Insider Trading Prior to Credit Rating Downgrades?

0 downloads 0 Views 477KB Size Report
Dec 8, 2013 - “Question 5e: Could a short sale on sovereign debt be considered covered by ... 1/ prior to the short sale, the short seller entered into one of the ...
Insider Trading Prior to Credit Rating Downgrades? Evidence from the European Sovereign Crisis Arturo Bris IMD, Yale ICF, ECGI December 2013

IMD, Chemin de Bellerive 23, P.O. Box 915, CH-1001 Lausanne, Switzerland. Tel: +41 21 6180111; email: [email protected].

Abstract Our paper presents strong evidence of securities lending market anticipation of sovereign bond dowgrades in Europe during the period 2008

2012. We construct a sample of bonds from 18 European countries which

includes members of the Euroland; EU, non-euro members, as well as non-EU countries. We …nd that, in the …ve days preceding a downgrade to speculative grade, securities lending demand and supply increase signi…cantly with respect to the non-downgrade periods for the same issue. We rule out that CRAs use this information to form their decision regarding the downgrade because of the special T + 3 + 1 settlement and reporting period for securities lending transactions in Europe. We …nd that the increases in lending demand and supply are only signi…cant for downgrades that surprise the market, i.e. downgrades that are not preceded by a previous downgrade or an outlook revision within the past 30 days. The di¤erence in results between these two types of downgrades is evidence against investors using publicly available information to bene…t from securities lending.

“Question 5e: Could a short sale on sovereign debt be considered covered by a repo contract executed in the days following the short sale but with the same settlement date as the short sale (e.g. a spot next repo for a T+3 cash sale)? Answer 5e: Yes, it is possible to cover a short sale by entering into a repo contract afterwards provided that: 1/ prior to the short sale, the short seller entered into one of the arrangements with a third party under article 13(1)(c) of the Regulation and article 7 of the ITS (e.g. obtained an “easy to purchase sovereign debt con…rmation” according to article 7(5)); 2/ the repo contract has an earlier or the same settlement date as the short sale, so that the delivery of the relevant sovereign debt can be e¤ected when it is due.” “Questions and Answers: Implementation of the Regulation on short selling and certain aspects of credit default swaps”, European Securities Markets Authority, September 13, 2012

I

Introduction

The Washington Summit of the G-20 Leaders on November 15 2008 marked the beginning of a regulatory revolution in …nancial markets that is still going on. On the background of such regulatory overhaul was the Lehman Brothers collapse and its devastating consequences, including the social pressure to control the incentives of …nancial institutions and their executives, to enhance transparency and accountability, and to restore trust in …nancial markets. Credit Rating Agencies (CRAs), as an important agent in these, were partly blamed for the crisis because of their inability to appropriately identify the risks of certain …rms and securities. World leaders encouraged …nancial regulators to design and enact rules that would ensure that “credit rating agencies meet the highest standards [...] and that they avoid con‡icts of interest, provide greater disclosure to investors and to issuers, and di¤ erentiate ratings for complex products.”1 Five years later, such call for stringent regulation of CRAs has mainly materialized into the speci…c provisions contained in the Dodd–Frank Wall Street Reform and Consumer Protection Act (2010), and especially in the EU Regulation on Credit Rating Agencies approved by the European Parliament on January 16, 2013. The role and impact of corporate credit ratings has been extensively studied in the literature (see Bongaerts et al., 2012; Boot et al., 2006; and Cantor, 2004 for earlier references). In the context of the recent European Sovereign Debt crisis, CRAs have been partly blamed for the fate of the Greek, Portuguese and Irish sovereign bonds. Indeed, public consultation by the European Commission highlighted some 1

Final Communiqué of the G20 Special Leaders Summit on the Financial Situation, Washington DC, November 14-15,

2008.

