can be executed in a single trade in a liquid bond.7If this explanation holds true we should expect to see less zero trading days in illiquid times without an increase in the total trading volume. As Figure 2.1 shows this happens during the subprime crisis. The top-right graph shows that the median number of percentage zeros in the regression sample decreases during the subprime crisis. For example, the median number of percentage zeros is 30% in the last quarter of 2008 while it is 62% in the first quarter of 2007. Also, we see in the bottom-left graph that volume in our regression sample decreases slightly during the crisis. Drawing conclusions from Figure 2.1 might be misleading since a bond in a given quarter is only included in the regression sample if it has a full set of accounting variables and trades at least four times that quarter (otherwise the Roll measure cannot be calculated). Thus, it is only the most liquid bonds that are included and there are less bonds included post-subprime than pre-subprime. The decrease in zero trading days might therefore be due to a smaller number of bonds included in the sample. To address this concern, Figure 2.3 shows time series of the quarterly average number of trades and average trade size for all straight coupon bullet bond transactions in our sample period. The top graphs are based on transactions of size $100,000 or more, which our regression results are based on, while the bottom graphs are based on all transactions. In both cases we clearly see an increase on the average number of trades and a decrease in the average trade size after the onset of the subprime crises. Overall, there is theoretical evidence both in favor of and against the 7Goldstein, Hotchkiss, and Sirri (2007) find that dealers behave differently when trading liquid and illiquid bonds. When trading liquid bonds they are more likely to buy the bond, have it as inventory and sell it in smaller amounts. When trading illiquid bonds they more often quickly sell the entire position, so they perform more of a matching function in these bonds. This is consistent with our argument that illiquid bonds trade more often, which can be illustrated with the following example. In a liquid bond the investor sells $1,000,000 to a dealer, who sells it to investors in two amounts of $500,000. In an illiquid bond the investor sells 500,000 to two different dealers, who each sells the $500,000 to an investor. The total number of trades in the illiquid bond is four while it is three in the liquid bond. 42 Essay 2 (2007) for all TRACE bonds for the years 2005, 2006, and 2007 based on all TRACE trades. The median roundtrip cost is 237 basis points, which appears too high compared to findings in Edwards, Harris, and Piwowar (2007), Goldstein, Hotchkiss, and Sirri (2007), and Bessembinder, Maxwell, and Venkaraman (2006). From a theoretical point of view the mixed results regarding the impact of zero trading days on spreads can be explained by results in Huberman and Stanzl (2005). They show that an investor trades more often when price impact of trades is high, because he attempts to reduce the total price impact by submitting more but smaller orders. All else equal more trades therefore occur in illiquid bonds since it is necessary to split a sell order in many small trades, while it can be executed in a single trade in a liquid bond.7If this explanation holds true we should expect to see less zero trading days in illiquid times without an increase in the total trading volume. As Figure 2.1 shows this happens during the subprime crisis. The top-right graph shows that the median number of percentage zeros in the regression sample decreases during the subprime crisis. For example, the median number of percentage zeros is 30% in the last quarter of 2008 while it is 62% in the first quarter of 2007. Also, we see in the bottom-left graph that volume in our regression sample decreases slightly during the crisis. Drawing conclusions from Figure 2.1 might be misleading since a bond in a given quarter is only included in the regression sample if it has a full set of accounting variables and trades at least four times that quarter (otherwise the Roll measure cannot be calculated). Thus, it is only the most liquid bonds that are included and there are less bonds included post-subprime than pre-subprime. The decrease in zero trading days might therefore be due to a smaller number of bonds included in the sample. To address this concern, Figure 2.3 shows time series of the quarterly average number of trades and average trade size for all straight coupon bullet bond transactions in our sample period. The top graphs are based on transactions of size $100,000 or more, which our regression results are based on, while the bottom graphs are based on all transactions. In both cases we clearly see an increase on the average number of trades and a decrease in the average trade size after the onset of the subprime crises. Overall, there is theoretical evidence both in favor of and against the 7Goldstein, Hotchkiss, and Sirri (2007) find that dealers behave differently when trading liquid and illiquid bonds. When trading liquid bonds they are more likely to buy the bond, have it as inventory and sell it in smaller amounts. When trading illiquid bonds they more often quickly sell the entire position, so they perform more of a matching function in these bonds. This is consistent with our argument that illiquid bonds trade more often, which can be illustrated with the following example. In a liquid bond the investor sells $1,000,000 to a dealer, who sells it to investors in two amounts of $500,000. In an illiquid bond the investor sells 500,000 to two different dealers, who each sells the $500,000 to an investor. The total number of trades in the illiquid bond is four while it is three in the liquid bond. Corporate Bond Liquidity Before and After the Onset of the Subprime Crisis 43 2005 2006 2007 2008 2009 15 20 25 30 35 average number of trades, institutional trades 2005 2006 2007 2008 2009 5 5.5 6 6.5 7 7.5 x 10 5 average trade size, institutional trades 2005 2006 2007 2008 2009 60 80 100 120 140 160 180 200 220 240 average number of trades, all trades 2005 2006 2007 2008 2009 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 x 10 5 average trade size, all trades Figure 3: Time series of average number of trades and average trade size in the full sample. This graph plots the time series of average number of trades in a quarter and average trade size along with a line marking the start of the subprime crisis (beginning in 2007.Q2). Number of trades and trade size are measured quarterly and the mean value across all observations each quarter is graphed. A bond is included in every quarter if it traded at least one time during the sample period 2005:Q1-2008:Q4. The top graphs is based on institutional trades, i.e. trades of size $100,000 or more, while the bottom graphs are based on all trades. 61 Figure 2.3: Time series of average number of trades and average trade size in the full sample. This graph plots the time series of average number of trades in a quarter and average trade size along with a line marking the start of the subprime crisis (beginning in 2007.Q2). Number of trades and trade size are measured quarterly and the mean value across all observations each quarter is graphed. A bond is included in every quarter if it traded at least one time during the sample period 2005:Q1-2008:Q4. The top graphs is based on institutional trades, i.e. trades of size $100,000 or more, while the bottom graphs are based on all trades. 44 Essay 2 use of zero trading days as a measure of illiquidity. Our empirical evidence is also mixed. We show that trading activity increases when the market becomes more illiquid, while at the same time Table 2.2 shows that bond zero trading days do tend to predict investment grade spreads after the onset of the subprime crisis. In any case, we do not find that zero trading days can be consistently used as a predictor of spreads. 2.5.3 Principal component analysis of liquidity In our analysis we include eight liquidity proxies that measure different aspects of liquidity. To see if most of the relevant information in the proxies can be captured by a few factors, we conduct a principal component analysis. Table 2.3 shows the loadings and the explanatory power of the eight principal components. We see that both the explanatory power and the loadings of each PC component are very stable in the two subperiods. Also we see that the PC components have clear interpretations. The first component explains 40% of the variation in the liquidity variables and is close to being an equally-weighted linear combination of the Amihud and URC measures and their associated liquidity risk measures. The second PC explains 20% and is a zero trading days measure, the third PC explains 13% and is a volume measure, and the fourth PC explains 9% and is a Roll measure. The last four PCs explain less than 20% and do not have clear interpretations. Table 2.4 shows results of adding each of the PCs in turn to our regression in the same way as we did with each liquidity variable in Table 2.2. Strikingly, the first PC is significant for all rating categories pre- and post-subprime. For 9 out of 10 regression coefficients the significance is at a 1% level. In addition, the remaining seven PCs are mostly insignificant and often with conflicting signs. This suggests that although liquidity has many different aspects, a single linear combination of measures of transaction costs, market depth, and liquidity risk explains much of the impact of liquidity on yield spreads. The factor is priced at all ratings pre- and postsubprime in contrast to previously proposed liquidity proxies, zero-trading days (Chen, Lesmond, and Wei (2007)) and the Roll measure (Bao, Pan, and Wang (2009)). The principal component loadings on the first PC in Table 2.3 lead us to define a factor that loads evenly on Amihud, URC, Amihud risk, and URC risk, and does not load on any of the other liquidity measures. The factor is simpler to calculate than the first PC while retaining its properties. We use this factor in our subsequent analysis and call it λ. To be precise: for each bond iand quarter twe calculate the measure Lj it where j= 1, .., 4 is an index for Amihud, URC, Amihud risk, and URC risk. We normalize each measure ˜ Lj it =Lj it−µj σjwhere µjand σjare the mean and standard deviation of Ljacross bonds and quarters and define our liquidity measure for each 44 Essay 2 use of zero trading days as a measure of illiquidity. Our empirical evidence is also mixed. We show that trading activity increases when the market becomes more illiquid, while at the same time Table 2.2 shows that bond zero trading days do tend to predict investment grade spreads after the onset of the subprime crisis. In any case, we do not find that zero trading days can be consistently used as a predictor of spreads. 2.5.3 Principal component analysis of liquidity In our analysis we include eight liquidity proxies that measure different aspects of liquidity. To see if most of the relevant information in the proxies can be captured by a few factors, we conduct a principal component analysis. Table 2.3 shows the loadings and the explanatory power of the eight principal components. We see that both the explanatory power and the loadings of each PC component are very stable in the two subperiods. Also we see that the PC components have clear interpretations. The first component explains 40% of the variation in the liquidity variables and is close to being an equally-weighted linear combination of the Amihud and URC measures and their associated liquidity risk measures. The second PC explains 20% and is a zero trading days measure, the third PC explains 13% and is a volume measure, and the fourth PC explains 9% and is a Roll measure. The last four PCs explain less than 20% and do not have clear interpretations. Table 2.4 shows results of adding each of the PCs in turn to our regression in the same way as we did with each liquidity variable in Table 2.2. Strikingly, the first PC is significant for all rating categories pre- and post-subprime. For 9 out of 10 regression coefficients the significance is at a 1% level. In addition, the remaining seven PCs are mostly insignificant and often with conflicting signs. This suggests that although liquidity has many different aspects, a single linear combination of measures of transaction costs, market depth, and liquidity risk explains much of the impact of liquidity on yield spreads. The factor is priced at all ratings pre- and postsubprime in contrast to previously proposed liquidity proxies, zero-trading days (Chen, Lesmond, and Wei (2007)) and the Roll measure (Bao, Pan, and Wang (2009)). The principal component loadings on the first PC in Table 2.3 lead us to define a factor that loads evenly on Amihud, URC, Amihud risk, and URC risk, and does not load on any of the other liquidity measures. The factor is simpler to calculate than the first PC while retaining its properties. We use this factor in our subsequent analysis and call it λ. To be precise: for each bond iand quarter twe calculate the measure Lj it where j= 1, .., 4 is an index for Amihud, URC, Amihud risk, and URC risk. We normalize each measure ˜ Lj it =Lj it−µj σjwhere µjand σjare the mean and standard deviation of Ljacross bonds and quarters and define our liquidity measure for each Corporate Bond Liquidity Before and After the Onset of the Subprime Crisis 45 Panel A: Principal Component loadings, pre-subprime (2004:Q4-2007:Q1) 1PC 2PC 3PC 4PC 5PC 6PC 7PC 8PC Amihud 0.45 0.05 −0.12 −0.05 0.44 0.70 −0.12 0.28 Roll 0.26 0.33 0.08 −0.86 −0.27 −0.06 0.06 0.02 firm zero −0.04 0.64 −0.02 0.39 −0.56 0.36 0.07 0.02 bond zero −0.00 0.67 −0.10 0.10 0.56 −0.45 0.05 0.11 turnover −0.02 0.07 0.98 0.07 0.15 0.08 0.01 0.03 URC 0.52 0.06 0.03 0.15 0.00 −0.10 −0.39 −0.73 Amihud risk 0.47 −0.11 0.01 0.16 −0.01 −0.09 0.85 −0.09 URC risk 0.49 −0.12 0.06 0.21 −0.29 −0.40 −0.31 0.60 cum. % explained 39% 59% 72% 81% 89% 94% 99% 100% Panel B: Principal Component loadings, post-subprime (2007:Q2-2009:Q2) 1PC 2PC 3PC 4PC 5PC 6PC 7PC 8PC Amihud 0.46 0.04 −0.10 −0.10 −0.07 0.73 0.43 0.21 Roll 0.06 0.47 0.35 −0.78 0.10 −0.02 −0.17 0.02 firm zero −0.11 0.59 −0.28 0.33 0.62 0.20 −0.17 0.00 bond zero −0.12 0.64 −0.07 0.21 −0.67 −0.16 0.21 0.12 turnover −0.14 0.05 0.88 0.39 0.08 0.20 0.12 0.01 URC 0.52 0.15 0.06 0.09 0.09 −0.26 0.28 −0.73 Amihud risk 0.46 0.03 0.07 0.21 −0.30 0.19 −0.78 −0.04 URC risk 0.51 0.02 0.09 0.13 0.23 −0.51 0.10 0.63 cum. % explained 39% 58% 71% 81% 88% 94% 99% 100% Table 2.3: Principal component loadings on the liquidity variables. This table shows the principal component analysis loadings on each of the eight liquidity variables along with the cumulative explanatory power of the components. bond and quarter as λit = 4  j=1 ˜ Lj it 2.5.4 Size of liquidity component To calculate the impact of corporate bond illiquidity on yield spreads we do the following. For each rating Rwe run the pooled regression spreadR it =αR+βRλit + credit risk controlsit +it where irefers to bond, tto time (measured in quarters of year), and λit is our liquidity measure. We define the liquidity score for a bond in a given quarter as βRλit. Within each rating (AAA, AA, A, BBB, spec), period (pre- or post subprime), and maturity (0-2y, 2-5y, 5-30y) we sort all observations according to their liquidity score. The liquidity component of an average bond is defined as the 50% quantile minus the 5% quantile of the liquidity score distribution. Thus, the liquidity component measures the difference in bond yields between a bond with average liquidity and a very liquid bond. Following Cameron, Gelbach, and Miller (2008) we calculate 46 Essay 2 Panel A: Multivariate liquidity regressions, pre-subprime (2004:Q4-2007:Q1) AAA AA A BBB spec intercept −0.4 (−1.24) 0.2 (1.20) −0.5 (−1.62) 2.2∗∗∗ (2.84) −0.1 (−0.03) 1PCA 0.01∗∗∗ (3.22) 0.02∗∗∗ (12.31) 0.03∗∗∗ (3.28) 0.05∗∗∗ (2.88) 0.30∗∗∗ (5.65) 2PCA 0.01 (0.58) −0.00 (−0.09) 0.04∗∗∗ (3.41) −0.06 (−1.30) −0.19 (−1.19) 3PCA −0.014∗∗∗ (−4.20) −0.006 (−0.72) 0.018∗∗∗ (2.66) −0.005 (−0.21) 0.093 (0.88) 4PCA −0.020∗∗ (−2.32) −0.022∗∗∗ (−2.94) −0.002 (−0.18) −0.015 (−0.67) 0.112∗ (1.92) 5PCA 0.00 (0.01) 0.02∗∗∗ (3.08) 0.03∗ (1.88) −0.05 (−1.22) −0.02 (−0.16) 6PCA 0.00 (0.69) 0.01 (0.81) 0.03∗∗∗ (4.19) 0.03 (0.65) 0.24∗ (1.91) 7PCA 0.00 (0.27) −0.00 (−0.28) −0.00 (−0.55) −0.02∗ (−1.70) −0.10∗ (−1.68) 8PCA 0.02∗∗∗ (3.07) 0.02 (1.43) −0.01 (−0.74) −0.23∗∗∗ (−2.58) −0.17 (−1.56) age 0.00 (0.08) −0.00 (−0.96) 0.00 (1.12) −0.01 (−1.26) −0.00 (−0.12) amount issued −0.025∗∗∗ (−3.52) −0.012 (−1.34) 0.032∗∗ (2.57) −0.108∗∗∗ (−2.65) −0.143 (−0.87) forecast dispersion 3.05 (1.64) 0.02 (1.30) 0.73∗∗ (2.12) 0.65∗∗ (2.04) 1.21 (1.37) coupon 0.02∗∗ (1.99) 0.02∗∗∗ (4.00) 0.01∗ (1.79) 0.07∗∗∗ (4.46) 0.29∗∗∗ (3.62) 10y swap −0.05∗ (−1.82) −0.03∗∗∗ (−3.76) −0.05∗∗∗ (−4.23) −0.06∗∗∗ (−4.03) −0.26 (−1.33) 10y-2y swap 0.005 (0.79) −0.030∗∗ (−2.28) −0.020∗∗∗ (−2.89) −0.107∗∗∗ (−5.31) −0.132 (−0.44) equity vol −0.002 (−0.33) 0.008∗∗∗ (15.21) 0.006∗ (1.68) 0.011∗∗∗ (4.17) 0.093∗∗∗ (5.88) pretax1 0.344∗∗∗ (3.53) 0.023∗∗∗ (2.88) 0.010 (0.57) −0.026 (−1.36) 0.027 (0.44) pretax2 −0.051∗∗∗ (−3.06) −0.016∗∗∗ (−4.90) −0.011∗ (−1.90) −0.013 (−1.54) −0.068 (−0.90) pretax3 −0.007 (−1.00) 0.000 (0.18) −0.001 (−0.35) 0.011∗∗ (2.18) 0.048 (0.95) pretax4 −0.003∗∗∗ (−3.78) 0.000 (0.03) 0.000 (0.26) −0.005∗∗∗ (−3.31) −0.022 (−1.30) sales to income −0.002 (−1.14) −0.000 (−0.53) −0.000 (−0.01) −0.005∗∗ (−2.14) −0.003∗∗ (−1.97) long term debt to asset −0.016∗∗ (−2.49) −0.002∗∗∗ (−4.13) 0.001 (1.16) 0.008∗∗∗ (2.92) −0.001 (−0.02) leverage ratio 0.009∗∗∗ (3.04) 0.001 (1.58) −0.001 (−1.00) 0.000 (0.10) 0.023 (0.91) time-to-maturity 0.016∗∗∗ (3.50) 0.019∗∗∗ (18.21) 0.022∗∗∗ (15.21) 0.040∗∗∗ (7.95) 0.043∗∗∗ (2.99) N533 1869 4148 1340 1075 R20.46 0.53 0.47 0.60 0.61 Table 2.4: Multivariate liquidity regressions. For each of the five rating classes a pooled regression with quarterly observations is run with variables measuring both liquidity and credit risk. Panel A shows the regression coefficients and t-statistics in parenthesis when using data from 2004:Q4 to 2007:Q1, while Panel B shows the results for data from 2007:Q2 to 2009:Q2. Standard errors are corrected for time series effects, firm fixed effects, and heteroscedasticity, and significance at 10% level is marked ’*’, at 5% marked ’**’, and at 1% marked ’***’. 46 Essay 2 Panel A: Multivariate liquidity regressions, pre-subprime (2004:Q4-2007:Q1) AAA AA A BBB spec intercept −0.4 (−1.24) 0.2 (1.20) −0.5 (−1.62) 2.2∗∗∗ (2.84) −0.1 (−0.03) 1PCA 0.01∗∗∗ (3.22) 0.02∗∗∗ (12.31) 0.03∗∗∗ (3.28) 0.05∗∗∗ (2.88) 0.30∗∗∗ (5.65) 2PCA 0.01 (0.58) −0.00 (−0.09) 0.04∗∗∗ (3.41) −0.06 (−1.30) −0.19 (−1.19) 3PCA −0.014∗∗∗ (−4.20) −0.006 (−0.72) 0.018∗∗∗ (2.66) −0.005 (−0.21) 0.093 (0.88) 4PCA −0.020∗∗ (−2.32) −0.022∗∗∗ (−2.94) −0.002 (−0.18) −0.015 (−0.67) 0.112∗ (1.92) 5PCA 0.00 (0.01) 0.02∗∗∗ (3.08) 0.03∗ (1.88) −0.05 (−1.22) −0.02 (−0.16) 6PCA 0.00 (0.69) 0.01 (0.81) 0.03∗∗∗ (4.19) 0.03 (0.65) 0.24∗ (1.91) 7PCA 0.00 (0.27) −0.00 (−0.28) −0.00 (−0.55) −0.02∗ (−1.70) −0.10∗ (−1.68) 8PCA 0.02∗∗∗ (3.07) 0.02 (1.43) −0.01 (−0.74) −0.23∗∗∗ (−2.58) −0.17 (−1.56) age 0.00 (0.08) −0.00 (−0.96) 0.00 (1.12) −0.01 (−1.26) −0.00 (−0.12) amount issued −0.025∗∗∗ (−3.52) −0.012 (−1.34) 0.032∗∗ (2.57) −0.108∗∗∗ (−2.65) −0.143 (−0.87) forecast dispersion 3.05 (1.64) 0.02 (1.30) 0.73∗∗ (2.12) 0.65∗∗ (2.04) 1.21 (1.37) coupon 0.02∗∗ (1.99) 0.02∗∗∗ (4.00) 0.01∗ (1.79) 0.07∗∗∗ (4.46) 0.29∗∗∗ (3.62) 10y swap −0.05∗ (−1.82) −0.03∗∗∗ (−3.76) −0.05∗∗∗ (−4.23) −0.06∗∗∗ (−4.03) −0.26 (−1.33) 10y-2y swap 0.005 (0.79) −0.030∗∗ (−2.28) −0.020∗∗∗ (−2.89) −0.107∗∗∗ (−5.31) −0.132 (−0.44) equity vol −0.002 (−0.33) 0.008∗∗∗ (15.21) 0.006∗ (1.68) 0.011∗∗∗ (4.17) 0.093∗∗∗ (5.88) pretax1 0.344∗∗∗ (3.53) 0.023∗∗∗ (2.88) 0.010 (0.57) −0.026 (−1.36) 0.027 (0.44) pretax2 −0.051∗∗∗ (−3.06) −0.016∗∗∗ (−4.90) −0.011∗ (−1.90) −0.013 (−1.54) −0.068 (−0.90) pretax3 −0.007 (−1.00) 0.000 (0.18) −0.001 (−0.35) 0.011∗∗ (2.18) 0.048 (0.95) pretax4 −0.003∗∗∗ (−3.78) 0.000 (0.03) 0.000 (0.26) −0.005∗∗∗ (−3.31) −0.022 (−1.30) sales to income −0.002 (−1.14) −0.000 (−0.53) −0.000 (−0.01) −0.005∗∗ (−2.14) −0.003∗∗ (−1.97) long term debt to asset −0.016∗∗ (−2.49) −0.002∗∗∗ (−4.13) 0.001 (1.16) 0.008∗∗∗ (2.92) −0.001 (−0.02) leverage ratio 0.009∗∗∗ (3.04) 0.001 (1.58) −0.001 (−1.00) 0.000 (0.10) 0.023 (0.91) time-to-maturity 0.016∗∗∗ (3.50) 0.019∗∗∗ (18.21) 0.022∗∗∗ (15.21) 0.040∗∗∗ (7.95) 0.043∗∗∗ (2.99) N533 1869 4148 1340 1075 R20.46 0.53 0.47 0.60 0.61 Table 2.4: Multivariate liquidity regressions. For each of the five rating classes a pooled regression with quarterly observations is run with variables measuring both liquidity and credit risk. Panel A shows the regression coefficients and t-statistics in parenthesis when using data from 2004:Q4 to 2007:Q1, while Panel B shows the results for data from 2007:Q2 to 2009:Q2. Standard errors are corrected for time series effects, firm fixed effects, and heteroscedasticity, and significance at 10% level is marked ’*’, at 5% marked ’**’, and at 1% marked ’***’. Corporate Bond Liquidity Before and After the Onset of the Subprime Crisis 47 Panel B: Multivariate liquidity regressions, post-subprime (2007:Q2-2009:Q2) AAA AA A BBB spec intercept −2.5∗∗ (−2.00) −2.6 (−1.00) 1.0∗∗∗ (2.66) 24.9 (1.42) 30.2∗ (1.65) 1PCA 0.05∗ (1.91) 0.48∗∗∗ (4.50) 0.45∗∗∗ (4.64) 0.67∗∗∗ (3.18) 1.16∗∗∗ (4.33) 2PCA −0.08 (−0.57) 0.15 (1.60) 0.26∗∗ (2.27) −0.03 (−0.05) −0.73 (−1.21) 3PCA 0.066 (1.21) 0.153∗∗∗ (2.96) 0.146∗∗∗ (3.27) 0.389∗ (1.75) 0.349 (0.90) 4PCA −0.125 (−1.35) 0.283∗∗∗ (5.14) 0.267∗∗∗ (4.07) 0.110∗ (1.81) 0.900 (1.40) 5PCA −0.35∗∗∗ (−2.75) −0.18 (−1.17) −0.17∗∗∗ (−7.65) −0.46 (−0.90) 0.52 (0.97) 6PCA −0.09∗ (−1.76) −0.17 (−1.30) −0.41∗ (−1.67) −0.30∗ (−1.70) 1.00∗∗ (2.57) 7PCA 0.07 (0.68) −0.39∗ (−1.79) −0.22 (−1.24) −0.44 (−1.08) −0.58∗∗ (−1.98) 8PCA 0.12∗ (1.72) 0.07 (0.30) −0.29∗∗ (−2.14) 1.04 (1.11) 0.63 (0.54) age −0.03∗∗∗ (−4.83) −0.02 (−0.84) 0.02 (0.52) 0.10 (1.02) 0.18∗∗∗ (3.12) amount issued 0.087∗∗∗ (4.22) 0.101 (1.27) 0.009 (0.09) −0.715 (−1.04) −0.571 (−0.72) forecast dispersion 18.32∗∗∗ (3.07) 0.13∗∗∗ (3.75) 0.15∗∗∗ (3.34) 0.76∗∗∗ (7.31) 1.06∗∗∗ (4.31) coupon 0.10∗∗∗ (4.46) 0.10∗∗ (2.07) 0.02 (0.17) −0.50 (−1.34) −0.09 (−0.19) 10y swap −0.32∗∗∗ (−6.18) 0.07 (0.24) −0.09 (−0.22) −1.33∗∗∗ (−3.25) −3.18∗∗∗ (−3.05) 10y-2y swap −0.400∗∗ (−2.17) −0.490 (−1.58) −0.820∗ (−1.95) −0.962 (−1.23) −1.962∗∗∗ (−2.59) equity vol 0.096∗∗∗ (6.22) 0.055∗∗∗ (3.82) 0.050∗∗∗ (3.64) 0.050∗∗∗ (3.06) 0.097∗∗∗ (3.24) pretax1 −0.836∗∗ (−2.17) 0.004 (0.21) −0.098∗ (−1.80) −0.051 (−0.53) 0.001 (0.44) pretax2 0.422∗∗∗ (5.33) 0.033 (0.93) −0.000 (−0.00) −0.073 (−0.53) −0.442 (−0.53) pretax3 0.144 (0.78) −0.041∗∗∗ (−2.59) −0.003 (−0.37) 0.076 (0.81) 0.000 (NaN) pretax4 0.003 (0.65) 0.052∗ (1.83) 0.008 (0.50) −0.067 (−0.62) 0.000 (NaN) sales to income −0.108∗ (−1.68) −0.003∗∗∗ (−4.56) −0.001∗∗∗ (−3.79) −0.002∗∗∗ (−7.81) −0.013 (−1.25) long term debt to asset −0.256∗∗∗ (−2.67) −0.009 (−0.71) 0.044∗∗ (2.40) 0.058 (1.56) −0.108∗∗∗ (−4.57) leverage ratio 0.184∗ (1.92) 0.000 (0.00) −0.026∗∗∗ (−3.55) −0.005 (−0.17) 0.106∗∗∗ (13.08) time-to-maturity 0.024∗∗∗ (6.00) −0.015 (−0.96) −0.035∗ (−1.72) −0.064 (−1.43) −0.124∗∗∗ (−2.63) N414 1549 2533 539 464 R20.84 0.71 0.67 0.79 0.72 Table 2.4: continued. 