1

market participants’ concerns regarding the methodology and information that CRAs use to establish their sovereign ratings. Others consider instead that the criticism that CRAs precipitated the euro area crisis is largely unjusti…ed, because their downgrades merely re‡ected the seriousness of the problems that some Member States were currently facing. In that regard–it is argued–ratings followed markets rather than the opposite.2 In this paper we investigate a particular aspect of the CRAs activity— the interaction with market participants. In the last years, there have been accusations of insider trading against: S&P in the US3 , S&P and Moody’s in Italy4 , and Moody’s in Australia5 . These claims are based on potential leakages of information from employees of rating agencies or government o¢ cials to investors, who in turn exploit such information in their bene…t, before the downgrade has been properly announced to the market. While the accusations have not been yet sustained, the fear that CRAs information is made privately available and exploited by a few is a regulator’s concern. In the original draft of the CRA European Directive there was a proposal to force CRAs to communicate their ratings to the sovereigns involved three days prior to the public announcement. The proposal was disregarded by the European authorities on the ground that an extension of the period from the current twelve hours to three days would increase the risk of market abuse.6 In a recent paper, Michaelides et al (2011) analyze the stock market performance around 874 rating changes in 65 countries, and …nd signi…cant abnormal returns in the …ve days prior to the public announcement. Henry et al. (2011) …nd that short selling is signi…cantly large in the year prior to a corporate bond downgrade. These papers therefore suggest that somehow investors anticipate the behavior of CRAs. However, establishing the causality of the relationship between market variables and CRAs behavior is quite di¢ cult. Indeed, Michaelides et al (2011) conclude that, because their abnormal returns are larger in countries with weak legal systems, this is probably evidence of leakages of private information. However, Henry et al. (2011) show that, as short sales are not signi…cantly high in the days immediately preceding the downgrade announcement, it is more likely the case that CRAs use information from the market to make their decisions. Our focus is not on the equity or bond markets, but rather on the securities lending market. There are several advantages of studying this market. The …rst and most important one is that a special characteristic of the securities lending market in Europe is that transactions are known to the market at day T +4 relative 2

See House of Lords, European Union Committee, 21st Report of Session 2010–12: “Sovereign Credit Ratings: Shooting

the Messenger?” 3 New York Times, August 15, 2011, “Was There Insider Trading on S.&P.’s Downgrade? 4 Wall Street Journal, January 20, 2012. 5 The Australian, December 28, 2012. 6 See Public Consultation on Rating Agencies, European Commission, November 5, 2010.

2

to the transaction day. This is because the settlement period for loans of sovereign bonds in Europe is three days. Besides, because these trades happen over the counter, the database that is commonly used by market participants (DataExplorers) reports transactions at day T + 1 relative to the settlement date. It immediately follows that a securities lending transaction involving a European sovereign bond that happens two days before a downgrade of the issuer is not known by either other securities lenders/borrowers, nor by the rest of the market. A second advantage of the securities lending market, especially compared to the market for Credit Default Swaps, is that our data allow us to identify shifts in the supply and demand curves for a particular issue. In the spirit of Cohen et al. (2007), we can analyze market behavior and expectations beyond reporting prices and volumes. Finally, even though there is a simultaneous relationship between the CDS market and the securities lending market (Ammer et al., 2007; and Chan-Lau & Kim, 2004), the ability of the CDS community to avert in the last years a credit event to trigger swap payments (for instance in the case of the Greek debt restructuring), makes the securities lending market more appropriate for traders to bene…t from private information.7 Several authors have studied the securities lending market. Sa¢ and Sigurdsson (2011) analyze the impact of securities lending on market e¢ ciency. More recently, Kaplan et al. (2013) conduct a market experiment by exogenously increase the lending supply, and …nd no adverse impact of securities lending on security prices.8 However, most of the papers in the literature analyze the corporate bond market, not the sovereign market. The reason is that, by and large, the objective of government securities lending is not to generate a short sale, but rather to borrow money using the sovereign as collateral. Therefore, in the data is di¢ cult to distinguish a securities borrowing transaction (which is motivated by traders willing to short a sovereign bond) from a cash borrowing transaction (a repo). Additionally, lending sovereign bonds does not necessarily imply shorting sovereign bonds. Very often institutions borrow sovereign bonds in order to collateralize other transactions, including securities lending. For example, a broker-dealer can be borrowing equities from a lender who requires sovereign debt as collateral; therefore they will borrow the sovereign debt to collateralize the transaction. Other …nancial transactions that may require sovereign debt as collateral include derivatives and futures and options9 . We use the DataExplorers European government bond daily buy-side …les between June 2006 and October 2012, which contain information regarding lending amount and fees for government bonds (central government and governmental institutions) issued in 18 countries. Before June 28, 2006, the DataExplorers information is only published with a weekly frequency. We stop our sample on October 2012 because on 7

See Credit Lime, December 28, 2011: “When a short beats a CDS”. See also Du¢ e et al. (2002). 9 Securities Finance Trust Company (2012). 8