48 Essay 2 confidence bands by performing a wild cluster bootstrap of the regression residuals. Table 2.5 shows the size of the liquidity component. We see that the liquidity component becomes larger as the rating quality of the bond decreases. For investment grade ratings, the component is small with an average presubprime across maturity of 0.8bp for AAA, 1.0bp for AA, 2.4bp for A, and 3.9bp for BBB. For speculative grade the liquidity component is larger and estimated to be 57.6bp. Panel A: Liquidity component in basis points, pre-subprime (2004Q4-2007:Q1) average 0-2y 2-5y 5-30y N 0-2y N 2-5y N 5-30y AAA 0.80.6 (0.3;0.8) 0.9 (0.5;1.3) 1.1 (0.6;1.5) 162 178 193 AA 1.00.7 (0.3;1.1) 1.0 (0.4;1.7) 1.3 (0.5;2.2) 704 667 498 A 2.41.5 (0.6;2.3) 2.5 (1.1;3.9) 3.2 (1.4;4.9) 1540 1346 1260 BBB 3.92.8 (1.4;4.4) 4.0 (1.9;6.2) 4.7 (2.3;7.3) 517 270 553 spec 57.645.0 (32.3;57.4) 44.0 (31.5;56.0) 83.9 (60.2;106.8) 270 324 480 Panel B: Liquidity component in basis points, post-subprime (2007:Q2-2009:Q2) average 0-2y 2-5y 5-30y N 0-2y N 2-5y N 5-30y AAA 4.92.5 (0.5;4.4) 4.5 (0.9;8.0) 7.9 (1.7;14.1) 110 149 155 AA 41.823.5 (12.9;33.2) 37.1 (20.3;52.4) 64.7 (35.5;91.4) 493 572 483 A 50.726.6 (15.3;39.2) 51.0 (29.3;75.1) 74.5 (42.9;109.7) 762 878 890 BBB 92.764.3 (36.5;92.7) 115.6 (65.6;166.6) 98.1 (55.7;141.4) 123 159 256 spec 196.8123.6 (80.2;157.3) 224.0 (145.3;285.1) 242.7 (157.4;308.8) 133 129 201 Table 2.5: Liquidity Component in basis points. For each rating Rwe run the pooled regression spreadR it =αR+βRλit + credit risk controlsit +it where irefers to bond, tto time, and λit is our liquidity measure. The bond spread is measured with respect to the swap rate. Within each rating and maturity bucket (0-2y, 2-5y, and 5-30y) we sort increasingly all values of λit and find the median value λ50 and the 5% value λ5. The liquidity component in the bucket is defined as β(λ50 −λ5). This table shows for all buckets the liquidity component with standard errors in parenthesis. Confidence bands are found by a wild cluster bootstrap. 48 Essay 2 confidence bands by performing a wild cluster bootstrap of the regression residuals. Table 2.5 shows the size of the liquidity component. We see that the liquidity component becomes larger as the rating quality of the bond decreases. For investment grade ratings, the component is small with an average presubprime across maturity of 0.8bp for AAA, 1.0bp for AA, 2.4bp for A, and 3.9bp for BBB. For speculative grade the liquidity component is larger and estimated to be 57.6bp. Panel A: Liquidity component in basis points, pre-subprime (2004Q4-2007:Q1) average 0-2y 2-5y 5-30y N 0-2y N 2-5y N 5-30y AAA 0.8 0.6 (0.3;0.8) 0.9 (0.5;1.3) 1.1 (0.6;1.5) 162 178 193 AA 1.0 0.7 (0.3;1.1) 1.0 (0.4;1.7) 1.3 (0.5;2.2) 704 667 498 A 2.4 1.5 (0.6;2.3) 2.5 (1.1;3.9) 3.2 (1.4;4.9) 1540 1346 1260 BBB 3.9 2.8 (1.4;4.4) 4.0 (1.9;6.2) 4.7 (2.3;7.3) 517 270 553 spec 57.6 45.0 (32.3;57.4) 44.0 (31.5;56.0) 83.9 (60.2;106.8) 270 324 480 Panel B: Liquidity component in basis points, post-subprime (2007:Q2-2009:Q2) average 0-2y 2-5y 5-30y N 0-2y N 2-5y N 5-30y AAA 4.9 2.5 (0.5;4.4) 4.5 (0.9;8.0) 7.9 (1.7;14.1) 110 149 155 AA 41.8 23.5 (12.9;33.2) 37.1 (20.3;52.4) 64.7 (35.5;91.4) 493 572 483 A 50.7 26.6 (15.3;39.2) 51.0 (29.3;75.1) 74.5 (42.9;109.7) 762 878 890 BBB 92.7 64.3 (36.5;92.7) 115.6 (65.6;166.6) 98.1 (55.7;141.4) 123 159 256 spec 196.8 123.6 (80.2;157.3) 224.0 (145.3;285.1) 242.7 (157.4;308.8) 133 129 201 Table 2.5: Liquidity Component in basis points. For each rating Rwe run the pooled regression spreadR it =αR+βRλit + credit risk controlsit +it where irefers to bond, tto time, and λit is our liquidity measure. The bond spread is measured with respect to the swap rate. Within each rating and maturity bucket (0-2y, 2-5y, and 5-30y) we sort increasingly all values of λit and find the median value λ50 and the 5% value λ5. The liquidity component in the bucket is defined as β(λ50 −λ5). This table shows for all buckets the liquidity component with standard errors in parenthesis. Confidence bands are found by a wild cluster bootstrap. Corporate Bond Liquidity Before and After the Onset of the Subprime Crisis 49 There is a strong increase in the liquidity component in the post-subprime period as Panel B in Table 2.5 shows. The component increases by a factor 10 or more in investment grade bonds of rating AA, A, and BBB while it increases by a factor 3-4 in speculative grade bonds. This shows that liquidity has dried out under the subprime crisis and part of the spread widening for bonds is due to a higher liquidity premium. Figure 2.1 shows the evolution of liquidity variables over the sample, and we see that the liquidity variables entering our measure of liquidity (Amihud, URC, Amihud risk, URC risk) all increase strongly after the onset of the subprime crisis. Thus, the higher liquidity premium is due to an increase in the sensitivity of spreads to illiquidity as well as higher levels of illiquidity. While liquidity components in all ratings increase, we see that in absolute terms the increase in AAA bonds is modest. Even after the onset of the subprime crisis the component is 8 basis points or less, which is small compared to the component of other bonds. We see in Table 2.5 that the regression coefficient for AAA on the first principal component is small post-subprime compared to those of other rating classes, so the sensitivity of AAA-rated bonds to liquidity is small.8This suggests that there is a flight-to-quality into AAA bonds, namely that investors are buying high-quality AAA-rated bonds regardless of their liquidity. The average liquidity premium in speculative grade bonds was 57.6bp pre-subprime, so even in this liquidity-rich period speculative grade bonds commanded a sizeable liquidity premium. Post-subprime the liquidity premium increased to 196.8bp for speculative grade bonds. An A rated bond has an average liquidity premium of 50.7bp post-subprime, so the illiquidity of such a bond post-subprime is similar to that of a speculative grade bond pre-subprime. The size of the liquidity component pre-subprime is comparable in magnitude to the nondefault component in investment grade corporate bond spreads found by subtracting the CDS premium from the corporate - swap spread (swap basis), see Longstaff, Mithal, and Neis (2005), Blanco, Brennan, and Marsh (2005), and Han and Zhou (2008).9These papers look at recent periods before the subprime crisis and our pre-subprime results agree with their results in that there is a modest liquidity premium in investment grade corporate bond yields. The nondefault component for speculative bonds extracted from the swap basis is smaller and often negative, and the evidence presented here suggests that other factors than corporate bond liq- 8Strictly speaking, we use our measure λto calculate liquidity components, but the regression coefficient on λis close to the coefficient on 1PC. 9Longstaff, Mithal, and Neis (2005) find an average nondefault component of -7.2bp for AAA/AA, 10.5bp for A, and 9.7bp for BBB, Han and Zhou (2008) find the nondefault component to be 0.3bp for AAA, 3.3bp for AA, 6.7bp for A, and 23.5bp for BBB, while Blanco, Brennan, and Marsh (2005) find it to be 6.9bp for AAA/AA, 0.5bp for A, and 14.9bp for BBB. 56 Essay 2 pre-subprime post-subprime β λ β λ AAA −0.0034 (−1.34) −0.0085 (−0.84) −0.0056∗∗∗ (−3.26) 0.0033∗∗∗ (2.65) 0.0159 (1.26) 0.0234∗∗ (2.38) AA 0.0012 (0.23) 0.1823∗ (1.94) 0.0067 (1.06) 0.0017 (0.60) 0.1720∗∗ (2.14) 0.1712∗∗∗ (3.82) A−0.0004 (−0.14) 0.2631∗∗ (2.22) 0.0021 (0.65) 0.0106∗∗ (2.57) 0.2314∗∗ (2.15) 0.1211∗∗ (2.03) BBB 0.0044 (1.34) 0.2171∗∗∗ (4.05) 0.0012 (0.34) 0.0254∗∗∗ (4.33) 0.3187∗∗∗ (3.44) 0.3242∗∗∗ (2.91) spec 0.0102 (0.90) 1.3538∗∗∗ (2.60) 0.0162 (1.31) 0.1502∗∗∗ (4.64) 1.3140∗∗ (2.73) 0.4155∗∗∗ (7.08) Table 2.8: βregressions. For each rating class Rpooled regressions are run where yield spreads are regressed on each bonds liquidity βand our liquidity measure λtwith credit risk controls SpreadR it =αR+γR 1λit +γR 2βi+ credit risk controlsit +it where iis for bond in rating Rand tis time measured in quarter. Each bond’s βiis calculated as the covariance between this bond’s monthly λit and a size-weighted monthly market λMt. Two regressions for each rating pre- and post-subprime are run; one with only βincluded and one with both βand λincluded. Standard errors are corrected for time series effects, firm fixed effects, and heteroscedasticity, and significance at 10% level is marked ’*’, at 5% marked ’**’, and at 1% marked ’***’. . they use stock and Treasury bond market liquidity to measure aggregate liquidity, our measure specifically captures corporate bond market liquidity. We saw in the previous section that the contribution to spreads of liquidity was small for AAA bonds after the onset of the crisis, and the insignificant liquidity beta coefficient for AAA in the crisis period confirms that there is a flight-to-quality effect in AAA-rated bonds. 2.6.2 Lead underwriter Brunnermeier and Pedersen (2009) provide a model that links an asset’s market liquidity and traders’ funding liquidity, and find that when funding liquidity is tight, traders become reluctant to take on positions, especially ”capital intensive” positions in high-margin securities. This lowers market liquidity. Empirical support for this prediction is found in Comerton-Forde, Hendershott, Jones, Moulton, and Seasholes (2010) who find for equities traded on NYSE that balance sheet and income statement variables for mar- 56 Essay 2 pre-subprime post-subprime β λ β λ AAA −0.0034 (−1.34) −0.0085 (−0.84) −0.0056∗∗∗ (−3.26) 0.0033∗∗∗ (2.65) 0.0159 (1.26) 0.0234∗∗ (2.38) AA 0.0012 (0.23) 0.1823∗ (1.94) 0.0067 (1.06) 0.0017 (0.60) 0.1720∗∗ (2.14) 0.1712∗∗∗ (3.82) A−0.0004 (−0.14) 0.2631∗∗ (2.22) 0.0021 (0.65) 0.0106∗∗ (2.57) 0.2314∗∗ (2.15) 0.1211∗∗ (2.03) BBB 0.0044 (1.34) 0.2171∗∗∗ (4.05) 0.0012 (0.34) 0.0254∗∗∗ (4.33) 0.3187∗∗∗ (3.44) 0.3242∗∗∗ (2.91) spec 0.0102 (0.90) 1.3538∗∗∗ (2.60) 0.0162 (1.31) 0.1502∗∗∗ (4.64) 1.3140∗∗ (2.73) 0.4155∗∗∗ (7.08) Table 2.8: βregressions. For each rating class Rpooled regressions are run where yield spreads are regressed on each bonds liquidity βand our liquidity measure λtwith credit risk controls SpreadR it =αR+γR 1λit +γR 2βi+ credit risk controlsit +it where iis for bond in rating Rand tis time measured in quarter. Each bond’s βiis calculated as the covariance between this bond’s monthly λit and a size-weighted monthly market λMt. Two regressions for each rating pre- and post-subprime are run; one with only βincluded and one with both βand λincluded. Standard errors are corrected for time series effects, firm fixed effects, and heteroscedasticity, and significance at 10% level is marked ’*’, at 5% marked ’**’, and at 1% marked ’***’. . they use stock and Treasury bond market liquidity to measure aggregate liquidity, our measure specifically captures corporate bond market liquidity. We saw in the previous section that the contribution to spreads of liquidity was small for AAA bonds after the onset of the crisis, and the insignificant liquidity beta coefficient for AAA in the crisis period confirms that there is a flight-to-quality effect in AAA-rated bonds. 2.6.2 Lead underwriter Brunnermeier and Pedersen (2009) provide a model that links an asset’s market liquidity and traders’ funding liquidity, and find that when funding liquidity is tight, traders become reluctant to take on positions, especially ”capital intensive” positions in high-margin securities. This lowers market liquidity. Empirical support for this prediction is found in Comerton-Forde, Hendershott, Jones, Moulton, and Seasholes (2010) who find for equities traded on NYSE that balance sheet and income statement variables for mar- Corporate Bond Liquidity Before and After the Onset of the Subprime Crisis 57 ket makers explain time variation in liquidity. Since the TRACE data do not reveal the identity of the traders, we cannot perform direct tests of the Brunnermeier and Pedersen (2009)-model for the U.S. corporate bond market. However, if we assume that the original underwriter is more likely to make a market (as is the case in equity markets, see Ellis, Michaely, and O’Hara (2000)), we can provide indirect evidence by observing bond liquidity of bonds underwritten by Bear Stearns and Lehman Brothers, two financial institutions in distress during the subprime crisis. We therefore calculate for all bonds with Lehman Brothers as lead underwriter their average λweighted by amount outstanding - on a monthly basis. Likewise, we does this for bonds with Bear Stearns as lead underwriter and for all bonds in the sample. We obtain underwriter information from FISD. The results are plotted in Figure 2.5. The liquidity of bonds with Bear Stearns as lead underwriter was roughly the same as an average bond entering the summer of 2007. During the week of July 16, 2007 Bear Stearns disclosed that two of their hedge funds had lost nearly all of the value, and the graph shows that the ’illiquidity gap’ between Bear Stearns underwritten bonds and average bonds increased this month. On August 6, Bear Stearns said that it was weathering the worst storm in financial markets in more than 20 years, in November 2007 Bear Stearns wrote down $1.62 billion and booked a fourth quarter loss, and in December 2007 there was a further write-down of $1.90 billion. During these months, the ’illiquidity gap’ steadily increased. Bear Stearns were in severe liquidity problems in beginning of March, and they were taken over by JPMorgan on March 16. In this month the ’illiquidity gap’ peaked but returned to zero in June 2008 after Bear Stearns shareholders approved JPMorgan’s buyout of the investment bank on May 29. The liquidity of bonds underwritten by Lehman was close to the liquidity of an average bond in the market up until August 2008, but this changed in September 2008 when the ’liquidity gap’ between Lehman underwritten bonds and average market bonds increased strongly in September in response to Lehman filing for bankruptcy on September 15. The gap stayed at high levels during the rest of the sample period suggesting that after the Lehman default, bonds they had underwritten became permanently more illiquid. 2.6.3 Industry Bonds issued by financial firms might by more or less liquid compared to bonds issued by industrial firms. They might be less liquid because financial firms are more opaque, especially in times of financial distress, and their bonds might be more affected by asymmetric information. They might be more liquid because financial firms are more connected to capital markets and are liquidity providers to the market. The empirical evidence is mixed. Longstaff, Mithal, and Neis (2005) 58 Essay 2 Jan05 Apr05 Jul05 Oct05 Jan06 Apr06 Jul06 Oct06 Jan07 Apr07 Jul07 Oct07 Jan08 Apr08 Jul08 Oct08 Jan09 Apr09 −2 0 2 4 6 8 10 12 λ Market Lehman Brothers Bear Stearns Figure 5: Illiquidity of bonds underwritten by Lehman Brothers and Bear Stearns. This graph shows the time series variation in illiquidity of bonds with Lehman Brothers as lead underwriter, bonds with Bear Stearns as lead underwriter, and all bonds in the sample. For every bond underwritten by Lehman Brothers their (il)liquidity measure λis calculated each month and a monthly weighted average is calculated using amount outstanding for each bond as weight. The graph shows the time series of monthly averages. Likewise, a time series of monthly averages is calculated for bonds with Bear Stearns as a lead underwriter and all bonds in the sample. Higher values on the y-axis imply more illiquid bonds. 63 Figure 2.5: Illiquidity of bonds underwritten by Lehman Brothers and Bear Stearns. This graph shows the time series variation in illiquidity of bonds with Lehman Brothers as lead underwriter, bonds with Bear Stearns as lead underwriter, and all bonds in the sample. For every bond underwritten by Lehman Brothers their (il)liquidity measure λis calculated each month and a monthly weighted average is calculated using amount outstanding for each bond as weight. The graph shows the time series of monthly averages. Likewise, a time series of monthly averages is calculated for bonds with Bear Stearns as a lead underwriter and all bonds in the sample. Higher values on the y-axis imply more illiquid bonds. 58 Essay 2 Jan05 Apr05 Jul05 Oct05 Jan06 Apr06 Jul06 Oct06 Jan07 Apr07 Jul07 Oct07 Jan08 Apr08 Jul08 Oct08 Jan09 Apr09 −2 0 2 4 6 8 10 12 λ Market Lehman Brothers Bear Stearns Figure 5: Illiquidity of bonds underwritten by Lehman Brothers and Bear Stearns. This graph shows the time series variation in illiquidity of bonds with Lehman Brothers as lead underwriter, bonds with Bear Stearns as lead underwriter, and all bonds in the sample. For every bond underwritten by Lehman Brothers their (il)liquidity measure λis calculated each month and a monthly weighted average is calculated using amount outstanding for each bond as weight. The graph shows the time series of monthly averages. Likewise, a time series of monthly averages is calculated for bonds with Bear Stearns as a lead underwriter and all bonds in the sample. Higher values on the y-axis imply more illiquid bonds. 63 Figure 2.5: Illiquidity of bonds underwritten by Lehman Brothers and Bear Stearns. This graph shows the time series variation in illiquidity of bonds with Lehman Brothers as lead underwriter, bonds with Bear Stearns as lead underwriter, and all bonds in the sample. For every bond underwritten by Lehman Brothers their (il)liquidity measure λis calculated each month and a monthly weighted average is calculated using amount outstanding for each bond as weight. The graph shows the time series of monthly averages. Likewise, a time series of monthly averages is calculated for bonds with Bear Stearns as a lead underwriter and all bonds in the sample. Higher values on the y-axis imply more illiquid bonds. Corporate Bond Liquidity Before and After the Onset of the Subprime Crisis 59 Jan05 Apr05 Jul05 Oct05 Jan06 Apr06 Jul06 Oct06 Jan07 Apr07 Jul07 Oct07 Jan08 Apr08 Jul08 Oct08 Jan09 Apr09 −2 −1 0 1 2 3 4 λ Bonds of industrial firms Bonds of financial firms Figure 6: Illiquidity of bonds of industrial and financial firms. This graph shows the time series variation in illiquidity of bonds of industrial and financial firms. For every bond issued by a financial firm their (il)liquidity measure λis calculated each month and a monthly weighted average is calculated using amount outstanding for each bond as weight. The graph shows the time series of monthly averages. Likewise, a time series of monthly averages is calculated for bonds issued by industrial firms. Higher values on the y-axis imply more illiquid bonds. 64 Figure 2.6: Illiquidity of bonds of industrial and financial firms. This graph shows the time series variation in illiquidity of bonds of industrial and financial firms. For every bond issued by a financial firm their (il)liquidity measure λis calculated each month and a monthly weighted average is calculated using amount outstanding for each bond as weight. The graph shows the time series of monthly averages. Likewise, a time series of monthly averages is calculated for bonds issued by industrial firms. Higher values on the y-axis imply more illiquid bonds. 60 Essay 2 find in a study of 68 bonds that bonds issued by financial firms are more illiquid and command a higher liquidity premium. In contrast, Friewald, Jankowitsch, and Subrahmanyam (2009) find that there is no difference, except during the subprime crisis where bonds of financial firms are in fact more liquid. We address the issue by calculating a value-weighted average monthly illiquidity λof financial respectively industrial firms and plotting the time series behavior in Figure 2.6. We obtain bond industry from FISD. In general, there is little systematic difference. For both financial and industrial bonds, illiquidity goes up at the onset of the crisis. There are, however, additional spikes in illiquidity for financial firms around the takeover of Bear Stearns in March 2008, around the Lehman bankruptcy in September 2008, and around the stock market decline in the first quarter of 2009. That is, in times of severe financial distress, illiquidity of financial bonds increases relative to that of industrial bonds, while in other times illiquidity is similar. By calculating monthly averages, we are able to draw more high-frequency inferences compared to other papers, since averaging λover longer periods of time, the approach taken in Longstaff, Mithal, and Neis (2005) and Friewald, Jankowitsch, and Subrahmanyam (2009), would wash out the effects we see. 2.7 Conclusion The subprime crisis dramatically increased corporate bond spreads and while default risk certainly has increased because of funding constraints and the slowing of the real economy, it is also widely believed that deteriorating liquidity has contributed to the widening of spreads. The difficulty is how to measure this contribution. In this paper, we show that an equally weighted sum of four (normalized) measures of liquidity and liquidity risk consistently contributes to corporate bond spreads across time and across ratings. The four measures are the Amihud measure of price impact, a measure of roundtrip trading costs and the variability of these two measures. The equally weighted sum is a close approximation to the first factor in a principal component analysis of eight liquidity measures, and this is true both before and after the onset of the crisis. Our measure dominates other liquidity measures, such as the Roll measure and zero trading days. The measure is used to analyze the contribution of illiquidity to corporate bond spreads before and after the onset of the subprime crisis. We find that before the crisis, the contribution to spreads from illiquidty was small for investment grade bonds both measured in basis points and as a fraction of total spreads. The contribution increased strongly at the onset of the crisis for all bonds except AAA-rated bonds, which is consistent with a flight-to- quality into AAA-rated bonds. Liquidity premia in investment grade bonds rose steadily during the crisis and peaked when the stock market declined 60 Essay 2 find in a study of 68 bonds that bonds issued by financial firms are more illiquid and command a higher liquidity premium. In contrast, Friewald, Jankowitsch, and Subrahmanyam (2009) find that there is no difference, except during the subprime crisis where bonds of financial firms are in fact more liquid. We address the issue by calculating a value-weighted average monthly illiquidity λof financial respectively industrial firms and plotting the time series behavior in Figure 2.6. We obtain bond industry from FISD. In general, there is little systematic difference. For both financial and industrial bonds, illiquidity goes up at the onset of the crisis. There are, however, additional spikes in illiquidity for financial firms around the takeover of Bear Stearns in March 2008, around the Lehman bankruptcy in September 2008, and around the stock market decline in the first quarter of 2009. That is, in times of severe financial distress, illiquidity of financial bonds increases relative to that of industrial bonds, while in other times illiquidity is similar. By calculating monthly averages, we are able to draw more high-frequency inferences compared to other papers, since averaging λover longer periods of time, the approach taken in Longstaff, Mithal, and Neis (2005) and Friewald, Jankowitsch, and Subrahmanyam (2009), would wash out the effects we see. 2.7 Conclusion The subprime crisis dramatically increased corporate bond spreads and while default risk certainly has increased because of funding constraints and the slowing of the real economy, it is also widely believed that deteriorating liquidity has contributed to the widening of spreads. The difficulty is how to measure this contribution. In this paper, we show that an equally weighted sum of four (normalized) measures of liquidity and liquidity risk consistently contributes to corporate bond spreads across time and across ratings. The four measures are the Amihud measure of price impact, a measure of roundtrip trading costs and the variability of these two measures. The equally weighted sum is a close approximation to the first factor in a principal component analysis of eight liquidity measures, and this is true both before and after the onset of the crisis. Our measure dominates other liquidity measures, such as the Roll measure and zero trading days. The measure is used to analyze the contribution of illiquidity to corporate bond spreads before and after the onset of the subprime crisis. We find that before the crisis, the contribution to spreads from illiquidty was small for investment grade bonds both measured in basis points and as a fraction of total spreads. The contribution increased strongly at the onset of the crisis for all bonds except AAA-rated bonds, which is consistent with a flight-to- quality into AAA-rated bonds. Liquidity premia in investment grade bonds rose steadily during the crisis and peaked when the stock market declined Corporate Bond Liquidity Before and After the Onset of the Subprime Crisis 61 strongly in the first quarter of 2009, while premia in speculative grade bonds peaked during the Lehman default and returned almost to pre-crisis levels in mid-2009. The number of zero trading days did not increase with the crisis and we find evidence that this was because trades in less liquid bonds were split into trades of smaller size. Our measure is useful for analyzing other important aspects of corporate bond liquidity. From the covariation between an individual bond’s liquidity measure and a value-weighted average of all bonds’ liquidity measures, we define a liquidity beta which is shown to have little effect on spreads before the onset of the crisis, but does have a positive effect for all bonds except AAA-bonds after the crisis. This is consistent with the regime-dependent role of liquidity betas found in Acharya, Amihud, and Bharath (2010) but it narrows the flight-to-quality story from general investment grade bonds to AAA-rated bonds only. We also use our measure to study the impact on bond liquidity of funding shocks to lead underwriters and to compare illiquidity of corporate bonds issued by financial firms with that of industrial firms. Financial distress of lead underwriters clearly affects the liquidity of the bonds for which they have served as lead underwriters. Bonds issued by financial firms are not permanently more or less liquid than industrials but they do, however, have illiquidity spikes around the take-over of bear Sterns, the collapse of Lehman and the March 2009 rapid stock market decline. 62 Essay 2 2.8 Appendix: Robustness checks In this Appendix we discuss possible misspecification in our regression analysis. We test for endogeneity, show that our results are robust to the choice of benchmark riskfree rate, and show that results are robust to how we define the liquidity component. 2.8.1 Endogeneity There may be a two way causal relationship between contemporaneous measures of liquidity and credit risk and failing to account for such a relationship in regressions results in inconsistent OLS estimates. This simultaneity bias is not a concern in our regressions since liquidity measures lag our measure of credit spreads. Spreads are measured on the last day in each quarter while liquidity measures are based on transactions during the quarter, so liquidity measures are lagged in time relative to spreads. To test for potential endogeneity bias, we use a residual augmented two stage least squares t-test as in Davidson and MacKinnon (1993), equivalent to the Durbin-Wu-Hausman test. We do this for every marginal regression in Table 2.2, that is, test every liquidity variable separately. If the test is not significant the liquidity variable can be regarded as exogenous. As instrument we use bond age and therefore exclude it in the yield spread regressions13. Table 2.9 shows the R2’s for the first stage regressions and the t-statistic tests for endogeneity. Most R2’s are relatively high indicating that the control variables including the instrument are able to explain a large portion of the variation in the liquidity measures. Out of the 80 test statistics 80% are insignificant at a 10% level indicating that endogeneity is not a major concern. 2.8.2 Benchmark riskfree rate The size of the nondefault component in corporate bond spreads investigated by among others Huang and Huang (2003) and Longstaff, Mithal, and Neis (2005) depend strongly on the chosen riskfree rate. In Longstaff, Mithal, and Neis (2005) the difference is around 60 basis points. As Table 2.10 shows the estimated liquidity component when the Treasury rate is used as riskfree rate instead of the swap rate does not change much. The change in estimated liquidity is often less than one basis point and is for all rating categories less than 10 basis points. Therefore, our findings on the size of the liquidity premium in basis points are insensitive to the choice of benchmark 13Another potential instrument is amount issued. Since this variable is significant in most of the regressions in Table 2.4, omitting it from the regressions in the test creates a new endogeneity problem. The tests in this case would likely show an endogeneity problem even if it is not there, and if we use amount issued as instrument, this is indeed the case. 62 Essay 2 2.8 Appendix: Robustness checks In this Appendix we discuss possible misspecification in our regression analysis. We test for endogeneity, show that our results are robust to the choice of benchmark riskfree rate, and show that results are robust to how we define the liquidity component. 2.8.1 Endogeneity There may be a two way causal relationship between contemporaneous measures of liquidity and credit risk and failing to account for such a relationship in regressions results in inconsistent OLS estimates. This simultaneity bias is not a concern in our regressions since liquidity measures lag our measure of credit spreads. Spreads are measured on the last day in each quarter while liquidity measures are based on transactions during the quarter, so liquidity measures are lagged in time relative to spreads. To test for potential endogeneity bias, we use a residual augmented two stage least squares t-test as in Davidson and MacKinnon (1993), equivalent to the Durbin-Wu-Hausman test. We do this for every marginal regression in Table 2.2, that is, test every liquidity variable separately. If the test is not significant the liquidity variable can be regarded as exogenous. As instrument we use bond age and therefore exclude it in the yield spread regressions13. Table 2.9 shows the R2’s for the first stage regressions and the t-statistic tests for endogeneity. Most R2’s are relatively high indicating that the control variables including the instrument are able to explain a large portion of the variation in the liquidity measures. Out of the 80 test statistics 80% are insignificant at a 10% level indicating that endogeneity is not a major concern. 2.8.2 Benchmark riskfree rate The size of the nondefault component in corporate bond spreads investigated by among others Huang and Huang (2003) and Longstaff, Mithal, and Neis (2005) depend strongly on the chosen riskfree rate. In Longstaff, Mithal, and Neis (2005) the difference is around 60 basis points. As Table 2.10 shows the estimated liquidity component when the Treasury rate is used as riskfree rate instead of the swap rate does not change much. The change in estimated liquidity is often less than one basis point and is for all rating categories less than 10 basis points. Therefore, our findings on the size of the liquidity premium in basis points are insensitive to the choice of benchmark 13Another potential instrument is amount issued. Since this variable is significant in most of the regressions in Table 2.4, omitting it from the regressions in the test creates a new endogeneity problem. The tests in this case would likely show an endogeneity problem even if it is not there, and if we use amount issued as instrument, this is indeed the case. Corporate Bond Liquidity Before and After the Onset of the Subprime Crisis 63 Panel A: Endogeneity tests, pre-subprime (2004:Q4-2007:Q1) AAA AA A BBB spec Amihud −0.43 (33%) −1.00 (20%) 0.98 (18%) 1.31 (9%) 0.71 (34%) Roll 0.66 (47%) −0.98 (30%) 0.98 (32%) 1.16 (24%) −0.45 (25%) firm zero −0.25 (88%) 1.08 (34%) −0.83 (23%) −1.18 (25%) 0.27 (46%) bond zero −0.41 (83%) 1.04 (67%) −0.69 (68%) 0.85 (45%) −0.87 (61%) turnover −0.18 (19%) −1.13 (28%) 0.86 (15%) −1.05 (29%) 1.04 (39%) URC 0.51 (34%) −1.08 (18%) 0.95 (19%) 1.45 (23%) 0.13 (37%) Amihud risk 0.45 (19%) −1.09 (10%) 0.89 (11%) 1.43 (13%) 0.31 (31%) URC risk 0.46 (13%) −1.08 (12%) 0.90 (11%) 1.29 (14%) −0.03 (33%) Panel B: Endogeneity tests, post-subprime (2007:Q2-2009:Q2) AAA AA A BBB spec Amihud −5.03∗∗∗ (41%) −1.06 (31%) −0.20 (30%) −0.60 (27%) −2.82∗∗∗ (42%) Roll −5.24∗∗∗ (33%) −1.15 (15%) 0.51 (21%) 0.77 (16%) −2.89∗∗∗ (23%) firm zero 5.50∗∗∗ (87%) −1.12 (35%) −0.40 (24%) −0.82 (44%) −3.06∗∗∗ (58%) bond zero 6.40∗∗∗ (79%) 1.10 (73%) −0.21 (70%) −0.70 (68%) −3.26∗∗∗ (76%) turnover −6.17∗∗∗ (27%) −1.15 (16%) 0.32 (17%) 0.73 (20%) 2.91∗∗∗ (36%) URC −4.94∗∗∗ (50%) −0.84 (42%) −0.26 (49%) 0.77 (39%) −2.72∗∗∗ (63%) Amihud risk −5.07∗∗∗ (21%) −1.05 (22%) −0.36 (34%) −0.59 (45%) −2.69∗∗∗ (50%) URC risk −4.82∗∗∗ (39%) −0.74 (34%) 0.57 (48%) −0.75 (34%) −2.75∗∗∗ (55%) Table 2.9: Endogeneity tests. For each rating class Rand each liquidity variable L we test for potential endogeneity bias by using a Durbin-Wu-Hausman test. In total 56 tests are run (8 liquidity variables ×5 rating classes) pre- and post-subprime. This table shows for each test the t-statistics and R2for the first stage regression in parenthesis. The proxies are described in detail in Section 2.4 and are calculated quarterly from 2004 : Q4 to 2009 : Q2. Panel A shows the coefficients using data before the subprime crisis, while Panel B shows the coefficients using data after the onset of the subprime crisis. Significance at 10% level is marked ’*’, at 5% marked ’**’, and at 1% marked ’***’. 64 Essay 2 Panel A: Liquidity component in basis points, pre-subprime (2004Q4-2007:Q1) average 0-2y 2-5y 5-30y N 0-2y N 2-5y N 5-30y AAA 1.61.1 (0.8;1.4) 1.7 (1.2;2.1) 2.0 (1.4;2.5) 162 178 193 AA 1.71.1 (0.8;1.5) 1.8 (1.3;2.3) 2.3 (1.6;3.0) 704 667 498 A 2.81.7 (0.9;2.6) 2.9 (1.5;4.3) 3.8 (1.9;5.5) 1540 1346 1260 BBB 4.02.9 (1.4;4.4) 4.1 (1.9;6.2) 4.9 (2.3;7.3) 517 270 553 spec 57.845.2 (33.9;57.4) 44.1 (33.1;56.0) 84.2 (63.2;106.9) 270 324 480 Panel B: Liquidity component in basis points, post-subprime (2007:Q2-2009:Q2) average 0-2y 2-5y 5-30y N 0-2y N 2-5y N 5-30y AAA 1.00.5 (0.3;5.4) 0.8 (0.5;8.1) 1.7 (0.9;16.6) 110 149 155 AA 40.622.9 (11.5;35.2) 36.1 (18.2;55.5) 63.0 (31.8;96.8) 493 572 483 A 47.625.0 (12.9;37.6) 47.9 (24.7;72.1) 70.0 (36.1;105.4) 762 878 890 BBB 94.065.2 (36.0;97.4) 117.2 (64.8;175.1) 99.5 (55.0;148.6) 123 159 256 spec 189.9119.3 (79.4;154.9) 216.3 (144.0;280.9) 234.2 (156.0;304.2) 133 129 201 Table 2.10: Liquidity Component in basis points when the Treasury rate is used as riskfree rate. For each rating Rwe run the pooled regression spreadR it =αR+βRλit + credit risk controlsit +it where irefers to bond, tto time, and λit is our liquidity measure. The bond spread is measured with respect to the Treasury yield. Within each rating and maturity bucket (0-2y, 2-5y, and 5-30y) we sort increasingly all values of λit and find the median value λ50 and the 5% value λ5. The liquidity component in the bucket is defined as β(λ50−λ5). This table shows for all buckets the liquidity component with standard errors in parenthesis. Confidence bands are found by a wild cluster bootstrap. 64 Essay 2 Panel A: Liquidity component in basis points, pre-subprime (2004Q4-2007:Q1) average 0-2y 2-5y 5-30y N 0-2y N 2-5y N 5-30y AAA 1.6 1.1 (0.8;1.4) 1.7 (1.2;2.1) 2.0 (1.4;2.5) 162 178 193 AA 1.7 1.1 (0.8;1.5) 1.8 (1.3;2.3) 2.3 (1.6;3.0) 704 667 498 A 2.8 1.7 (0.9;2.6) 2.9 (1.5;4.3) 3.8 (1.9;5.5) 1540 1346 1260 BBB 4.0 2.9 (1.4;4.4) 4.1 (1.9;6.2) 4.9 (2.3;7.3) 517 270 553 spec 57.8 45.2 (33.9;57.4) 44.1 (33.1;56.0) 84.2 (63.2;106.9) 270 324 480 Panel B: Liquidity component in basis points, post-subprime (2007:Q2-2009:Q2) average 0-2y 2-5y 5-30y N 0-2y N 2-5y N 5-30y AAA 1.0 0.5 (0.3;5.4) 0.8 (0.5;8.1) 1.7 (0.9;16.6) 110 149 155 AA 40.6 22.9 (11.5;35.2) 36.1 (18.2;55.5) 63.0 (31.8;96.8) 493 572 483 A 47.6 25.0 (12.9;37.6) 47.9 (24.7;72.1) 70.0 (36.1;105.4) 762 878 890 BBB 94.0 65.2 (36.0;97.4) 117.2 (64.8;175.1) 99.5 (55.0;148.6) 123 159 256 spec 189.9 119.3 (79.4;154.9) 216.3 (144.0;280.9) 234.2 (156.0;304.2) 133 129 201 Table 2.10: Liquidity Component in basis points when the Treasury rate is used as riskfree rate. For each rating Rwe run the pooled regression spreadR it =αR+βRλit + credit risk controlsit +it where irefers to bond, tto time, and λit is our liquidity measure. The bond spread is measured with respect to the Treasury yield. Within each rating and maturity bucket (0-2y, 2-5y, and 5-30y) we sort increasingly all values of λit and find the median value λ50 and the 5% value λ5. The liquidity component in the bucket is defined as β(λ50−λ5). This table shows for all buckets the liquidity component with standard errors in parenthesis. Confidence bands are found by a wild cluster bootstrap. Corporate Bond Liquidity Before and After the Onset of the Subprime Crisis 65 (while our findings on the fraction out of the total spread of course depend on the benchmark riskfree rate). 2.8.3 Alternative definition of liquidity component The liquidity component is calculated as the the median minus 5% quantile of the liquidity score and has the natural interpretation as the liquidity premium of an average bond in the corporate bond market relative to a very liquid bond. To check that our main results are robust to the definition of the liquidity component, Table 2.11 shows the liquidity component when it is defined as the 75% quantile minus 5% quantile. The component in this table can be interpreted as that of an illiquid bond relative to a very liquid bond. Table 2.11 shows that the liquidity component is larger for an illiquid bond compared to an average bond (which by definition must the case). Also, Table 2.11 shows that the main results of the paper are unchanged: liquidity premia are increasing in maturity, the liquidity premium is higher post-subprime compared to pre-subprime, and the liquidity premium for investment grade bonds is small pre-subprime. 72 Essay 3 3.2.2 Stock index rebalancing returns and explanations. Whereas all studies agree on the changes in trading volume around revisions, not all agree on how the abnormal returns react and specifically what happens in the longer run. Shleifer (1986) and Chen, Noronha, and Singal (2004) find no significant price changes around index inclusion or exclusion before the introduction of the announcement service in September 1976. Over the period from September 1976 until September 1989 Chen, Noronha, and Singal (2004) report an average abnormal return of 3.7% on the first trading day after the announcement for stocks included into the S&P 500, which is at the level of Shleifer (1986), Harris and Gurel (1986), Jain (1987), Dhillon and Johnson (1991) and Beneish and Whaley (1996) for various sub-periods. Harris and Gurel (1986) find a complete reversal of the abnormal return over the next 10-14 trading days, whereas the remaining studies only find a partial reversal of the abnormal return. For the newer period from October 1989 until 2000, where the announcement date and the change date for the S&P500 are separated by 5 days on average, Chen, Noronha, and Singal (2004) report a 5.4% abnormal return on the announcement day for stocks being included and a further 3.5% increase from the announcement day to the effectuation day. After the effectuation day, the cumulative abnormal return is again partially reversed over time. This return pattern is also found by Beneish and Whaley (1996), Lynch and Mendenhall (1997), Blume and Edelen (2002), Denis, McConnell, and Ovtchinnikov (2003), Hegde and Mc- Dermott (2003), Elliot and Warr (2003), Cai (2007), Petajisto (2009), Elliott, Ness, Walker, and Warr (2006) for S&P 500 inclusions for various periods after 1989. The same pattern is also documented by Madhavan (2003) and Cari˜no and Pritamani (2007) for inclusions to the Russell-indices, for revisions of the Nikkei225 (Greenwood 2005), for revisions of the Toronto300 index (Kaul, Mehrotra, and Morck 2000) and for inclusions to the FTSE100 (Mase 2007). Even though index inclusions have received the most attention in the literature, index exclusions show roughly the same pattern with an abnormal (negative) return on the announcement date and a further decrease up to the change date followed by a reversal (Lynch and Mendenhall (1997), Blume and Edelen (2002) and Chen, Noronha, and Singal (2004)). Elliott, Ness, Walker, and Warr (2006) list five competing explanations for the abnormal return patterns around stock index revisions - price pressure, downward-sloping demand curves, improved liquidity, improved operating performance, and increased investor awareness. The explanations are not necessarily exclusive. According to the price pressure hypothesis the abnormal return at inclusion stems from short-run liquidity constraints temporarily driving prices above the fundamental value in order to compensate liquidity providers. Empirically only Harris and Gurel (1986) find full support of the price pressure hypothesis since they find a complete reversal of the abnormal returns. As 72 Essay 3 3.2.2 Stock index rebalancing returns and explanations. Whereas all studies agree on the changes in trading volume around revisions, not all agree on how the abnormal returns react and specifically what happens in the longer run. Shleifer (1986) and Chen, Noronha, and Singal (2004) find no significant price changes around index inclusion or exclusion before the introduction of the announcement service in September 1976. Over the period from September 1976 until September 1989 Chen, Noronha, and Singal (2004) report an average abnormal return of 3.7% on the first trading day after the announcement for stocks included into the S&P 500, which is at the level of Shleifer (1986), Harris and Gurel (1986), Jain (1987), Dhillon and Johnson (1991) and Beneish and Whaley (1996) for various sub-periods. Harris and Gurel (1986) find a complete reversal of the abnormal return over the next 10-14 trading days, whereas the remaining studies only find a partial reversal of the abnormal return. For the newer period from October 1989 until 2000, where the announcement date and the change date for the S&P500 are separated by 5 days on average, Chen, Noronha, and Singal (2004) report a 5.4% abnormal return on the announcement day for stocks being included and a further 3.5% increase from the announcement day to the effectuation day. After the effectuation day, the cumulative abnormal return is again partially reversed over time. This return pattern is also found by Beneish and Whaley (1996), Lynch and Mendenhall (1997), Blume and Edelen (2002), Denis, McConnell, and Ovtchinnikov (2003), Hegde and Mc- Dermott (2003), Elliot and Warr (2003), Cai (2007), Petajisto (2009), Elliott, Ness, Walker, and Warr (2006) for S&P 500 inclusions for various periods after 1989. The same pattern is also documented by Madhavan (2003) and Cari˜no and Pritamani (2007) for inclusions to the Russell-indices, for revisions of the Nikkei225 (Greenwood 2005), for revisions of the Toronto300 index (Kaul, Mehrotra, and Morck 2000) and for inclusions to the FTSE100 (Mase 2007). Even though index inclusions have received the most attention in the literature, index exclusions show roughly the same pattern with an abnormal (negative) return on the announcement date and a further decrease up to the change date followed by a reversal (Lynch and Mendenhall (1997), Blume and Edelen (2002) and Chen, Noronha, and Singal (2004)). Elliott, Ness, Walker, and Warr (2006) list five competing explanations for the abnormal return patterns around stock index revisions - price pressure, downward-sloping demand curves, improved liquidity, improved operating performance, and increased investor awareness. The explanations are not necessarily exclusive. According to the price pressure hypothesis the abnormal return at inclusion stems from short-run liquidity constraints temporarily driving prices above the fundamental value in order to compensate liquidity providers. Empirically only Harris and Gurel (1986) find full support of the price pressure hypothesis since they find a complete reversal of the abnormal returns. As Index Driven Price Pressure for Corporate Bonds 73 listed above several other studies find that part of the abnormal return is reversed in the days subsequent to the revision, but that a part of the price increase is permanent, at least for the newer periods. This suggests that price pressure at most explains part of the return pattern. Scholes (1972), Kraus and Stoll (1972) and Keim and Madhavan (1998) all study large block sales of stocks and do also find evidence of short-run price pressure in the stock market. However, contrary to the stock market, large trades in the corporate bond market are not expected to move prices apriori. Edwards, Harris, and Piwowar (2007) and Feldh¨utter (2009) present evidence that transaction costs are a decreasing function of trading volume for corporate bonds and that larger trades are executed at lower bid-ask spreads. This happens because larger trades are carried out by sophisticated investors with high bargaining power in an over-the-counter market as modeled in Duffie, Gˆarleanu, and Pedersen (2007). Even though corporate bond index trackers are sophisticated investors they may still be subject to price pressure since they are constrained by their investment strategy to minimize tracking error, which could reduce their bargaining power. This leaves the question of index driven price pressure for corporate bonds as an empirical question. In classical asset pricing theory the demand curves for stocks are horizontal at the risk adjusted fundamental value, which implies that a demand or a supply shift would not affect the price of the stock. To the extent that index inclusions are information free events, they provide an excellent test of the slope of the demand curve for stocks. When stocks are included into an index, demand for the stock will increase as a consequence of index tracking. Shleifer (1986) is the first to suggest this test of downward sloping demand curves and as listed earlier a range of studies do find evidence that part of the price increase following index inclusions is permanent. Especially Greenwood (2005)’s study of the Nikkei225 and Kaul, Mehrotra, and Morck (2000) of the Toronto300 present strong evidence in support of downward sloping demand curves for stocks. In both studies index weights were changed due to an index redefinition which resulted in price changes for stocks still in the index, but now with a different weight and hence a different demand from index trackers. Also Wurgler and Zhuravskaya (2002) find support of downward sloping demand curves for stocks studying S&P 500 inclusions. They find that index arbitrageurs are unable to find close substitutes for the stocks included which leaves the arbitrageurs with unhedged risk that make them unwilling to trade at the pre-inclusion price of the stock. The improved liquidity hypothesis states that when stocks are included into an index it improves liquidity and reduces asymmetric information permanently. This results in lower transaction costs and in the end a permanently higher price as in the model of Amihud and Mendelson (1986). The liquidity may be improved upon inclusion because the number of institutional investors increases which increases the level of monitoring of the firm thereby lowering the level of asymmetric information. Beneish and Whaley 74 Essay 3 (1996) find a temporary reduction in bid-ask spreads, Hegde and McDermott (2003) find the same but with a permanent reduction and Madhavan (2003) ascribes the increased trading volume following inclusion into the Russell index family as improved liquidity. It seems unlikely that a bond index inclusion should improve liquidity. Inclusion into major bond indices are based on mechanical