3

November 2012 the European Union enacted new rules prohibiting naked short sales of European government bonds. Data Explorers records daily bond loan information from about 100 participants representing approximately 85% of the OTC securities lending markets. Such information is voluntarily reported in exchange for having access to the consolidated information. This information includes transaction data, which DataExplorers do not make available to academics. Instead, we only have access to the total securities lending supply and demand on the day, and the average lending fees. We have securities lending information from 6; 052 bonds from 18 countries, and we complement this information with price and yield bond information from Datastream. Our …nal sample consists of more than 1:2 million of daily observations at the issue level. We also collect information regarding the history of bond rating changes in those countries by S&P, Moody’s, and Fitch. Our dataset includes changes in the outlook of the sovereign as well. There are 72 downgrade events in the sample period: 68 downgrades and 2 upgrades. We further distinguish between downgrades between di¤erent rating categories (notch downgrades), and downgrades from investment to speculative grade (junk downgrades). There are 6 downgrades of the latter type, all corresponding to only three countries: Greece (3), Ireland (1), and Portugal (2). We then generate indicators of the pre-rating periods for all of these events, and try to identify abnormal shifts in demand, supply, and utilization (the ratio of demand to supply) rates, as well as changes in lending fees around the downgrade. In order to distinguish between securities lending and cash lending transactions, we exploit the magnitude and richness of the dataset. Intuitively, repo transactions (cash lending agreements) are in the bene…t of the securities lender, so typically cash ‡ows are bene…cial to the securities borrower. Instead, securities lending transactions are in the bene…t of the securities borrower, so the lending fee will be generally higher than the interest received on the collateral. Consequently, securities lending transactions are by de…nition more costly to the borrower, so even in the absence of an indicator of the type of transaction, we can identify securities lending transactions are those where the lending fee is high enough. We therefore identify special bonds (high lending fees) and consider that these are normally the underlying asset in a securities lending transaction. We then estimate panel regressions with country controls and issue-…xed e¤ects, distinguishing between cheap-to-borrow and expensive-to-borrow bonds. We …rst …nd a signi…cant increase in utilization rates in the days prior to the announcement of a junk downgrade. Average utilization of Greek, Irish, and Portuguese bonds is 21:6% of the available lending supply …ve days before the corresponding downgrade announcement, and it increases to 23:37% one day before. It then reaches up to 25% at t = +3, and goes back to its original levels at date t = +8. Because of the settlement and disclosure dates described above, the securities lending activity reported at t + 3 is happening before the downgrade announcement. The increase in utilization rates is not re‡ected in the prices of the bonds involved. We estimate 4

cumulative abnormal returns before, at, and after the announcement of a credit rating event. We …nd that in the …ve days preceding a notch downgrade, bond prices experience a positive and signi…cant abnormal return of 0:11%. However the notch downgrade announcement is associated with a one-day abnormal return of

0:19% (signi…cant at the 1% level). In the case of junk downgrades, we do not …nd signi…cant abnormal

returns either prior or at the announcement in Greece, but they are both negative and signi…cant in the case of the Ireland downgrade. For Portugal, we …nd positive and signi…cant pre-announcement returns, and negative and signi…cant at-announcement returns. We then run panel regressions to explain the dynamics of securities lending demand, supply, utilization and fees. We control for macroeconomic conditions (euro-dollar exchange rates, the German benchmark ten-year bond yield), country-speci…c variables (the MSCI country stock market index), bond characteristics (yield, daily return, maturity controls, bond liquidity), as well as determinants of stock lending (the euro overnight rate index or EONIA, the average utilization for all secuties loans that day, CDS spreads). We also estimate regressions with issue-…xed e¤ects. We isolate securities trading transactions motivated by shorting demand by controlling for the specialness of the bond. We …nd very weak evidence that investors anticipate notch downgrades in Europe. For the whole sample of downgraded bonds there is no signi…cant movement in either lending supply or demand. In the subsample that includes only Greece, Ireland, and Portugal, we …nd that, in the …ve days prior to the downgrade, the securities lending demand reduces sign…cantly by 0:1%. By rating agency, this e¤ect is concentrated among Moody’s and Fitch’s notch downgrades: in the …ve days before the announcement, securities demand reduces

0:69% and

1:69% per day on average, respectively (signi…cant at the 1% level).