rules like that of the Russell indices but the indices are not limited to contain a certain number of bonds. Most of the bonds included are issued by large firms which already have a number of bonds outstanding, so there is no reason to think that the issuance of yet another bond from the same firm would lower the asymmetric information level. If more bonds from the same firm get included into the index it might even lower the liquidity of the bonds, since the issuance of many bonds might indicate that the firm is in financial trouble, which tend to lower the bond liquidity (see e.g. Dick-Nielsen, Feldh¨utter, and Lando (2009)). Investors may not be aware of all stocks in the market which would make them overinvested in the stocks of which they are aware (Merton 1987). In order to hold less known stocks investors require a return premium (a shadow cost). Inclusion into a stock index is likely to increase investor awareness of the stock and it thus reduces the shadow costs. For stocks included into the index the empirical implications of this hypothesis and of the downward sloping demand curve hypothesis are identical. However, index exclusions do allow us to empirically separate the investor awareness hypothesis and the downward sloping demand curve hypothesis. Chen, Noronha, and Singal (2004) argue that investor awareness should not decrease after an index exclusion which should result in a smaller price change for stocks removed from the index. Empirically, Chen, Noronha, and Singal (2004) do find an asymmetric price response in favor of the investor awareness hypothesis. For the large majority of corporate bond indices, inclusion is closely related to the issuance of the bond, since the bonds are included close after issuance. This means that index inclusion is not a sudden change in the status of the bond and at the same time index membership is not limited to a selected number as it is for most stocks indices. All together, it is unlikely that investor awareness should change upon a bond index inclusion. All of the above mentioned hypotheses assume that index inclusion is an information free event in itself. This is a reasonable assumption since index revision rules for the Nikkei225, the Toronto300 and the Russell index family are all based on mechanical rules and Standard and Poor’s states that inclusion into the S&P 500 is only based upon publicly available information. However, Standard and Poor’s do make extensive analysis of different candidate firms in order to insure a low turnover in index membership. Being a rating company which specializes in financial analysis of firms, Standard and Poor’s may unwillingly or unknowingly use the public information in a way superior to other investors and in the end convey information about the firm when they select it for inclusion. Denis, McConnell, and Ovtchinnikov (2003) 74 Essay 3 (1996) find a temporary reduction in bid-ask spreads, Hegde and McDermott (2003) find the same but with a permanent reduction and Madhavan (2003) ascribes the increased trading volume following inclusion into the Russell index family as improved liquidity. It seems unlikely that a bond index inclusion should improve liquidity. Inclusion into major bond indices are based on mechanical rules like that of the Russell indices but the indices are not limited to contain a certain number of bonds. Most of the bonds included are issued by large firms which already have a number of bonds outstanding, so there is no reason to think that the issuance of yet another bond from the same firm would lower the asymmetric information level. If more bonds from the same firm get included into the index it might even lower the liquidity of the bonds, since the issuance of many bonds might indicate that the firm is in financial trouble, which tend to lower the bond liquidity (see e.g. Dick-Nielsen, Feldh¨utter, and Lando (2009)). Investors may not be aware of all stocks in the market which would make them overinvested in the stocks of which they are aware (Merton 1987). In order to hold less known stocks investors require a return premium (a shadow cost). Inclusion into a stock index is likely to increase investor awareness of the stock and it thus reduces the shadow costs. For stocks included into the index the empirical implications of this hypothesis and of the downward sloping demand curve hypothesis are identical. However, index exclusions do allow us to empirically separate the investor awareness hypothesis and the downward sloping demand curve hypothesis. Chen, Noronha, and Singal (2004) argue that investor awareness should not decrease after an index exclusion which should result in a smaller price change for stocks removed from the index. Empirically, Chen, Noronha, and Singal (2004) do find an asymmetric price response in favor of the investor awareness hypothesis. For the large majority of corporate bond indices, inclusion is closely related to the issuance of the bond, since the bonds are included close after issuance. This means that index inclusion is not a sudden change in the status of the bond and at the same time index membership is not limited to a selected number as it is for most stocks indices. All together, it is unlikely that investor awareness should change upon a bond index inclusion. All of the above mentioned hypotheses assume that index inclusion is an information free event in itself. This is a reasonable assumption since index revision rules for the Nikkei225, the Toronto300 and the Russell index family are all based on mechanical rules and Standard and Poor’s states that inclusion into the S&P 500 is only based upon publicly available information. However, Standard and Poor’s do make extensive analysis of different candidate firms in order to insure a low turnover in index membership. Being a rating company which specializes in financial analysis of firms, Standard and Poor’s may unwillingly or unknowingly use the public information in a way superior to other investors and in the end convey information about the firm when they select it for inclusion. Denis, McConnell, and Ovtchinnikov (2003) Index Driven Price Pressure for Corporate Bonds 75 present evidence that analyst earnings forecasts rise for firms, whose stocks are included into the S&P 500 and that these firms also realize the higher earnings. The improved operating performance hypothesis states that this may not just be a result of new information about the firm. The improved operating performance is a result of closer scrutiny of the management, which in the end improves the performance (as also suggested by Shleifer 1986). Both Denis, McConnell, and Ovtchinnikov (2003) and Dhillon and Johnson (1991) find empirically that upon inclusion into the S&P 500 the firms’ stocks, as well as it’s bonds, increase in value. The increased bond values indicate that the index inclusion is linked to the fundamental value of the firm and not only isolated frictions relating to stocks as an asset class. As with the Nikkei225, the Toronto300 and the Russell index family, inclusion to or exclusion from corporate bond indices are non-informational events and like the investor awareness hypothesis it seems unlikely that inclusion into the bond index should lead to improved operating performance, since inclusion does not make the firm move visible. 3.2.3 Demand Curves for Corporate Bonds This paper defines index trackers as investors that seek to replicate the return of an index while they seek to minimize their tracking error. The definition captures the fact that index trackers do not seek or wish to outperform the index as documented in Blume and Edelen (2002) for S&P 500 index trackers. Blume and Edelen (2002) show that index trackers actually sacrifice trading gain to reduce the tracking error. No papers have so far looked at the impact of corporate bond index tracking on the individual bonds being included to or excluded from the index. However, two papers, Newman and Rierson (2004) and Chen, Lookman, Sch¨urhoff, and Seppi (2009), present evidence of price pressure or short-term downward sloping demand curves for corporate bonds around events closely related to corporate bond index tracking. Between October 1999 and July 2001 the European telecom-sector issued bonds which increased the overall amount outstanding by 300% for the sector. The new issues raised funds to the sector that could support bids for government auctions on cellular bandwidth licenses. Newman and Rierson (2004) study the price reaction on the already issued bonds in the sector around the new issuances. They find price pressure on the already issued bonds in the sector that temporarily decreases the price. One explanation for the price pressure could be that the new issues would enter into a corporate bond index. This new entry would then decrease the weight of the other bonds in the index forcing index trackers with limited funds to sell out of the bonds already in the index in order to buy the new issues and hence still track the index. This explanation is in line with what happened in Kaul, Mehrotra, and Morck (2000) and Greenwood (2005). However, whereas Kaul, Mehrotra, and Morck (2000) and Greenwood (2005) docu- 76 Essay 3 ment the effect of the reweighting in the index to happen on the same day that the reweightening was effectuated, Newman and Rierson (2004) show that the price pressure surrounding the new issuances in the telecom sector was centered around the issuance date of the bonds. As we will describe later on, newly issued bonds will not enter the major indices until the last day of the month in which they are issued. Nothing in Newman and Rierson (2004) suggests that the actual index inclusion day itself should be special. This indicates that the price pressure they find is not driven by index trackers, since we have no reason to think that index trackers would be active around the issuance date (see Blume and Edelen 2002). Chen, Lookman, Sch¨urhoff, and Seppi (2009) look at the effect of a rating rule change in the Lehman/Barclay index family. In January 2005, Lehman announced that it would change its practise on index rating definitions (only relevant for split-rated bonds). Before the change, a corporate bond was considered investment grade if it had an investment grade rating by both S&P and Moody’s. After the change, a corporate bond was considered investment grade if it had an investment grade rating by at least two out the three major rating agencies S&P, Moody’s or Fitch. Lehman used the index rating to determine bond membership of its indices, e.g. membership of the major US Aggregate Bond Index and of the Corporate Bond index required an investment grade index rating. The index rating rule change had an effect on bonds that were mechanically upgraded into investment grade because of the rule change and on bonds that now had a more save investment grade rating. Chen, Lookman, Sch¨urhoff, and Seppi (2009) show that the mechanically upgraded bonds experienced a cumulative abnormal return of 200 bps in the two weeks following the announcement. The price pressure was transitory and disappeared again after 20 to 30 trading days. While the index rating rule change without doubt affected the portfolio construction of index trackers it is unlikely that the 200 bps abnormal return was driven by index trackers. The actual rule change was not effectuated until the end of June 2005, long after the announcement return effect in January 2005, had disappeared and Chen, Lookman, Sch¨urhoff, and Seppi (2009) find no abnormal return around the actual index change. As stated in Chen, Lookman, Sch¨urhoff, and Seppi (2009) the abnormal return following the announcement is more likely to have been driven by institutional investors. Institutional investors are in many cases constrained in their portfolio choice to investment grade assets and in the case of split ratings industry practise have been to follow the Lehman index rating rule. 3.3 The Lehman/Barclay Corporate Bond Index The largest bond index funds (i.e. Vanguard, Schwab and Fidelity Total Bond Market Index Funds) are tracking the performance of the Barclays 76 Essay 3 ment the effect of the reweighting in the index to happen on the same day that the reweightening was effectuated, Newman and Rierson (2004) show that the price pressure surrounding the new issuances in the telecom sector was centered around the issuance date of the bonds. As we will describe later on, newly issued bonds will not enter the major indices until the last day of the month in which they are issued. Nothing in Newman and Rierson (2004) suggests that the actual index inclusion day itself should be special. This indicates that the price pressure they find is not driven by index trackers, since we have no reason to think that index trackers would be active around the issuance date (see Blume and Edelen 2002). Chen, Lookman, Sch¨urhoff, and Seppi (2009) look at the effect of a rating rule change in the Lehman/Barclay index family. In January 2005, Lehman announced that it would change its practise on index rating definitions (only relevant for split-rated bonds). Before the change, a corporate bond was considered investment grade if it had an investment grade rating by both S&P and Moody’s. After the change, a corporate bond was considered investment grade if it had an investment grade rating by at least two out the three major rating agencies S&P, Moody’s or Fitch. Lehman used the index rating to determine bond membership of its indices, e.g. membership of the major US Aggregate Bond Index and of the Corporate Bond index required an investment grade index rating. The index rating rule change had an effect on bonds that were mechanically upgraded into investment grade because of the rule change and on bonds that now had a more save investment grade rating. Chen, Lookman, Sch¨urhoff, and Seppi (2009) show that the mechanically upgraded bonds experienced a cumulative abnormal return of 200 bps in the two weeks following the announcement. The price pressure was transitory and disappeared again after 20 to 30 trading days. While the index rating rule change without doubt affected the portfolio construction of index trackers it is unlikely that the 200 bps abnormal return was driven by index trackers. The actual rule change was not effectuated until the end of June 2005, long after the announcement return effect in January 2005, had disappeared and Chen, Lookman, Sch¨urhoff, and Seppi (2009) find no abnormal return around the actual index change. As stated in Chen, Lookman, Sch¨urhoff, and Seppi (2009) the abnormal return following the announcement is more likely to have been driven by institutional investors. Institutional investors are in many cases constrained in their portfolio choice to investment grade assets and in the case of split ratings industry practise have been to follow the Lehman index rating rule. 3.3 The Lehman/Barclay Corporate Bond Index The largest bond index funds (i.e. Vanguard, Schwab and Fidelity Total Bond Market Index Funds) are tracking the performance of the Barclays Index Driven Price Pressure for Corporate Bonds 77 Capital U.S Aggregated Bond Index (formerly Lehman U.S Aggregated Bond Index). The index is a broad mixture of Government, Agency, Corporate and Mortgage Backed bonds all with an investment grade rating. We focus on the corporate bond part of the index which by itself is known as the Barclays Capital Corporate Bond Index. As of July 1, 2005, the index consists of all corporate bond issues which have an investment grade rating by at least two of the three major rating agencies. The issue size must be $250 millions or above and time to maturity must be above 1 year. Once a bond is issued and if it complies with all the rules it is included in the index at the next rebalancing date. If the bond at some point no longer fulfills all the criteria, it is excluded from the index at the next rebalancing date. Reason N Average amt. ($1,000) Average Duration Average Coupon Panel A: Inclusions New issue 4007 663,834 7.4 5.6 Upgrade 554 587,277 5.8 7.0 Other 200 676,832 6.7 5.6 Panel B: Exclusions Maturity<1 1817 538,114 0.92 6.0 Called 244 332,439 0.82 7.5 Downgrade 917 587,548 5.0 6.9 Other 1685 247,617 5.8 6.7 Table 3.1: Corporate Bond Index Inclusions and Exclusions Statistics. This table shows statistics for inclusions and exclusions from the Lehman/Barclay Corporate Bond Index. The same bonds enters and exits several Lehman/Barclay indices at the same time. The statistics are accumulated over the period from July 2002 to August 2009. Average amt. is the average amount outstanding in $1,000 at the time of the index revision. The index is rebalanced on the last trading day of each month at 3:00 PM EST. Inclusions to the index mainly happen because a bond is newly issued or because it is upgraded. Exclusions from the index happen mainly when a bond is downgraded to speculative grade or when the time to maturity drops below 1 year. From July 2002 to August 2009, 4,007 bonds were included as newly issued. In table 3.1 we can see that they had an average duration of 7.4 years, which is higher than the total index duration of around 5 years. 554 bonds were upgraded into the index. These bonds have a lower duration and a higher coupon since most of them were originally issued as speculative grade before they were later upgraded. Effective July 2005 Lehman changed the index rules so that a split rated bond was no longer seen as a speculative grade, but had the middle rating from Standard and Poor’s, Moody’s and Fitch (the present rating rule). The specific event involving the rule change 78 Essay 3 is researched in Chen, Lookman, Sch¨urhoff, and Seppi (2009). The other bonds that were included into the index were included for various reasons e.g. they changed status from non-public to publicly traded. 1,817 bonds were excluded from the index because the maturity dropped below one year. A smaller fraction of the bonds were called and 917 bonds left the index because of a downgrade. The average coupon for the downgraded bonds was higher than for the maturing bonds, reflecting that the downgraded bonds were issued with a lower rating in the first place. A large fraction of the bonds were excluded for other reasons. The average amount outstanding for these bonds is below $ 250 million, which is the lower limit for being in the index today. Most of these bonds were excluded exactly because the amount outstanding dropped below the limit, either because the amount outstanding changed or because the index limit was raised. Since the inception of the index the minimum amount outstanding limit has been raised several times. Typically, each mutual fund family (Vanguard, Schwab, Fidelity etc.) has more funds tracking different parts of the total aggregated index. Barclays keeps short-term, intermediate and long-term indices, which have similar rules for membership as the aggregate index. The only difference is the maturity of the bonds in the indices. None of the bond index funds track the indices by replication, which is the most common way for stock index funds to track an index (Blume and Edelen 2002). Instead they are sampling the index (see e.g. Vanguard (2009) or Schwab (2009)). Typically the fund chooses a selection of the bonds currently in the index and designs a portfolio of these that match the index with respect to duration, cash flow, quality and callability. As of December 31, 2008 Vanguard Total Bond Market Index Fund held 3,731 different bonds out of the 9,168 bonds present in the index at that time. The corporate bond index has around 3,400 bonds in the index right now (see figure 3.1). Since index inception the rules for index membership have been tightened so that smaller issues have been excluded. It happened once in October, 2003 and again in July, 2004. The motivation has been to keep the index from getting too large, but as can be seen in figure 3.1 the index has still increased immensely both in the number of bonds and the amount outstanding. In order to reduce transaction costs from rebalancing, the index funds also invest outside the index. Typically 80% of the assets in the funds are invested in bonds currently in the index. The remaining 20% are invested outside the index for example in non-public bonds, lower rated bonds, non-corporate bonds or derivatives such as futures, options and swaps. The criteria for investing outside the index are rather loose and in that way it is not possible to know exactly which assets the funds have on their balance sheets. Still, even though the bond funds track by sampling they obtain a rather low tracking error. The average yearly absolute return tracking error for the shares of Vanguard Total Bond Market Fund over 1995- 2009 is 23.5 bps. The tracking error record is dominated by 2002 which had 78 Essay 3 is researched in Chen, Lookman, Sch¨urhoff, and Seppi (2009). The other bonds that were included into the index were included for various reasons e.g. they changed status from non-public to publicly traded. 1,817 bonds were excluded from the index because the maturity dropped below one year. A smaller fraction of the bonds were called and 917 bonds left the index because of a downgrade. The average coupon for the downgraded bonds was higher than for the maturing bonds, reflecting that the downgraded bonds were issued with a lower rating in the first place. A large fraction of the bonds were excluded for other reasons. The average amount outstanding for these bonds is below $ 250 million, which is the lower limit for being in the index today. Most of these bonds were excluded exactly because the amount outstanding dropped below the limit, either because the amount outstanding changed or because the index limit was raised. Since the inception of the index the minimum amount outstanding limit has been raised several times. Typically, each mutual fund family (Vanguard, Schwab, Fidelity etc.) has more funds tracking different parts of the total aggregated index. Barclays keeps short-term, intermediate and long-term indices, which have similar rules for membership as the aggregate index. The only difference is the maturity of the bonds in the indices. None of the bond index funds track the indices by replication, which is the most common way for stock index funds to track an index (Blume and Edelen 2002). Instead they are sampling the index (see e.g. Vanguard (2009) or Schwab (2009)). Typically the fund chooses a selection of the bonds currently in the index and designs a portfolio of these that match the index with respect to duration, cash flow, quality and callability. As of December 31, 2008 Vanguard Total Bond Market Index Fund held 3,731 different bonds out of the 9,168 bonds present in the index at that time. The corporate bond index has around 3,400 bonds in the index right now (see figure 3.1). Since index inception the rules for index membership have been tightened so that smaller issues have been excluded. It happened once in October, 2003 and again in July, 2004. The motivation has been to keep the index from getting too large, but as can be seen in figure 3.1 the index has still increased immensely both in the number of bonds and the amount outstanding. In order to reduce transaction costs from rebalancing, the index funds also invest outside the index. Typically 80% of the assets in the funds are invested in bonds currently in the index. The remaining 20% are invested outside the index for example in non-public bonds, lower rated bonds, non-corporate bonds or derivatives such as futures, options and swaps. The criteria for investing outside the index are rather loose and in that way it is not possible to know exactly which assets the funds have on their balance sheets. Still, even though the bond funds track by sampling they obtain a rather low tracking error. The average yearly absolute return tracking error for the shares of Vanguard Total Bond Market Fund over 1995- 2009 is 23.5 bps. The tracking error record is dominated by 2002 which had Index Driven Price Pressure for Corporate Bonds 79 an exceptionally bad tracking error of 2% under the actual return of the Lehman/Barclay index. Without that year the mean absolute error is 11.0 bps. Compared to Fidelitys tracking error of 2.3 bps for tracking the S&P 500, the error for the Vanguard bond fund is much higher. Still, the tracking error record for the Vanguard fund suggests that the goal of the fund is to track the index very closely. In only one of the years between 1995-2009, have the Vanguard fund shares had a higher return than the index it tracks, which is probably due to the transaction costs from the sampling strategy. 2600 2800 3000 3200 3400 3600 Month # Bonds in the Index 2001M4 2003M10 2006M6 2009M3 1400 1600 1800 2000 2200 2400 Month Total Index Amt. (billion $) 2001M4 2003M10 2006M6 2009M3 Figure 3.1: Corporate Bond Index Composition. The left graph shows the evolution of the number of bonds in the index. During the period the index membership rules where changed twice. First in October, 2003 and again in July, 2004. The left graph shows the total amount of debt outstanding in the index. We obtain information about the index rules and composition of the Lehman/Barclays Capital Corporate Bond Index from their website 2. From the same place we obtain return series for different benchmark indices. Finally we use the TRACE database from WRDS for transaction level information about the individual bonds in the index. Before using the TRACE data for calculating returns we filter the data as in Dick-Nielsen (2009) in order to avoid biases from the way reporting errors are recorded and cumulating in the TRACE system. 2See index rules at www.lehman.com/fi/indices/pdf/US Corporate Index.pdf and index dynamics at https://live.lehman.com/LL/lehmanlive 80 Essay 3 3.4 Measuring Abnormal Corporate Bond Returns Calculating daily corporate bond returns provide quite a challenge compared to stock market returns. First, an average corporate bond only trades once in a couple of months making it hard to get time series data. Second, trades are usually clustered and the same bond might trade several times on the same day at different prices (see e.g. Feldh¨utter (2009)) making it difficult to get a unique daily price. Bessembinder, Kahle, Maxwell, and Xu (2009) show in an extensive simulation analysis of corporate bond event studies that tests relying on transactions data have far more power than tests using daily data quotes from e.g. DataStream. This finding is in line with Sarig and Warga (1989) and Dick-Nielsen, Feldh¨utter, and Lando (2009) who show that quoted prices often have little connection to actual transaction prices. For each bond entering or leaving the index we calculate a daily price on days with at least one trade above $100,000 in nominal value (100 bonds) as the trading volume weighted average price of all trades on that day above $100,000. This method is suggested by Bessembinder, Kahle, Maxwell, and Xu (2009) as the best way to calculate daily prices. Bonds might not have transactions on all days surrounding our index events. To circumvent this problem we follow the approach in Cai, Helwege, and Warga (2007) and calculate event returns as the logarithmic difference between the price on the closest day prior to the event and the closet day after or including the event date: rb,a i= log pb i−log pa i where iis a bond identifier, bis the event day or the day closest to the event day, but still after the event and ais the day closest to the event date, but still before the event. If we define day number -1 as the day before the event, day 0 as the event day and day 1 as the day after the event, then a= max(day number with a return) <0 b= min(day number with a return) ≥0 We extend the method to cumulative returns by restricting the window in the following way a= max(day number with a return) < c b= min(day number with a return) ≥d where c≤0 and d≥0 so that the window spans a period. We restrict the sample to returns calculated using daily prices not more than five trading days away from the event date or restriction points (c−a < 6 and b−d < 6). In comparison Ambrose, Cai, and Helwege (2009) use a return window as 80 Essay 3 3.4 Measuring Abnormal Corporate Bond Returns Calculating daily corporate bond returns provide quite a challenge compared to stock market returns. First, an average corporate bond only trades once in a couple of months making it hard to get time series data. Second, trades are usually clustered and the same bond might trade several times on the same day at different prices (see e.g. Feldh¨utter (2009)) making it difficult to get a unique daily price. Bessembinder, Kahle, Maxwell, and Xu (2009) show in an extensive simulation analysis of corporate bond event studies that tests relying on transactions data have far more power than tests using daily data quotes from e.g. DataStream. This finding is in line with Sarig and Warga (1989) and Dick-Nielsen, Feldh¨utter, and Lando (2009) who show that quoted prices often have little connection to actual transaction prices. For each bond entering or leaving the index we calculate a daily price on days with at least one trade above $100,000 in nominal value (100 bonds) as the trading volume weighted average price of all trades on that day above $100,000. This method is suggested by Bessembinder, Kahle, Maxwell, and Xu (2009) as the best way to calculate daily prices. Bonds might not have transactions on all days surrounding our index events. To circumvent this problem we follow the approach in Cai, Helwege, and Warga (2007) and calculate event returns as the logarithmic difference between the price on the closest day prior to the event and the closet day after or including the event date: rb,a i= log pb i−log pa i where iis a bond identifier, bis the event day or the day closest to the event day, but still after the event and ais the day closest to the event date, but still before the event. If we define day number -1 as the day before the event, day 0 as the event day and day 1 as the day after the event, then a= max(day number with a return) <0 b= min(day number with a return) ≥0 We extend the method to cumulative returns by restricting the window in the following way a= max(day number with a return) < c b= min(day number with a return) ≥d where c≤0 and d≥0 so that the window spans a period. We restrict the sample to returns calculated using daily prices not more than five trading days away from the event date or restriction points (c−a < 6 and b−d < 6). In comparison Ambrose, Cai, and Helwege (2009) use a return window as Index Driven Price Pressure for Corporate Bonds 81 large as min(a)=−100 and max(b) = 99. However, a window that long could have a significant influence on the variance of the return, which is why we choose a shorter window. There could still be a problem with the variance of the return, but Ambrose, Cai, and Helwege (2009) and Cai, Helwege, and Warga (2007) run various robustness checks and show that there are no problems with small windows like the one we use. Chen, Lookman, Sch¨urhoff, and Seppi (2009) propose a different method to circumvent the infrequent trading problem. They calculate cumulative returns from a given pre-event day to a range of later dates, for each date disregarding bonds that did not trade on that particular date. Then they use a value-weighted average of all returns on each of the later days to get a portfolio return for that day. In this way they get a portfolio time-series, but with a possibly different portfolio on each time series date. However, this method still requires trading on the post-event dates, which is the major problem in our event study. Bessembinder, Kahle, Maxwell, and Xu (2009) and Chen, Lookman, Sch¨urhoff, and Seppi (2009) argue that the best way to calculate abnormal returns is to use a benchmark portfolio of bonds similar to the individual bond in the event study (in line with Barber and Lyon (1997) for stock returns). We follow these studies and in each of the four following event types we explain which benchmark portfolio we use. 3.5 Maturity Less Than 1 year In contrast to stocks, bonds have a maximum lifetime in an index, since they are excluded 1 year prior to maturity. Corporate bonds that fall below 1 year to maturity are excluded from the corporate bond index at the last trading day of the month. Exclusion because of low maturity is by far the most common reason for a bond to leave the index (as seen in table 3.1). Figure 3.2 shows the aggregate trading volume around the index rebalancing date. The event date is the last day of the month, negative days are before the event date and positive days are after the event date. For each bond excluded because of low maturity the total trading volume (without the sign of the trade direction) is added for each day around the event across bonds and calendar dates. Similar to stock index trackers’ timing of their trades (Blume and Edelen (2002)), the trading activity is highest at the date of the exclusion. Because bond funds only sample an index, not all bonds that are excluded from the index exhibits abnormal trading activity. The average turnover on the event day is 1.2% of the volume outstanding for the 1/3 most traded bonds on the event day (as stated in section 3.3 bond index funds typically only hold 1/3 of the bonds in an index). It is clear from figure 3.2 that the event date is the exclusion date and not the day where the maturity actually falls below 1 year. Had the latter been the case, then the trading activity should has been higher 10 to 21 days before the exclusion day, which 88 Essay 3 0 10 20 30 40 50 10 15 20 25 30 Days around event Aggregate Trading Volume ($ billion) Figure 3.5: Total trading volume for newly issued bonds included into the index. Trading volume for each bond is added on a daily basis (without trading direction) and then the daily volume is added across bonds according to distance from the event. 2014 bonds trade on the event date. The average turnover for the 1/3 most traded bonds is 2.8%. The most traded bonds are selected on a monthly basis as the 1/3 bonds with highest turnover that month. We have excluded a few bonds which where issued during the last 10 calendar days of the month in order not to mix up any index effect with the abnormal trading volume on the first days of trading in the secondary market. 88 Essay 3 0 10 20 30 40 50 10 15 20 25 30 Days around event Aggregate Trading Volume ($ billion) Figure 3.5: Total trading volume for newly issued bonds included into the index. Trading volume for each bond is added on a daily basis (without trading direction) and then the daily volume is added across bonds according to distance from the event. 2014 bonds trade on the event date. The average turnover for the 1/3 most traded bonds is 2.8%. The most traded bonds are selected on a monthly basis as the 1/3 bonds with highest turnover that month. We have excluded a few bonds which where issued during the last 10 calendar days of the month in order not to mix up any index effect with the abnormal trading volume on the first days of trading in the secondary market. Index Driven Price Pressure for Corporate Bonds 89 the secondary market. They show that the bonds are heavily traded in the first week of trading on the secondary market. In order not to mix up this newly issuance effect from the possible index effect on the last day of the month, all bonds issued (app. 5%) on the last 10 trading days of the month are excluded from the following analysis. Figure 3.5 shows the trading activity around the index inclusion date (event date) for newly issued bonds. Since no bonds are issued on the last 10 trading days of the month the number of bonds that could possibly trade is the same on all the plotted days. Cai, Helwege, and Warga (2007) and Goldstein and Hotchkiss (2008) make similar graphs with the first day of trading on the secondary market as the event date. On average the trading activity for newly issued bonds is decreasing towards a constant level over the first 60 days of trading. The same decreasing pattern can be recognized from figure 3.5 from +1 to +50 days after the inclusion event. The trading activity increases up to the inclusion day where it spikes. The average turnover on the inclusion day is 2.8% for the 1/3 most actively traded bonds. The spike indicates that some investors are actually tracking the index and are trying to do so with a low tracking error as in Blume and Edelen (2002). Abnormal returns around the event are shown in table 3.3. The benchmark return is chosen for each bond as the part of the Lehman/Barclay corporate bond index with matching rating and maturity as either intermediate term (<10 years) or longer term bonds (>10 years). Panel A of table 3.3 shows that the event return for inclusion is 9.6 bps and that the return stays up around 5 days after the event before it is fully reversed to the preevent level after 10 days of trading. In panel B the same abnormal returns are calculated, but only for the 1/3 most traded bonds. Each month the 1/3 bonds with highest turnover are selected to be included in this calculation. The reason for picking only the 1/3 most traded bonds is that on average a bond index fund only sample the index, which means that they only hold about 1/3 of all the bonds in the index. The majority of the remaining bonds might not be part of any index tracking strategy, in which case we would not expect them to have an abnormal return on the event date. Leaving them in could bias the abnormal returns towards zero. The abnormal returns are slightly higher for the most traded bonds. From day -1 to +5 the abnormal return is 15.4 bps compared to 9.8 bps for the entire bond index universe. As before the abnormal return is fully reversed, but for the most traded bonds it takes till day +15 after the event. The size of the price pressure and reversal returns can be compared to the underpricing returns found in Cai, Helwege, and Warga (2007) and Goldstein and Hotchkiss (2008). Cai, Helwege, and Warga (2007) find an insignificant first day abnormal return of 2 bps for investment grade bond IPOs in a study with data from 1995 to 1999. For a newer data sample matching the one in this paper, Goldstein and Hotchkiss (2008) find a significant first day return of around 40 bps. However, they do not construct an abnormal return correcting for market 90 Essay 3 movements. Panel A: All Bonds From (Day) To (Day) N Return (bps) T-Stat -1 1 2056 9.6 6.50*** -1 5 2050 9.8 5.05*** -1 10 2042 0.3 0.10 5 10 2140 -11.0 -5.58*** Panel B: Most traded From (Day) To (Day) N Return (bps) T-Stat -1 1 808 10.6 4.73*** -1 5 805 15.4 5.00*** -1 10 805 4.7 1.15 -1 15 803 -5.6 -1.18 5 10 827 -9.7 -2.91*** 5 15 824 -18.7 -4.24*** Table 3.3: Abnormal returns at index inclusion of newly issued bonds. The abnormal return is a size weighted average over all bonds for which it has been possible to calculate a return. The table excludes a few bonds that were issued on the last 10 trading days of the month in order not to mix up any underpricing with the index effect. The benchmark return for each bond is the corresponding Lehman/Barclys index that match on maturity and rating. The test statistics are cluster robust to time series and firm fixed effects. Panel A shows the results across all bonds. Panel B shows the results when only the most traded bonds are included in the analysis. The most traded bonds are selected on a monthly basis as the 1/3 bonds with highest turnover that month. The presence of the reversal return in our study supports that the index trackers trading behavior leads to a price pressure, where the price temporarily increases, because of the temporarily increased demand. The inclusion event is completely free of any information content and it is not expected to increase investor awareness, since the index is not limited to a fixed number of bonds. Altogether the return and trading activity evidence support the price pressure hypothesis/short term downward sloping demand curves for bonds. It could also be that demand curves are downward sloping in the long run. The empirical evidence is not contradicting this hypothesis, but since the pre and post event demand in the long run cannot be separated easily from the newly issued bond effect where demand is decreasing (Cai, Helwege, and Warga (2007) and Goldstein and Hotchkiss (2008)) it is hard to tell if there is any demand shift in the long run caused by the index inclusion. However, since there is no reason to expect any increased investor awareness we would not expect any demand shift even if we could separate 90 Essay 3 movements. Panel A: All Bonds From (Day) To (Day) N Return (bps) T-Stat -1 1 2056 9.6 6.50*** -1 5 2050 9.8 5.05*** -1 10 2042 0.3 0.10 5 10 2140 -11.0 -5.58*** Panel B: Most traded From (Day) To (Day) N Return (bps) T-Stat -1 1 808 10.6 4.73*** -1 5 805 15.4 5.00*** -1 10 805 4.7 1.15 -1 15 803 -5.6 -1.18 5 10 827 -9.7 -2.91*** 5 15 824 -18.7 -4.24*** Table 3.3: Abnormal returns at index inclusion of newly issued bonds. The abnormal return is a size weighted average over all bonds for which it has been possible to calculate a return. The table excludes a few bonds that were issued on the last 10 trading days of the month in order not to mix up any underpricing with the index effect. The benchmark return for each bond is the corresponding Lehman/Barclys index that match on maturity and rating. The test statistics are cluster robust to time series and firm fixed effects. Panel A shows the results across all bonds. Panel B shows the results when only the most traded bonds are included in the analysis. The most traded bonds are selected on a monthly basis as the 1/3 bonds with highest turnover that month. The presence of the reversal return in our study supports that the index trackers trading behavior leads to a price pressure, where the price temporarily increases, because of the temporarily increased demand. The inclusion event is completely free of any information content and it is not expected to increase investor awareness, since the index is not limited to a fixed number of bonds. Altogether the return and trading activity evidence support the price pressure hypothesis/short term downward sloping demand curves for bonds. It could also be that demand curves are downward sloping in the long run. The empirical evidence is not contradicting this hypothesis, but since the pre and post event demand in the long run cannot be separated easily from the newly issued bond effect where demand is decreasing (Cai, Helwege, and Warga (2007) and Goldstein and Hotchkiss (2008)) it is hard to tell if there is any demand shift in the long run caused by the index inclusion. However, since there is no reason to expect any increased investor awareness we would not expect any demand shift even if we could separate Index Driven Price Pressure for Corporate Bonds 91 the index effect from the newly issued effect in the long run. For the later period of the sample the transaction data contains information on trade direction where trades are marked as either a dealer sell (to a customer), a dealer buy or an interdealer trade. From this information a cumulative dealer inventory can be calculated. For each bond all sells and buys are added (with sign) on each date, so that a negative number indicates that on that day dealers have been net selling from their inventory. The inventory changes are then added across bonds and cumulated over time with day -5 before the event as the benchmark. Figure 3.6 shows the cumulative dealer inventory for the bonds included to the index over the period November 2008-August 2009. On the event date where the trading activity spiked in figure 3.5, figure 3.6 with the dealer inventory shows no abnormal change. Over the period the dealers seem to sell steadily out of their inventory. The interdealer trading volume does spike on the event date (not shown) which suggests that the dealers are not providing the liquidity for the index trackers but only perform a matching function, unlike what they do in the other three cases of index rebalancing. However, Edwards, Harris, and Piwowar (2007) and Feldh¨utter (2009) find that the bid-ask spread is around 20 bps for large trades across the period. So the dealers can make a higher profit from just providing the matching function, than they can from selling out of their inventory. Another reason could be that some dealers, usually some of the underwriters (see Dick-Nielsen, Feldh¨utter, and Lando (2009)), have market making responsibilities in the bond, which can make it unattractive to just unload their entire inventory at once. It is reasonable to think that index trackers prefer some bonds over others. Since they only sample the index they have to choose bonds that fit into a robust portfolio matching the duration, convexity and rating of the index. If some bonds are more attractive than other we would expect higher temporary demand for these and hence more price pressure. In order to test this hypothesis, we regress the bond specific abnormal return (day -1 to 1) on different bond characteristics and common bond return predictors. Table 3.4 shows the estimates and robust standard errors (see Petersen (2009)) for 4 different specifications of the regression. As bond specific predictors we use bond rating, coupon, maturity and size. We also include systematic abnormal return predictors, which are the slope and level of the treasury curve, credit risk factors (level and monthly change of a BAA yield curve) and illiquidity factors (level and monthly change). The daily illiquidity factor is calculated as a size weighted average over a daily Amihud price impact measure for all corporate bonds on the market. The daily illiquidity factor is then transformed to a monthly factor by taking the median over the month. The levels of the credit risk factor and the illiquidity factor are highly correlated (84%), so only one of the two factor types are included into the regression. In both regression specifications for all bonds, maturity and issue size are significant, which indicates a preference for large bond 92 Essay 3 −5 0 5 10 15 20 −7000 −6000 −5000 −4000 −3000 −2000 −1000 0 Days around event Dealer Inventory ($ million) Figure 3.6: Dealer inventory around inclusion of newly issued bonds. The graph shows the cumulative dealer inventory change. From November 2008 to August 2009 transactions in TRACE are marked as dealers sell, dealer buy or interdealer trade. Using this marking, we calculate dealer inventory change as the volume difference between dealer buys and sells each day for each bond. Then we add the bond specific inventory changes together according to distance from the event. Finally, we cumulate the aggregate changes using day -5 as benchmark. 92 Essay 3 −5 0 5 10 15 20 −7000 −6000 −5000 −4000 −3000 −2000 −1000 0 Days around event Dealer Inventory ($ million) Figure 3.6: Dealer inventory around inclusion of newly issued bonds. The graph shows the cumulative dealer inventory change. From November 2008 to August 2009 transactions in TRACE are marked as dealers sell, dealer buy or interdealer trade. Using this marking, we calculate dealer inventory change as the volume difference between dealer buys and sells each day for each bond. Then we add the bond specific inventory changes together according to distance from the event. Finally, we cumulate the aggregate changes using day -5 as benchmark. Index Driven Price Pressure for Corporate Bonds 93 All Bonds Most Traded Bonds Intercept -53.5 -57.7 -35.5 -63.1 -288.3 (-1.32) (-1.63) (-0.54) (-1.11) (-1.55) AAA 1.1 -4.6 13.0 0.5 -35.8 (0.12) (-0.5) (0.97) (0.04) (-0.66) AA 4.2 -1.1 10.3 -1.7 0.7 (0.75) (-0.21) (1.34) (-0.24) (0.03) A 6.1* 0.6 7.4 -3.0 9.4 (1.66) (0.16) (1.34) (-0.57) (0.53) BAA - - - - - ----- Coupon 2.9* -1.3 2.1 -4.7** -4.3 (1.73) (-0.79) (0.8) (-1.98) (-0.79) TTM 0.4** 0.6*** 0.5* 0.8*** 0.9 (2.07) (3.08) (1.84) (3.14) (1.06) Log(Size) 6.1** 6.9*** 1.1 5.3 5.0 (2.56) (2.97) (0.29) (1.43) (0.4) Treasury Level -10.6*** -7.4* -1.6 2.4 44.4 (-3.18) (-1.78) (-0.33) (0.41) (1.12) Treasury Slope -7.7** -7.6** -2.7 -1.3 39.9 (-2.55) (-2.5) (-0.63) (-0.31) (0.75) BAA yield 0.015 1.9 (0.01) (0.42) ∆BAA yield 14.2*** 6.9 (2.71) (0.91) Illiq 1765.7** 2266.2* 6178.7 (2.06) (1.86) (1.02) ∆Illiq 0.022*** 0.029*** 0.036*** (11.59) (10.55) (4.54) ∆Inventory -0.9** (-2.37) R22.6 8.5 1.5 14.0 20.1 N 2055 2053 807 805 164 Table 3.4: Abnormal event return regression for newly issued bonds included in the index. The table shows the regression coefficients from regressions of the abnormal event return (day -1 to +1) for newly issued bonds included into the index on different issue specific characteristics and common bond return predictors. The bond characteristics are rating, coupon, maturity and the logarithm of the issuance size. The common bond return predictors are the level of a Moody’s BAA yield curve, the monthly change in this yield curve, treasury level (1 year yield), treasury slope (10 - 1 year yield), and the level and change of an illiquidity factor. The illiquidity factor is a size weighted average of a bond specific monthly Amihud price impact measure across the entire corporate bond market. The first 4 columns show the results when all bonds are used and the remaining columns show the results for the same regressions when only the most traded bonds are used. The most traded bonds are selected on a monthly basis as the 1/3 bonds with highest turnover that month. All standard error are robust i.e. controlled for time series and firm fixed effects. 94 Essay 3 issues with long maturities. This makes sense since larger issues weigh more in the index than smaller issues and when buying longer term bonds index trackers reduce transaction costs. Running the same regression on the most heavily traded bonds show that maturity still remains significant. The issuance size has dropped out of the regression but that is mainly due to the conditioning on the most traded bonds, which is mainly the large issues. Even though the levels of the credit risk factor and the illiquidity factor is highly correlated, the illiquidity factors do a better job at explaining the returns than the credit risk factors based on the R2. The driving difference is the monthly change in the illiquidity factor. The change in the illiquidity factor is not as much correlated with the corresponding change in the credit risk factor. When the market is illiquid or credit risk is high the abnormal returns are higher. Also when the illiquidity or credit risk have increased over the month abnormal returns are higher. The latter could be seen as extra compensation for the liquidity providers, who have probably bought the bond on the primary market and then hold it to the last day of the month. If liquidity has gone down, then the liquidity provision strategy has become more risky during the month and the index trackers may also have more problems locating the bond, which would lower their bargaining power against the liquidity provider. In the last column of table 3.4 we include the bond specific change to dealer inventory over day -1 to +1 around the inclusion. The inventory change is negative and significant (on a 5% level). This means that the more dealers net sell from their inventory the higher the abnormal return. One could also argue for the causality to be turned around, so that dealers only participate in the liquidity provision when the abnormal return is high (enough). 3.7 Downgraded Bonds When a bond is downgraded from an index rating of investment grade to speculative grade it gets excluded from the index at the last trading day of the month. Still, whereas the downgrade itself could contain new information to the market the subsequent index exclusion should not contain any information. A corporate bond that gets downgraded from investment grade to speculative grade is also known as a fallen angel. Price pressure for fallen angels have been separately studied in Ambrose, Cai, and Helwege (2009), but without paying any attention to the index exclusion event. Figure 3.7 shows the aggregate trading volume for all bonds excluded because of a downgrade. There is a clear spike on the event date indicating that index trackers sell out of their portfolio in order to minimize their tracking error. The bonds are downgraded between day -21 and -1, with most bonds being downgraded on the 15th calendar day of the month, which explains the tall spike on that day. When the bonds become speculative grade they start trading less than before (even when the period is extended 94 Essay 3 issues with long maturities. This makes sense since larger issues weigh more in the index than smaller issues and when buying longer term bonds index trackers reduce transaction costs. Running the same regression on the most heavily traded bonds show that maturity still remains significant. The issuance size has dropped out of the regression but that is mainly due to the conditioning on the most traded bonds, which is mainly the large issues. Even though the levels of the credit risk factor and the illiquidity factor is highly correlated, the illiquidity factors do a better job at explaining the returns than the credit risk factors based on the R2. The driving difference is the monthly change in the illiquidity factor. The change in the illiquidity factor is not as much correlated with the corresponding change in the credit risk factor. When the market is illiquid or credit risk is high the abnormal returns are higher. Also when the illiquidity or credit risk have increased over the month abnormal returns are higher. The latter could be seen as extra compensation for the liquidity providers, who have probably bought the bond on the primary market and then hold it to the last day of the month. If liquidity has gone down, then the liquidity provision strategy has become more risky during the month and the index trackers may also have more problems locating the bond, which would lower their bargaining power against the liquidity provider. In the last column of table 3.4 we include the bond specific change to dealer inventory over day -1 to +1 around the inclusion. The inventory change is negative and significant (on a 5% level). This means that the more dealers net sell from their inventory the higher the abnormal return. One could also argue for the causality to be turned around, so that dealers only participate in the liquidity provision when the abnormal return is high (enough). 3.7 Downgraded Bonds When a bond is downgraded from an index rating of investment grade to speculative grade it gets excluded from the index at the last trading day of the month. Still, whereas the downgrade itself could contain new information to the market the subsequent index exclusion should not contain any information. A corporate bond that gets downgraded from investment grade to speculative grade is also known as a fallen angel. Price pressure for fallen angels have been separately studied in Ambrose, Cai, and Helwege (2009), but without paying any attention to the index exclusion event. Figure 3.7 shows the aggregate trading volume for all bonds excluded because of a downgrade. There is a clear spike on the event date indicating that index trackers sell out of their portfolio in order to minimize their tracking error. The bonds are downgraded between day -21 and -1, with most bonds being downgraded on the 15th calendar day of the month, which explains the tall spike on that day. When the bonds become speculative grade they start trading less than before (even when the period is extended Index Driven Price Pressure for Corporate Bonds 95 −40 −20 0 20 40 12345 Days around event Aggregate Trading Volume ($ billion) Figure 3.7: Total trading volume for bonds excluded because of a downgrade. Trading volume for each bond is added on a daily basis (without trading direction) and then the daily volume is added across bonds according to distance from the event. 412 bonds trade on the event date. The average turnover is 1.5%. We have excluded a few bonds which where downgraded during the last 5 calendar days of the month in order not to mix up any index effect with the abnormal trading volume around the actual downgrade. 96 Essay 3 back in time on the graph before any firm distress). Measured by outstanding volume Ford and GM bonds make up a large part of the market and an even larger part when only looking at fallen angels. The large volume spike in the downgrade period is mainly caused by trading in Ford and GM and the spike at day +21 is because Ford gets an index upgrade again the following month when Lehman included Fitch in the determination of spilt ratings (as explained in section 3.3). Table 3.5 shows the abnormal returns for different periods around the exclusion event. Panel A shows the results for all bonds excluded whereas panel B shows the results without GM and Ford. Since Ford and GM make up such a large part of the market they might trade differently than other bonds. The event return from day -1 to +1 without GM and Ford is - 136.0 bps consistent with a price decrease when index trackers are trying to sell the bonds. The negative return is reversed over the next 5 to 10 days. The pattern is the same when Ford and GM are included, although the event return is virtually 0 bps. The price falls on average with -2,145.7 bps in the downgrade period from day -21 to -5. This is not surprising since the downgrade probably contains some information for some of the bonds. Ambrose, Cai, and Helwege (2009) find that not all bonds fall in price at the downgrade to speculative grade, but mainly the bonds for which the stock price also decreases, which suggests that the price only falls when the downgrade conveys information. What is interesting in table 3.5 is that the average bond also decreases in price between day -5 to 0, even though no firms are downgraded in that specific period. It could be because the index trackers are selling out in this period or just because the market is illiquid, so that it takes some time for the downgrade information to get impounded into the price. The following reversal in price indicates that index trackers cause some price pressure. In panel B the event return of -136 bps is almost fully reversed after 10 days. Figure 3.8 shows the cumulative change in dealer inventory over the shorter period from November 2008 to August 2009. The dealers unload (part of) their inventory before the downgrade, which can be seen from the decrease in inventory between day -30 and -21. It does not mean that the dealers can forecast the downgrade, since the graph is conditional upon downgrade and dealers might also decrease their inventory in other bonds which did not end up being downgraded. Comparing with the abnormal returns in table 3.5 it seems smart to unload inventory in this period, since the abnormal return between day -30 to -21 is 13.3 bps for all bonds and -119.7 when excluding Ford and GM. Both returns are far less than the downgrade month return of more than -2,000 bps. In the downgrade month from day -21 to -5 (where all the bonds in the sample get downgraded), the dealers keep a constant inventory. The constant inventory should be seen in connection with figure 3.9. Figure 3.9 shows the aggregate trading volume across all the bonds with day 0 being the actual downgrade date (and not 96 Essay 3 back in time on the graph before any firm distress). Measured by outstanding volume Ford and GM bonds make up a large part of the market and an even larger part when only looking at fallen angels. The large volume spike in the downgrade period is mainly caused by trading in Ford and GM and the spike at day +21 is because Ford gets an index upgrade again the following month when Lehman included Fitch in the determination of spilt ratings (as explained in section 3.3). Table 3.5 shows the abnormal returns for different periods around the exclusion event. Panel A shows the results for all bonds excluded whereas panel B shows the results without GM and Ford. Since Ford and GM make up such a large part of the market they might trade differently than other bonds. The event return from day -1 to +1 without GM and Ford is - 136.0 bps consistent with a price decrease when index trackers are trying to sell the bonds. The negative return is reversed over the next 5 to 10 days. The pattern is the same when Ford and GM are included, although the event return is virtually 0 bps. The price falls on average with -2,145.7 bps in the downgrade period from day -21 to -5. This is not surprising since the downgrade probably contains some information for some of the bonds. Ambrose, Cai, and Helwege (2009) find that not all bonds fall in price at the downgrade to speculative grade, but mainly the bonds for which the stock price also decreases, which suggests that the price only falls when the downgrade conveys information. What is interesting in table 3.5 is that the average bond also decreases in price between day -5 to 0, even though no firms are downgraded in that specific period. It could be because the index trackers are selling out in this period or just because the market is illiquid, so that it takes some time for the downgrade information to get impounded into the price. The following reversal in price indicates that index trackers cause some price pressure. In panel B the event return of -136 bps is almost fully reversed after 10 days. Figure 3.8 shows the cumulative change in dealer inventory over the shorter period from November 2008 to August 2009. The dealers unload (part of) their inventory before the downgrade, which can be seen from the decrease in inventory between day -30 and -21. It does not mean that the dealers can forecast the downgrade, since the graph is conditional upon downgrade and dealers might also decrease their inventory in other bonds which did not end up being downgraded. Comparing with the abnormal returns in table 3.5 it seems smart to unload inventory in this period, since the abnormal return between day -30 to -21 is 13.3 bps for all bonds and -119.7 when excluding Ford and GM. Both returns are far less than the downgrade month return of more than -2,000 bps. In the downgrade month from day -21 to -5 (where all the bonds in the sample get downgraded), the dealers keep a constant inventory. The constant inventory should be seen in connection with figure 3.9. Figure 3.9 shows the aggregate trading volume across all the bonds with day 0 being the actual downgrade date (and not Index Driven Price Pressure for Corporate Bonds 97 Panel A: All Bonds From (Day) To (Day) N Return (bps) T-Stat -1 1 360 0.2 0.003 -5 0 346 -935.1 -5.28*** -5 10 317 -586.5 -3.14*** 0 4 315 348.7 6.17*** 0 10 348 255.8 4.28*** -30 -21 279 13.3 0.31 -30 10 271 -2682.2 -6.20*** -21 -5 808 -2145.7 -7.34*** Panel B: Without GM and Ford From (Day) To (Day) N Return (bps) T-Stat -1 1 334 -136.0 -2.06** -5 0 320 -1303.5 -6.25*** -5 10 291 -930.7 -4.35*** 0 4 289 270.2 4.29*** 0 10 322 125.0 1.90* -30 -21 253 -119.7 -2.53** -30 10 245 -3652.5 -7.28*** -21 -5 808 -2742.3 -8.23*** Table 3.5: Abnormal returns at index exclusion of downgraded bonds. The abnormal return is a size weighted average over all bonds for which it has been possible to calculate a return. The table excludes a few bonds that were downgraded on the last 5 trading days of the month in order not to mix up any downgrade day effect with the index effect. The benchmark return for each bond is the corresponding high yield Lehman/Barclys index that match on maturity and rating. The test statistics are cluster robust to time series and firm fixed effects. Panel A shows the results across all bonds. Panel B shows the results without Ford and GM bonds, which together make up a large part of the sample measured by volume. 104 Essay 3 Panel A: All Bonds From (Day) To (Day) N Return (bps) T-Stat -1 1 188 21.7 3.96*** -5 0 208 32.1 3.96*** -5 10 197 63.6 5.13*** 0 4 161 -16.8 -2.60** 0 10 194 32.8 3.83*** -30 -21 165 257.2 8.35*** -30 10 155 259.6 8.31*** -21 -5 186 -13.6 -1.03 Panel B: Without index rule redefinition months From (Day) To (Day) N Return (bps) T-Stat -1 1 136 2.5 0.41 -5 0 152 9.3 1.07 -5 10 140 2.0 0.15 0 4 114 4.8 0.71 0 10 138 -7.4 -0.83 -30 -21 113 16.1 0.97 -30 10 101 53.8 1.95* -21 -5 128 43.1 2.69*** Table 3.6: Abnormal returns at index inclusion of upgraded bonds. The abnormal return is a size weighted average over all bonds for which it has been possible to calculate a return. The table excludes a few bonds that were upgraded on the last 5 trading days of the month in order not to mix up any upgrade day effect with the index effect. The benchmark return for each bond is the corresponding Lehman/Barclys index that match on maturity and rating. The test statistics are cluster robust to time series and firm fixed effects. Panel A shows the results across all bonds. Panel B excludes the bonds that where mechanically upgraded because of the rating rule change in July 2005. 104 Essay 3 Panel A: All Bonds From (Day) To (Day) N Return (bps) T-Stat -1 1 188 21.7 3.96*** -5 0 208 32.1 3.96*** -5 10 197 63.6 5.13*** 0 4 161 -16.8 -2.60** 0 10 194 32.8 3.83*** -30 -21 165 257.2 8.35*** -30 10 155 259.6 8.31*** -21 -5 186 -13.6 -1.03 Panel B: Without index rule redefinition months From (Day) To (Day) N Return (bps) T-Stat -1 1 136 2.5 0.41 -5 0 152 9.3 1.07 -5 10 140 2.0 0.15 0 4 114 4.8 0.71 0 10 138 -7.4 -0.83 -30 -21 113 16.1 0.97 -30 10 101 53.8 1.95* -21 -5 128 43.1 2.69*** Table 3.6: Abnormal returns at index inclusion of upgraded bonds. The abnormal return is a size weighted average over all bonds for which it has been possible to calculate a return. The table excludes a few bonds that were upgraded on the last 5 trading days of the month in order not to mix up any upgrade day effect with the index effect. The benchmark return for each bond is the corresponding Lehman/Barclys index that match on maturity and rating. The test statistics are cluster robust to time series and firm fixed effects. Panel A shows the results across all bonds. Panel B excludes the bonds that where mechanically upgraded because of the rating rule change in July 2005. Index Driven Price Pressure for Corporate Bonds 105 −20 −10 0 10 20 −40 −20 0 20 40 60 Days around event Dealer Inventory ($ million) Figure 3.13: Dealer inventory around inclusion of upgraded bonds. The graph shows the cumulative dealer inventory change. From November 2008 to August 2009 transactions in TRACE are marked as dealers sell, dealer buy or interdealer trade. Using this marking, we calculate dealer inventory change as the volume difference between dealer buys and sells each day for each bond. Then we add the bond specific inventory changes together according to distance from the event. Finally, we cumulate the aggregate changes using day -21 as benchmark. 106 Essay 3 mechanically upgraded because of the index rule change in 2005 the event returns become insignificant. The remaining bonds show a positive and significant abnormal return in the period of the actual upgrade of 43.1 bps, which seems to be a lasting price increase. When including all bonds there is a significant and positive return of 257.2 bps from day -30 to -21, which is before the upgrade month. However, since the rule change was announced 5 month prior investors could at this point forecast that specific bonds would be upgraded. The increase in price before the upgrade month, which is mainly driven by the mechanically upgraded bonds could be explained by the incentives for the investors to front run the market in anticipation of a further price increase once the bond is actually upgraded. Figure 3.13 shows the cumulative change in dealer inventory around the index inclusion. The graph is only based on 28 bonds for the period November 2008 to August 2009, so it is not very robust. It seems like the dealers are buying up the bonds around the actual upgrade and then sell out in the days up to the index inclusion. Such a dealer strategy would be consistent with price pressure caused by the index trackers, but we could not see that effect in the returns in panel B of table 3.6. 3.9 Conclusion Similar to stock index trackers, corporate bond index trackers seek to minimize their tracking error. This results in an increased trading activity at the rebalancing date in bonds that are included to or excluded from the Lehman/Barclays index. The trading activity spikes both when bonds are included because they are newly issued, when they are included because they are upgraded, when they are excluded because of low maturity and when they are excluded because they have been downgraded from investment grade to speculative grade. The information deciding whether a bond should be included or excluded is always available before the index revision and in most cases even long before. Hence, the informational content leading to the index revision usually gets impounded into prices before the bond is excluded or included. Parallel with the increased trading activity from index trackers at the rebalancing date, the bonds experience a price pressure. When index trackers buy up bonds the price temporarily increases, causing an abnormal positive return. The abnormal return is fully reversed 5 to 10 trading days after the index inclusion. The opposite happens when index trackers are selling a bond that leaves the index. The price temporarily decreases, only to be reversed afterwards. The abnormal price pressure return and reversal are significant in all four cases, except for upgraded bonds where there is no price reaction to the index inclusion. For bonds excluded because of low maturity and for bonds excluded because of a downgrade, the reversal returns are economically significant and 106 Essay 3 mechanically upgraded because of the index rule change in 2005 the event returns become insignificant. The remaining bonds show a positive and significant abnormal return in the period of the actual upgrade of 43.1 bps, which seems to be a lasting price increase. When including all bonds there is a significant and positive return of 257.2 bps from day -30 to -21, which is before the upgrade month. However, since the rule change was announced 5 month prior investors could at this point forecast that specific bonds would be upgraded. The increase in price before the upgrade month, which is mainly driven by the mechanically upgraded bonds could be explained by the incentives for the investors to front run the market in anticipation of a further price increase once the bond is actually upgraded. Figure 3.13 shows the cumulative change in dealer inventory around the index inclusion. The graph is only based on 28 bonds for the period November 2008 to August 2009, so it is not very robust. It seems like the dealers are buying up the bonds around the actual upgrade and then sell out in the days up to the index inclusion. Such a dealer strategy would be consistent with price pressure caused by the index trackers, but we could not see that effect in the returns in panel B of table 3.6. 3.9 Conclusion Similar to stock index trackers, corporate bond index trackers seek to minimize their tracking error. This results in an increased trading activity at the rebalancing date in bonds that are included to or excluded from the Lehman/Barclays index. The trading activity spikes both when bonds are included because they are newly issued, when they are included because they are upgraded, when they are excluded because of low maturity and when they are excluded because they have been downgraded from investment grade to speculative grade. The information deciding whether a bond should be included or excluded is always available before the index revision and in most cases even long before. Hence, the informational content leading to the index revision usually gets impounded into prices before the bond is excluded or included. Parallel with the increased trading activity from index trackers at the rebalancing date, the bonds experience a price pressure. When index trackers buy up bonds the price temporarily increases, causing an abnormal positive return. The abnormal return is fully reversed 5 to 10 trading days after the index inclusion. The opposite happens when index trackers are selling a bond that leaves the index. The price temporarily decreases, only to be reversed afterwards. The abnormal price pressure return and reversal are significant in all four cases, except for upgraded bonds where there is no price reaction to the index inclusion. For bonds excluded because of low maturity and for bonds excluded because of a downgrade, the reversal returns are economically significant and Index Driven Price Pressure for Corporate Bonds 107 above an average bid-ask spread for large trades. We present empirical evidence that shows dealers only participate as liquidity providers for the index trackers when the reversal return is higher than bid-ask spreads. Hence, dealers increase their inventory up to the rebalancing date, when buying up excluded bonds. After the rebalancing date dealers decrease their inventory again to the pre-event level, so that they profit from the reversal return. Consistent with a profit maximization strategy, dealers do not trade against their inventory, when the reversal return is lower than an average bid-ask spread. This happens when newly issued bonds are included into the index and at the actual downgrade date. In both of these cases dealers only perform a matching function. Index trackers seek to minimize transaction costs and thus only sample the bond index. In effect they do not hold all bonds in the index. We present evidence on which bonds are most attractive for the index trackers. In a regression using the price pressure return of newly issued bonds included into the index, we find that maturity and issuance size help predict the abnormal return and e.g. bond rating do not. This indicates that index trackers prefer bonds which weigh more in the index and which have a longer time in the index, both characteristics reduce the need for frequent trading. Summary English Summary Essay 1: Liquidity Biases in TRACE The transactions database TRACE is rapidly becoming the standard data source for empirical research on US corporate bonds. This paper is the first to thoroughly discuss the assumptions needed to clean the disseminated TRACE data and to suggest that different filters should be used depending upon the application. According to the FINRA rule 6700 series all members (dealers) are required to report an over-the-counter corporate bond transaction in the secondary market using the TRACE system. The TRACE system was introduced for the first time in July 2002 and the dissemination of transactions from TRACE has gradually been expanded to include the entire corporate bond universe. The dissemination of the transactions is a great improvement for empirical research, but the construction of the system means that errors are accumulating. Around 7.7% of all trade reports in TRACE are errors and in some cases up to 18% of the reports should be deleted. The errors accumulate because TRACE is essentially a one day system. If a dealer makes a reporting error, he can easily correct it if the correction is made within the same day as the report was filed. In this case the dealer files a new report, but the old report containing the error still remains in TRACE. In the disseminated data this yields two reports. The first report contains the error, the second report either cancels or corrects the wrong report but none of them replaces the first report. Hence, one or both reports should be deleted from the sample before the data is used for research. While same-day corrections can easily be matched with the original trade report, corrections on a later date cannot. In order to find the original report in the latter case a range of assumptions are needed. This paper explains the assumptions and sets up a filter that deletes almost all error reports from TRACE. For the 10 most frequently traded bonds in 2007 the deviation between filtered TRACE reports and the official FINRA number of reports is in the range of 0.05%. Failing to filter the data before use will most likely result in different biases. As an example this paper shows 109 110 Essay 3 that popular liquidity measures will be biased towards a more liquid market. The median bias for the daily turnover will be 7.4% and for a quarter of the bonds the Amihud price impact measure will be underestimated by at least 14.6%. These biases are encountered if nothing is done about the data after download. WRDS supplies the TRACE data for academic research together with some sample programs. However, if one uses the sample programs, they will encounter exactly the biases just described. A naive filter would be to delete all trade reports marked as error reports. This is what Bloomberg does when they report statistics based on TRACE data. However, the naive filter deletes less than half of what should be deleted, since they do nothing about the original report containing the error. Even after applying the error filter there is still a lot of identical trade reports in TRACE. They are a result of the way certain agency transactions are reported. I suggest deleting the duplicates if the price sequence is important, since the sequence otherwise would be biased. Finally, I show that normal price based filters cannot replace the error filter. In a normal price based filter trade reports are deleted if the transaction price falls outside a certain range based on the surrounding price sequence. A standard price sequence filter only deletes a very small fraction of the errors detected by the error filter. Price sequence filters are usually motivated by a desire to detect and delete typing errors from the dealers. However, it is worth noting that the TRACE system by itself performs a price sequence test. If the price deviates too much the trade report is dismissed and the dealer has to overwrite the system. The TRACE system then expands the allowed price range. If the price is still outside the range the report is dismissed again and the dealer has to phone in the report and explain why the price deviates. Essay 2: Corporate Bond Liquidity Before and After the Onset of the Subprime Crisis The subprime crisis dramatically increased corporate bond spreads and while default risk certainly has increased because of funding constraints and the slowing of the real economy, it is also widely believed that deteriorating liquidity has contributed to the widening of spreads. The difficulty is how to measure this contribution. We analyze liquidity components of corporate bond spreads by using transaction-level corporate bond prices from TRACE. The high data quality of TRACE allow us to compute a range of different liquidity measures and asses their performance over a period spanning both a pre-crisis period and the onset of the subprime crisis. When calculating the liquidity measures we restrict the transaction sample to only include trades with a nominal value above 100,000$. In this way we only look at trades from sophisticated traders, since retail traders rarely trade above 100,000$. The volume restriction together with the transaction level information significantly reduce the size of the liquidity measures 110 Essay 3 that popular liquidity measures will be biased towards a more liquid market. The median bias for the daily turnover will be 7.4% and for a quarter of the bonds the Amihud price impact measure will be underestimated by at least 14.6%. These biases are encountered if nothing is done about the data after download. WRDS supplies the TRACE data for academic research together with some sample programs. However, if one uses the sample programs, they will encounter exactly the biases just described. A naive filter would be to delete all trade reports marked as error reports. This is what Bloomberg does when they report statistics based on TRACE data. However, the naive filter deletes less than half of what should be deleted, since they do nothing about the original report containing the error. Even after applying the error filter there is still a lot of identical trade reports in TRACE. They are a result of the way certain agency transactions are reported. I suggest deleting the duplicates if the price sequence is important, since the sequence otherwise would be biased. Finally, I show that normal price based filters cannot replace the error filter. In a normal price based filter trade reports are deleted if the transaction price falls outside a certain range based on the surrounding price sequence. A standard price sequence filter only deletes a very small fraction of the errors detected by the error filter. Price sequence filters are usually motivated by a desire to detect and delete typing errors from the dealers. However, it is worth noting that the TRACE system by itself performs a price sequence test. If the price deviates too much the trade report is dismissed and the dealer has to overwrite the system. The TRACE system then expands the allowed price range. If the price is still outside the range the report is dismissed again and the dealer has to phone in the report and explain why the price deviates. Essay 2: Corporate Bond Liquidity Before and After the Onset of the Subprime Crisis The subprime crisis dramatically increased corporate bond spreads and while default risk certainly has increased because of funding constraints and the slowing of the real economy, it is also widely believed that deteriorating liquidity has contributed to the widening of spreads. The difficulty is how to measure this contribution. We analyze liquidity components of corporate bond spreads by using transaction-level corporate bond prices from TRACE. The high data quality of TRACE allow us to compute a range of different liquidity measures and asses their performance over a period spanning both a pre-crisis period and the onset of the subprime crisis. When calculating the liquidity measures we restrict the transaction sample to only include trades with a nominal value above 100,000$. In this way we only look at trades from sophisticated traders, since retail traders rarely trade above 100,000$. The volume restriction together with the transaction level information significantly reduce the size of the liquidity measures Index Driven Price Pressure for Corporate Bonds 111 towards a more liquid pre-crisis market than what is usually found in previous studies. The performance of each bond liquidity measure is assessed by running marginal regressions with only one liquidity measure at the time. We use quarter-end yield spreads as the depended variable in all regressions and a range of different credit risk controls as independent variables. The marginal regressions show that the Amihud price impact measure, a measure of round-trip transaction costs and the standard deviation of these two measure are all significant in explaining the liquidity component in credit spread both before and after the onset of the crisis. The bond turnover, the Roll measure, the number of zero-trading days and the number of firm zero trading days are not consistently significant. Furthermore, we find in a principal component analysis of the liquidity measures that an equally weighted linear combination of the four significant measures captures most of the liquidity-related variation of credit spreads. Using this new measure of bond liquidity we estimate the absolute and relative size of the liquidity component in credit spreads. Further, we use the measure to shed new light on flight-to-quality, liquidity risk, the impact of trading frequency, the role of funding shocks to lead underwriters, and the liquidity of corporate bonds issued by financial firms. We find that before the crisis, the contribution to spreads from illiquidity was small for investment grade bonds both measured in basis points and as a fraction of total spreads. The contribution increased strongly at the onset of the crisis for all bonds except AAA-rated bonds, which is consistent with a flight-to- quality into AAA-rated bonds. Liquidity premia in investment grade bonds rose steadily during the crisis and peaked when the stock market declined strongly in the first quarter of 2009, while premia in speculative grade bonds peaked during the Lehman default and returned almost to pre-crisis levels in mid-2009. The number of zero trading days did not increase with the crisis and we find evidence that this was because trades in less liquid bonds were split into trades of smaller size. Essay 3: Index Driven Price Pressure for Corporate Bonds The impact of stock index tracking has been intensely studied and there exists several competing theories seeking to explain the price reaction at index inclusion. This paper is the first to test similar theories for a corporate bond index. Unlike the S&P 500 index, inclusions and exclusions to the Lehman/Barclays Corporate Bond Index are monthly recurrent events. The rules for inclusion and exclusion are fully transparent and based on bonds characteristics. Another unique feature of bond indices is that they are not limited to a certain number of securities. The main reasons for inclusion are that the bonds are newly issued or that the bonds get upgraded from speculative grade to investment grade. The inclusion itself always happens on the last trading day of the month, which makes the inclusions information 112 Essay 3 free events. Trading activity at the date of inclusion is higher than normal and this spike indicates that some investors are in fact tracking the index. The incentive for index trackers to trade close to the rebalancing date is that they want to minimize tracking error (Blume and Edelen (2002)). A similar trading pattern can be observed for excluded bonds. Again, the trading activity spikes at the exclusion date, where index trackers want to sell the bonds. The main reason for bonds to be excluded from the index is because their maturity falls below one year or because they are downgraded from investment grade to speculative grade. As with the inclusions, exclusions are effectuated at the last trading day of the month. Parallel with the increased trading activity from index trackers, the bonds experience a price pressure. When index trackers buy up bonds the price temporarily increases, causing an abnormal positive return. The abnormal return is fully reversed 5 to 10 trading days after the index inclusion. The opposite happens when index trackers are selling a bond that leaves the index. The price temporarily decreases, only to be fully reversed afterwards. The abnormal price pressure return and reversal are both significant in all four cases, except for upgraded bonds. For bonds excluded because of low maturity and for bonds excluded because of a downgrade, the reversal return is significant and above an average bid-ask spread for large trades over the period. In these two cases, dealers participate as liquidity providers for the index trackers. Dealers increase their inventory up to the rebalancing date, when buying up from the index trackers. After the rebalancing date dealers decrease their inventory again to the pre-event level, while they earn the reversal return. Consistent with a profit maximization strategy, the dealers do not trade against their inventory, when the reversal return is lower than an average bid-ask spread. This happens both when newly issued bonds are included into the index and at the actual downgrade date, where dealers only perform a matching function. We find that the price pressure for newly issued bonds at index inclusion is positively correlated with maturity and issuance size and not with e.g. bond rating. This indicates that index trackers prefer bonds which weigh more in the index and which have a longer time in the index. These characteristics are important for the bond index trackers, since bond index trackers do not replicate the index, but only sample the index. Sampling the index means that they only hold part of the index, unlike S&P 500 index funds that holds all 500 securities. The Lehman/Barclay corporate bond index has around 3,400 bonds, so holding all the bonds would result in a very high level of transaction costs. In order to avoid the transaction costs, bond index funds only hold about 1/3 of the bonds in the index and they invest 20% of their portfolio outside the index. Despite the sampling strategy, we still find index driven price pressure in corporate bonds. 112 Essay 3 free events. Trading activity at the date of inclusion is higher than normal and this spike indicates that some investors are in fact tracking the index. The incentive for index trackers to trade close to the rebalancing date is that they want to minimize tracking error (Blume and Edelen (2002)). A similar trading pattern can be observed for excluded bonds. Again, the trading activity spikes at the exclusion date, where index trackers want to sell the bonds. The main reason for bonds to be excluded from the index is because their maturity falls below one year or because they are downgraded from investment grade to speculative grade. As with the inclusions, exclusions are effectuated at the last trading day of the month. Parallel with the increased trading activity from index trackers, the bonds experience a price pressure. When index trackers buy up bonds the price temporarily increases, causing an abnormal positive return. The abnormal return is fully reversed 5 to 10 trading days after the index inclusion. The opposite happens when index trackers are selling a bond that leaves the index. The price temporarily decreases, only to be fully reversed afterwards. The abnormal price pressure return and reversal are both significant in all four cases, except for upgraded bonds. For bonds excluded because of low maturity and for bonds excluded because of a downgrade, the reversal return is significant and above an average bid-ask spread for large trades over the period. In these two cases, dealers participate as liquidity providers for the index trackers. Dealers increase their inventory up to the rebalancing date, when buying up from the index trackers. After the rebalancing date dealers decrease their inventory again to the pre-event level, while they earn the reversal return. Consistent with a profit maximization strategy, the dealers do not trade against their inventory, when the reversal return is lower than an average bid-ask spread. This happens both when newly issued bonds are included into the index and at the actual downgrade date, where dealers only perform a matching function. We find that the price pressure for newly issued bonds at index inclusion is positively correlated with maturity and issuance size and not with e.g. bond rating. This indicates that index trackers prefer bonds which weigh more in the index and which have a longer time in the index. These characteristics are important for the bond index trackers, since bond index trackers do not replicate the index, but only sample the index. Sampling the index means that they only hold part of the index, unlike S&P 500 index funds that holds all 500 securities. The Lehman/Barclay corporate bond index has around 3,400 bonds, so holding all the bonds would result in a very high level of transaction costs. In order to avoid the transaction costs, bond index funds only hold about 1/3 of the bonds in the index and they invest 20% of their portfolio outside the index. Despite the sampling strategy, we still find index driven price pressure in corporate bonds. Index Driven Price Pressure for Corporate Bonds 113 Dansk Resum´e Essay 1: Likviditetsbias’ i TRACE Transaktionsdatabasen TRACE er hurtigt ved at blive standarden indenfor empiriske undersøgelser af amerikanske virksomhedsobligationer. Denne artikel er den første, der grundigt diskuterer de forudsætninger, der ligger til grund for anvendelserne af dette datasæt. Inden man kan benytte den givne data er det nødvendigt at filtrere datasættet og slette alle transaktionsrapporter, der indeholder fejl, hvilket ikke kan gøres uden visse antagelser. Ifølge amerikansk lovgivning kræves det, at alle dealere indberetter handler i virksomhedsobligationer gennem TRACE systemet senest 15 min. efter, handlen er effektueret. TRACE systemet blev introduceret i juli 2002 og transaktionsrapporterne er efterfølgende blevet offentligt tilgængelige. Tilgængeligheden er gradvist blevet øget s˚aledes, at alle transaktioner i dag er offentligt tilgængelige s˚a snart, de er blevet rapporteret. TRACE systemet er s˚aledes en kæmpe fordel for empiriske studier, men selve konstruktionen af TRACE gør, at fejl akkumuleres i det offentliggjorte datasæt. Omkring 7.7% af transaktionsrapporterne i TRACE skal slettes før datasættet kan bruges til forskning, og i nogle tilfælde skal helt op til 18% slettes. Fejlrapporterne akkumuleres, fordi TRACE er konstrueret som et samme-dag-system. Hvis en dealer kommer til at sende en rapport med en fejl, kan han nemt rette den indenfor samme dag, som rapporten er sendt. For at rette en fejl indenfor samme dag sender dealeren en ny rapport, men uden at den gamle rapport slettes. Den nye rapport fortæller blot, at den gamle rapport var en fejl, som man skal se bort fra. Men systemet indeholder nu to rapporter for en enkelt handel. Før datasættet kan bruges til forskning skal mindst ´en af rapporterne slettes. Hvis en dealer først retter en fejl p˚a en dag, der ligger senere end den dag, hvor den oprindelige rapport blev sendt, sker det samme som lige beskrevet. Men i det offentlige datasæt kan vi nemt matche fejlrapporten med original rapporten indenfor samme dag, mens vi ikke kan lave samme match, n˚ar fejlrrapporten er sendt p˚a en senere dato. I det sidst nævnte tilfælde opstiller vi en række antagelser, der gør, at vi alligevel kan lave et match til en potentiel original rapport. Ud fra antagelserne opstilles et filter, der fjerner alle fejlrapporter fra det offentlige datasæt. For de 10 mest handlede virksomhedsobligationer i 2007 kan vi sammenligne resultatet af vores filter med det tal, som FINRA selv finder. Den relative forskel mellem vores filter og FINRA’s tal ligger alle i omegnen af 0.05%. Hvis man ignorer filteret og blot bruger data uden nogen modifikationer, vil man højst sandsynligt f˚a systematiske fejl i sine udregninger. Medianfejlen for den daglige omsætning vil ligge p˚a 7.4%, mens Amihud likviditetsm˚al vil blive mindst 14.6% for sm˚at for hver fjerde obligation. Disse fejl opn˚as, hvis man bruger datasættet ukritisk uden at gøre noget ved data inden brug. WRDS, som sikre offentlig adgang til datasættet, vedlægger samtidig ogs˚a 120 Bibliography Feldh¨utter, P. and D. Lando (2008). Decomposing Swap Spreads. Journal of Financial Economics 88, 375–405. FINRA (2003). Letter to Mr. Fazel from National Financial Services, LLC. FINRA Interpretive Letters. FINRA (2008). TRACE User Guide - version 2.2. FINRA TRACE Documentation. Friewald, N., R. Jankowitsch, and M. G. Subrahmanyam (2009). Illiquidity or credit deterioration: a study of liquidity in the US corporate bond market during financial crises. Working Paper. G¨untay, L. and D. Hackbarth (2006). Liquidity of Corporate Bond Credit Spreads and Forecast Dispersion. Working Paper. Available at SSRN: http://ssrn.com/abstract=891698. Goldstein, M. A. and E. S. Hotchkiss (2008). Dealer Behavior and the Trading of Newly Issued Corporate Bonds. Working Paper, Boston College. Goldstein, M. A., E. S. Hotchkiss, and E. R. Sirri (2007). Transparency and Liquidity: A Controlled Experiment on Corporate Bonds. The Review of Financial Studies 20(2), 235–273. Green, R., B. Hollifield, and N. Sch¨urhoff (2007a). Dealer Intermediation and Price Behavior in the Aftermarket for New Bond Issues. Journal of Financial Economics 86, 643–682. Green, R., B. Hollifield, and N. Sch¨urhoff (2007b). Financial Intermediation and the Costs of Trading in an Opaque Market. Review of Financial Studies 20(2), 275–314. Greenwood, R. (2005). Short- and long-term demand curves for stocks: theory and evidence on the dynamics of arbitrage. Journal of Financial Economics 75(3), 607–649. Han, S. and H. Zhou (2008). Effects of Bond Liquidity on the Nondefault Component of Corporate Bond Spreads: Evidence from Intraday Transactions Data. Working Paper, Federal Reserve Board. Harris, L. and E. Gurel (1986). Price and Volume Effects Associated with Changes in the S&P 500 List: New Evidence for the Existence of Price Pressures. Journal of Finance 41(4), 815–829. Harris, L. E. (1990). Statistical Properties of the Roll Serial Covariance Bid/Ask Spread Estimator. Journal of Finance 45(2), 579–590. Hegde, S. P. and J. B. McDermott (2003). The liquidity effects of revisions to the S&P 500 index: an empirical analysis. Journal of Financial Markets 6(3), 413–459. 120 Bibliography Feldh¨utter, P. and D. Lando (2008). Decomposing Swap Spreads. Journal of Financial Economics 88, 375–405. FINRA (2003). Letter to Mr. Fazel from National Financial Services, LLC. FINRA Interpretive Letters. FINRA (2008). TRACE User Guide - version 2.2. FINRA TRACE Documentation. Friewald, N., R. Jankowitsch, and M. G. Subrahmanyam (2009). Illiquidity or credit deterioration: a study of liquidity in the US corporate bond market during financial crises. Working Paper. G¨untay, L. and D. Hackbarth (2006). Liquidity of Corporate Bond Credit Spreads and Forecast Dispersion. Working Paper. Available at SSRN: http://ssrn.com/abstract=891698. Goldstein, M. A. and E. S. Hotchkiss (2008). Dealer Behavior and the Trading of Newly Issued Corporate Bonds. Working Paper, Boston College. Goldstein, M. A., E. S. Hotchkiss, and E. R. Sirri (2007). Transparency and Liquidity: A Controlled Experiment on Corporate Bonds. The Review of Financial Studies 20(2), 235–273. Green, R., B. Hollifield, and N. Sch¨urhoff (2007a). Dealer Intermediation and Price Behavior in the Aftermarket for New Bond Issues. Journal of Financial Economics 86, 643–682. Green, R., B. Hollifield, and N. Sch¨urhoff (2007b). Financial Intermediation and the Costs of Trading in an Opaque Market. Review of Financial Studies 20(2), 275–314. Greenwood, R. (2005). Short- and long-term demand curves for stocks: theory and evidence on the dynamics of arbitrage. Journal of Financial Economics 75(3), 607–649. Han, S. and H. Zhou (2008). Effects of Bond Liquidity on the Nondefault Component of Corporate Bond Spreads: Evidence from Intraday Transactions Data. Working Paper, Federal Reserve Board. Harris, L. and E. Gurel (1986). Price and Volume Effects Associated with Changes in the S&P 500 List: New Evidence for the Existence of Price Pressures. Journal of Finance 41(4), 815–829. Harris, L. E. (1990). Statistical Properties of the Roll Serial Covariance Bid/Ask Spread Estimator. Journal of Finance 45(2), 579–590. Hegde, S. P. and J. B. McDermott (2003). The liquidity effects of revisions to the S&P 500 index: an empirical analysis. Journal of Financial Markets 6(3), 413–459. Bibliography 121 Houweling, P., A. Mentink, and T. Vorst (2005). Comparing Possible Proxies of Corporate Bond Liquidity. Journal of Banking and Finance 29(6), 1331–1358. Huang, J. and M. Huang (2003). How Much of the Corporate-Treasury Yield Spread is due to Credit Risk? Working Paper, Stanford University. Huberman, G. and W. Stanzl (2005). Optimal Liquidity Trading. Review of Finance 9, 165–200. Hull, J., M. Predescu, and A. White (2004). The Relationship Between Credit Default Swap Spreads, Bond Yields, and Credit Rating Announcements. Journal of Banking and Finance 28, 2789–2811. Jain, P. C. (1987). The Effect on Stock Price of Inclusion in or Exclusion from the S&P 500. Financial Analysts Journal, 58–64. Johnson, T. C. (2008). Volume, liquidity, and liquidity risk. Journal of Financial Economics 87, 388–417. Kaul, A., V. Mehrotra, and R. Morck (2000). Demand Curves for Stocks Do Slope Down: New Evidence from an Index Weights Adjustment. Journal of Finance 55(2), 893–912. Keim, D. and A. Madhavan (1998). Demand Curves for Stocks Do Slope Down: New Evidence from an Index Weights Adjustment. Financial Analysts Journal 54(4), 50–69. Kraus, A. and H. Stoll (1972). Price Impact of Block Trading on the New York Stock Exchange. Journal of Finance 27(3), 569–588. Kyle, A. S. (1985). Continuous Auctions And Insider Trading. Econometrica 53(6), 1315–1334. Longstaff, F., S. Mithal, and E. Neis (2005). Corporate yield spreads: default risk or liquidity? New evidence from the credit-default swap market. Journal of Finance 60(5), 2213–2253. Lynch, A. W. and R. R. Mendenhall (1997). New Evidence on Stock Price Effects Associated with Changes in the S&P 500 Index. Journal of Business 70(3), 351–383. Madhavan, A. (2003). The Russell Reconstitution Effect. Financial Analysts Journal 59(4), 51–63. Mahanti, S., A. Nashikkar, M. Subramanyam, G. Chacko, and G. Mallik (2008). Latent Liquidity: A New Measure of Liquidity with an Application to Corporate Bonds. Journal of Financial Economics 88, 272–298. Mase, B. (2007). The Impact of Changes in the FTSE 100 Index. Financial Review 42(3), 461–484. 122 Bibliography Mazouz, K. and B. Saadouni (2007). New evidence on the price and liquidity effects of the FTSE 100 index revisions. International Review of Financial Analysis 16(3), 223–241. Merton, R. C. (1987). A Simple Model of Capital Market Equilibrium with Incomplete Information. Journal of Finance 42(3), 483–510. Newman, Y. S. and M. A. Rierson (2004). Illiquidity Spillovers: Theory and Evidence From European Telecom Bond Issuance. Working Paper, 1–58. Petajisto, A. (2009). Why do Demand Curves for Stocks Slope Down? Journal of Financial and Quantitative Analysis 44(5), 1013–1044. Petersen, M. A. (2009). Estimating Standard Errors in Finance Panel Data Sets: Comparing Approaches. Review of Financial Studies 22(1), 435– 480. Roll, R. (1984). A Simple Implicit Measure of the Effective Bid-Ask Spread in an Efficient Market. Journal of Finance 39(2), 1127–1139. Sarig, O. and A. Warga (1989). Bond Price Data and Bond Market Liquidity. Journal of Financial and Quantitative Analysis 24(3), 367–378. Scholes, M. S. (1972). The Market for Securities: Substitution Versus Price Pressure and the Effects of Information on Share Prices. Chicago Journals 45(2), 179–211. Schwab (2009). Schwab Bond Funds Prospectus. Schwab Funds. Shleifer, A. (1986). Do Demand Curves for Stocks Slope Down? Journal of Finance 41(3), 579–590. Tang, D. Y. and H. Yan (2006). Liquidity, Liquidity Spillover, and Credit Default Swap Spreads. Working Paper. Vanguard (2009). Vanguard Bond Index Funds Prospectus. The Vanguard Group. Wurgler, J. and E. Zhuravskaya (2002). Does Arbitrage Flatten Demand Curves for Stocks? Journal of Business 75(4), 583–608. 122 Bibliography Mazouz, K. and B. Saadouni (2007). New evidence on the price and liquidity effects of the FTSE 100 index revisions. International Review of Financial Analysis 16(3), 223–241. Merton, R. C. (1987). A Simple Model of Capital Market Equilibrium with Incomplete Information. Journal of Finance 42(3), 483–510. Newman, Y. S. and M. A. Rierson (2004). Illiquidity Spillovers: Theory and Evidence From European Telecom Bond Issuance. Working Paper, 1–58. Petajisto, A. (2009). Why do Demand Curves for Stocks Slope Down? Journal of Financial and Quantitative Analysis 44(5), 1013–1044. Petersen, M. A. (2009). Estimating Standard Errors in Finance Panel Data Sets: Comparing Approaches. Review of Financial Studies 22(1), 435– 480. Roll, R. (1984). A Simple Implicit Measure of the Effective Bid-Ask Spread in an Efficient Market. Journal of Finance 39(2), 1127–1139. Sarig, O. and A. Warga (1989). Bond Price Data and Bond Market Liquidity. Journal of Financial and Quantitative Analysis 24(3), 367–378. Scholes, M. S. (1972). The Market for Securities: Substitution Versus Price Pressure and the Effects of Information on Share Prices. Chicago Journals 45(2), 179–211. Schwab (2009). Schwab Bond Funds Prospectus. Schwab Funds. Shleifer, A. (1986). Do Demand Curves for Stocks Slope Down? Journal of Finance 41(3), 579–590. Tang, D. Y. and H. Yan (2006). Liquidity, Liquidity Spillover, and Credit Default Swap Spreads. Working Paper. Vanguard (2009). Vanguard Bond Index Funds Prospectus. The Vanguard Group. Wurgler, J. and E. Zhuravskaya (2002). Does Arbitrage Flatten Demand Curves for Stocks? Journal of Business 75(4), 583–608. TITLER I PH.D.SERIEN: 2004 1. Martin Grieger Internet-based Electronic Marketplaces and Supply Chain Management 2. Thomas Basbøll LIKENESS A Philosophical Investigation 3. Morten Knudsen Beslutningens vaklen En systemteoretisk analyse of moderniseringen af et amtskommunalt sundhedsvæsen 1980-2000 4. Lars Bo Jeppesen Organizing Consumer Innovation A product development strategy that is based on online communities and allows some firms to benefit from a distributed process of innovation by consumers 5. Barbara Dragsted SEGMENTATION IN TRANSLATION AND TRANSLATION MEMORY SYSTEMS An empirical investigation of cognitive segmentation and effects of integrating a TM system into the translation process 6. Jeanet Hardis Sociale partnerskaber Et socialkonstruktivistisk casestudie af partnerskabsaktørers virkelighedsopfattelse mellem identitet og legitimitet 7. Henriette Hallberg Thygesen System Dynamics in Action 8. Carsten Mejer Plath Strategisk Økonomistyring 9. Annemette Kjærgaard Knowledge Management as Internal Corporate Venturing – a Field Study of the Rise and Fall of a Bottom-Up Process 10. Knut Arne Hovdal De profesjonelle i endring Norsk ph.d., ej til salg gennem Samfundslitteratur 11. Søren Jeppesen Environmental Practices and Greening Strategies in Small Manufacturing Enterprises in South Africa – A Critical Realist Approach 12. Lars Frode Frederiksen Industriel forskningsledelse – på sporet af mønstre og samarbejde i danske forskningsintensive virksomheder 13. Martin Jes Iversen The Governance of GN Great Nordic – in an age of strategic and structural transitions 1939-1988 14. Lars Pynt Andersen The Rhetorical Strategies of Danish TV Advertising A study of the first fifteen years with special emphasis on genre and irony 15. Jakob Rasmussen Business Perspectives on E-learning 16. Sof Thrane The Social and Economic Dynamics of Networks – a Weberian Analysis of Three Formalised Horizontal Networks 17. Lene Nielsen Engaging Personas and Narrative Scenarios – a study on how a usercentered approach influenced the perception of the design process in the e-business group at AstraZeneca 18. S.J Valstad Organisationsidentitet Norsk ph.d., ej til salg gennem Samfundslitteratur 19. Thomas Lyse Hansen Six Essays on Pricing and Weather risk in Energy Markets 20. Sabine Madsen Emerging Methods – An Interpretive Study of ISD Methods in Practice 21. Evis Sinani The Impact of Foreign Direct Investment on Efficiency, Productivity Growth and Trade: An Empirical Investigation 22. Bent Meier Sørensen Making Events Work Or, How to Multiply Your Crisis 23. Pernille Schnoor Brand Ethos Om troværdige brand- og virksomhedsidentiteter i et retorisk og diskursteoretisk perspektiv 24. Sidsel Fabech Von welchem Österreich ist hier die Rede? Diskursive forhandlinger og magtkampe mellem rivaliserende nationale identitetskonstruktioner i østrigske pressediskurser 25. Klavs Odgaard Christensen Sprogpolitik og identitetsdannelse i flersprogede forbundsstater Et komparativt studie af Schweiz og Canada 26. Dana B. Minbaeva Human Resource Practices and Knowledge Transfer in Multinational Corporations 27. Holger Højlund Markedets politiske fornuft Et studie af velfærdens organisering i perioden 1990-2003 28. Christine Mølgaard Frandsen A.s erfaring Om mellemværendets praktik i en transformation af mennesket og subjektiviteten 29. Sine Nørholm Just The Constitution of Meaning – A Meaningful Constitution? Legitimacy, identity, and public opinion in the debate on the future of Europe 2005 1. Claus J. Varnes Managing product innovation through rules – The role of formal and structured methods in product development 2. Helle Hedegaard Hein Mellem konflikt og konsensus – Dialogudvikling på hospitalsklinikker 3. Axel Rosenø Customer Value Driven Product Innovation – A Study of Market Learning in New Product Development 4. Søren Buhl Pedersen Making space An outline of place branding 5. Camilla Funck Ellehave Differences that Matter An analysis of practices of gender and organizing in contemporary workplaces 6. Rigmor Madeleine Lond Styring af kommunale forvaltninger 7. Mette Aagaard Andreassen Supply Chain versus Supply Chain Benchmarking as a Means to Managing Supply Chains 8. Caroline Aggestam-Pontoppidan From an idea to a standard The UN and the global governance of accountants’ competence 9. Norsk ph.d. 10. Vivienne Heng Ker-ni An Experimental Field Study on the Effectiveness of Grocer Media Advertising Measuring Ad Recall and Recognition, Purchase Intentions and Short-Term Sales 11. Allan Mortensen Essays on the Pricing of Corporate Bonds and Credit Derivatives 12. Remo Stefano Chiari Figure che fanno conoscere Itinerario sull’idea del valore cognitivo e espressivo della metafora e di altri tropi da Aristotele e da Vico fino al cognitivismo contemporaneo 13. Anders McIlquham-Schmidt Strategic Planning and Corporate Performance An integrative research review and a meta-analysis of the strategic planning and corporate performance literature from 1956 to 2003 14. Jens Geersbro The TDF – PMI Case Making Sense of the Dynamics of Business Relationships and Networks 15 Mette Andersen Corporate Social Responsibility in Global Supply Chains Understanding the uniqueness of firm behaviour 16. Eva Boxenbaum Institutional Genesis: Micro – Dynamic Foundations of Institutional Change 17. Peter Lund-Thomsen Capacity Development, Environmental Justice NGOs, and Governance: The Case of South Africa 18. Signe Jarlov Konstruktioner af offentlig ledelse 19. Lars Stæhr Jensen Vocabulary Knowledge and Listening Comprehension in English as a Foreign Language An empirical study employing data elicited from Danish EFL learners 20. Christian Nielsen Essays on Business Reporting Production and consumption of strategic information in the market for information 21. Marianne Thejls Fischer Egos and Ethics of Management Consultants 22. Annie Bekke Kjær Performance management i Procesinnovation – belyst i et social-konstruktivistisk perspektiv 23. Suzanne Dee Pedersen GENTAGELSENS METAMORFOSE Om organisering af den kreative gøren i den kunstneriske arbejdspraksis 24. Benedikte Dorte Rosenbrink Revenue Management Økonomiske, konkurrencemæssige & organisatoriske konsekvenser 25. Thomas Riise Johansen Written Accounts and Verbal Accounts The Danish Case of Accounting and Accountability to Employees 26. Ann Fogelgren-Pedersen The Mobile Internet: Pioneering Users’ Adoption Decisions 27. Birgitte Rasmussen Ledelse i fællesskab – de tillidsvalgtes fornyende rolle 28. Gitte Thit Nielsen Remerger – skabende ledelseskræfter i fusion og opkøb 29. Carmine Gioia A MICROECONOMETRIC ANALYSIS OF MERGERS AND ACQUISITIONS 30. Ole Hinz Den effektive forandringsleder: pilot, pædagog eller politiker? Et studie i arbejdslederes meningstilskrivninger i forbindelse med vellykket gennemførelse af ledelsesinitierede forandringsprojekter 31. Kjell-Åge Gotvassli Et praksisbasert perspektiv på dynamiske læringsnettverk i toppidretten Norsk ph.d., ej til salg gennem Samfundslitteratur 32. Henriette Langstrup Nielsen Linking Healthcare An inquiry into the changing performances of web-based technology for asthma monitoring 33. Karin Tweddell Levinsen Virtuel Uddannelsespraksis Master i IKT og Læring – et casestudie i hvordan proaktiv proceshåndtering kan forbedre praksis i virtuelle læringsmiljøer 34. Anika Liversage Finding a Path Labour Market Life Stories of Immigrant Professionals 35. Kasper Elmquist Jørgensen Studier i samspillet mellem stat og erhvervsliv i Danmark under 1. verdenskrig 36. Finn Janning A DIFFERENT STORY Seduction, Conquest and Discovery 37. Patricia Ann Plackett Strategic Management of the Radical Innovation Process Leveraging Social Capital for Market Uncertainty Management 2006 1. Christian Vintergaard Early Phases of Corporate Venturing 2. Niels Rom-Poulsen Essays in Computational Finance 3. Tina Brandt Husman Organisational Capabilities, Competitive Advantage & Project- Based Organisations The Case of Advertising and Creative Good Production 4. Mette Rosenkrands Johansen Practice at the top – how top managers mobilise and use non-financial performance measures 5. Eva Parum Corporate governance som strategisk kommunikations- og ledelsesværktøj 6. Susan Aagaard Petersen Culture’s Influence on Performance Management: The Case of a Danish Company in China 7. Thomas Nicolai Pedersen The Discursive Constitution of Organizational Governance – Between unity and differentiation The Case of the governance of environmental risks by World Bank environmental staff 8. Cynthia Selin Volatile Visions: Transactons in Anticipatory Knowledge 9. Jesper Banghøj Financial Accounting Information and Compensation in Danish Companies 10. Mikkel Lucas Overby Strategic Alliances in Emerging High- Tech Markets: What’s the Difference and does it Matter? 11. Tine Aage External Information Acquisition of Industrial Districts and the Impact of Different Knowledge Creation Dimensions A case study of the Fashion and Design Branch of the Industrial District of Montebelluna, NE Italy 12. Mikkel Flyverbom Making the Global Information Society Governable On the Governmentality of Multi- Stakeholder Networks 13. Anette Grønning Personen bag Tilstedevær i e-mail som interaktionsform mellem kunde og medarbejder i dansk forsikringskontekst 14. Jørn Helder One Company – One Language? The NN-case 15. Lars Bjerregaard Mikkelsen Differing perceptions of customer value Development and application of a tool for mapping perceptions of customer value at both ends of customer-suppli- er dyads in industrial markets 16. Lise Granerud Exploring Learning Technological learning within small manufacturers in South Africa 17. Esben Rahbek Pedersen Between Hopes and Realities: Reflections on the Promises and Practices of Corporate Social Responsibility (CSR) 18. Ramona Samson The Cultural Integration Model and European Transformation. The Case of Romania 2007 1. Jakob Vestergaard Discipline in The Global Economy Panopticism and the Post-Washington Consensus 2. Heidi Lund Hansen Spaces for learning and working A qualitative study of change of work, management, vehicles of power and social practices in open offices 3. Sudhanshu Rai Exploring the internal dynamics of software development teams during user analysis A tension enabled Institutionalization Model; ”Where process becomes the objective” 4. Norsk ph.d. Ej til salg gennem Samfundslitteratur 5. Serden Ozcan EXPLORING HETEROGENEITY IN ORGANIZATIONAL ACTIONS AND OUTCOMES A Behavioural Perspective 6. Kim Sundtoft Hald Inter-organizational Performance Measurement and Management in Action – An Ethnography on the Construction of Management, Identity and Relationships 7. Tobias Lindeberg Evaluative Technologies Quality and the Multiplicity of Performance 8. Merete Wedell-Wedellsborg Den globale soldat Identitetsdannelse og identitetsledelse i multinationale militære organisationer 9. Lars Frederiksen Open Innovation Business Models Innovation in firm-hosted online user communities and inter-firm project ventures in the music industry – A collection of essays 10. Jonas Gabrielsen Retorisk toposlære – fra statisk ’sted’ til persuasiv aktivitet 11. Christian Moldt-Jørgensen Fra meningsløs til meningsfuld evaluering. Anvendelsen af studentertilfredshedsmålinger på de korte og mellemlange videregående uddannelser set fra et psykodynamisk systemperspektiv 12. Ping Gao Extending the application of actor-network theory Cases of innovation in the telecommunications industry 13. Peter Mejlby Frihed og fængsel, en del af den samme drøm? Et phronetisk baseret casestudie af frigørelsens og kontrollens sameksistens i værdibaseret ledelse! 14. Kristina Birch Statistical Modelling in Marketing 15. Signe Poulsen Sense and sensibility: The language of emotional appeals in insurance marketing 16. Anders Bjerre Trolle Essays on derivatives pricing and dynamic asset allocation 17. Peter Feldhütter Empirical Studies of Bond and Credit Markets 18. Jens Henrik Eggert Christensen Default and Recovery Risk Modeling and Estimation 19. Maria Theresa Larsen Academic Enterprise: A New Mission for Universities or a Contradiction in Terms? Four papers on the long-term implications of increasing industry involvement and commercialization in academia 20. Morten Wellendorf Postimplementering af teknologi i den offentlige forvaltning Analyser af en organisations kontinuerlige arbejde med informationsteknologi 21. Ekaterina Mhaanna Concept Relations for Terminological Process Analysis 22. Stefan Ring Thorbjørnsen Forsvaret i forandring Et studie i officerers kapabiliteter under påvirkning af omverdenens forandringspres mod øget styring og læring 23. Christa Breum Amhøj Det selvskabte medlemskab om managementstaten, dens styringsteknologier og indbyggere 24. Karoline Bromose Between Technological Turbulence and Operational Stability – An empirical case study of corporate venturing in TDC 25. Susanne Justesen Navigating the Paradoxes of Diversity in Innovation Practice – A Longitudinal study of six very different innovation processes – in practice 26. Luise Noring Henler Conceptualising successful supply chain partnerships – Viewing supply chain partnerships from an organisational culture perspective 27. Mark Mau Kampen om telefonen Det danske telefonvæsen under den tyske besættelse 1940-45 28. Jakob Halskov The semiautomatic expansion of existing terminological ontologies using knowledge patterns discovered on the WWW – an implementation and evaluation 29. Gergana Koleva European Policy Instruments Beyond Networks and Structure: The Innovative Medicines Initiative 30. Christian Geisler Asmussen Global Strategy and International Diversity: A Double-Edged Sword? 31. Christina Holm-Petersen Stolthed og fordom Kultur- og identitetsarbejde ved skabelsen af en ny sengeafdeling gennem fusion 32. Hans Peter Olsen Hybrid Governance of Standardized States Causes and Contours of the Global Regulation of Government Auditing 33. Lars Bøge Sørensen Risk Management in the Supply Chain 34. Peter Aagaard Det unikkes dynamikker De institutionelle mulighedsbetingelser bag den individuelle udforskning i professionelt og frivilligt arbejde 35. Yun Mi Antorini Brand Community Innovation An Intrinsic Case Study of the Adult Fans of LEGO Community 36. Joachim Lynggaard Boll Labor Related Corporate Social Performance in Denmark Organizational and Institutional Perspectives 2008 1. Frederik Christian Vinten Essays on Private Equity 2. Jesper Clement Visual Influence of Packaging Design on In-Store Buying Decisions 3. Marius Brostrøm Kousgaard Tid til kvalitetsmåling? – Studier af indrulleringsprocesser i forbindelse med introduktionen af kliniske kvalitetsdatabaser i speciallægepraksissektoren 4. Irene Skovgaard Smith Management Consulting in Action Value creation and ambiguity in client-consultant relations 5. Anders Rom Management accounting and integrated information systems How to exploit the potential for management accounting of information technology 6. Marina Candi Aesthetic Design as an Element of Service Innovation in New Technologybased Firms 7. Morten Schnack Teknologi og tværfaglighed – en analyse af diskussionen omkring indførelse af EPJ på en hospitalsafdeling 8. Helene Balslev Clausen Juntos pero no revueltos – un estudio sobre emigrantes norteamericanos en un pueblo mexicano 9. Lise Justesen Kunsten at skrive revisionsrapporter. En beretning om forvaltningsrevisionens beretninger 10. Michael E. Hansen The politics of corporate responsibility: CSR and the governance of child labor and core labor rights in the 1990s 11. Anne Roepstorff Holdning for handling – en etnologisk undersøgelse af Virksomheders Sociale Ansvar/CSR