However we do …nd that securities lending demand increases signi…cantly before S&P notch downgrades ( +2:52%, signi…cant at the 10% level). The …nal impact of the notch downgrade on borrowing costs is not signi…cant before the announcement. In the case of downgrades to speculative grade (junk downgrades, which correspond to only three countries in the sample but to all three CRAs), we …nd that, prior to the downgrade announcement, both securities lending supply and demand rates increase sign…cantly. Such e¤ect is associated to a signi…cant reduction in borrowing fees. There is strong evidence of market anticipation which translates into securities lenders willing to increase their supply of bonds–which, in the expectation of a downgrade, will be settled at a lower value three days later, and securities borrowers increasing their demand for shorting purposes. Such shifts in the demand and supply curves result in lower borrowing fees and higher bond utilization rates, which is a potential indication of increasing (naked) short sales prior to the downgrade of a sovereign issue. These results have two potential explanations. One explanation is that (similar to Henry et al., 2011) investors anticipate sovereign downgrades on the basis of public information. A second explanation is 5

that some investors possess private information (acquired from CRAs or from the governments involved) regarding an upcoming downgrade. We separate out these two hypothesis by considering only those rating events which are not preceded by market news: we de…ne a new downgrade as one where no previous downgrade has happened in the previous 30 days, and where no CRA has published an outlook revision. There are 32 new notch downgrades in the sample, but only …ve junk downgrades that qualify as new downgrades: one by S&P (Portugal, January 13, 2012), three by Moody’s (Greece, June 14, 2010; Ireland, July 13, 2011, and Portugal, July 5, 2011), and one by Fitch (Portugal, January 13, 2012). The relevance of our results lies on the fact that the previous e¤ects are mostly concentrated among new ratings. This evidence contradicts the public information hypothesis, because we do not …nd signi…cantly higher securities lending activity when there are clear public signals of a potential downgrade within the previous month, but rather the opposite: securities lending is more intense when the information regarding the downgrade is more valuable because it is not known by the market. Furthermore, the behavior of securities lending markets prior to junk downgrades is consistent across CRAs. For all …ve of the new downgrades there is evidence of increasing lending supply and demand, and lower lending fees. Our results are also robust to excluding Germany for the sample, restricting the sample period to the post-2008 observations, and restricting the pre-downgrade window to exclude securities lending transactions that are settled after the downgrade announcement. “ESMA also observed cases of external communication consultants supporting one or more CRAs in the disclosure of rating actions, with an outsourcing agreement which allows con…dential rating information to be shared with the consultants before a rating action is published. ESMA is concerned that these practices may impair the CRAs’ability to directly control con…dential information on credit ratings from being: i) disclosed; and ii) used or shared for the purpose of trading in …nancial instruments, or for any other purpose other than the credit rating activity. CRAs must ensure that any con…dentiality agreements adequately protect the con…dentiality of ratings information, and that they set up appropriate controls to actively monitor and verify that there is no inappropriate use of the information.”10 The next Section of the paper describes the sovereign rating process to di¤erentiate it from the better known corporate bond rating process. In Section III we describe our bond data, securities lending variables, as well as rating downgrade information. Section IV analyzes the securities lending market and explains the di¤erences between securities lending and cash lending transactions, and how we use such di¤erence in the paper. The results of our study are presented in Section V. We conclude after a …nal section (Section 10

European Securities and Markets Authority, “Credit Rating Agencies, Sovereign ratings investigation. ESMA’s assessment

of governance, con‡icts of interest, resourcing adequacy and con…dentiality controls”, December 2014.

6

VI) that presents some robustness tests.

II A

The Mechanics of Sovereign Bond Ratings Rating Process

The process of rating a sovereign bond is somehow di¤erent from the mechanics that apply to corporate bonds (see Gaillard 2012 for an excellent description of the process). Usually sovereign ratings start with a formal request by a government to rate a particular debt issue (Standard and Poors, 2012; FitchRatings, 2012), unless the rating is unsolicited. The 2009 European Directive requires that CRAs state prominently in the credit rating “whether or not the rated entity or related third party participated in the credit rating process and whether the credit rating agency had access to the accounts and other relevant internal documents of the rated entity or a related third party” (EU Regulation 1060/2009). The directive also mandates CRAs to specify their policies regarding the issuance of unsolicited ratings11 . Among the countries in our sample, S&P issues unsolicited ratings on Belgium, France12 , Germany, Italy, Netherlands, Switzerland and the UK.13 Moody’s list of unsolicited issuers in Europe includes France, Germany, Italy, Netherlands, Switzerland, and the UK. Fitch publishes unsolicited ratings for France, Portugal, the UK, Germany, Netherlands, Switzerland, Austria, Spain, Norway and Sweden. Not surprisingly, only AAA countries (Germany, Netherlands, Switzerland and the UK) do not request ratings because these are mostly demand driven and produced by the CRAs anyway.14 That the rating is unsolicited does not mean that the sovereign does not participate in the rating process. S&P reports (House of Lords, 2011) that out of the 15 unsolicited ratings it produces, only Germany, Netherlands and Switzerland do not meet with them on a regular basis as of 2011. When a rating agreement exists, the initial process usually takes four to six weeks to complete, and includes an analysis of publicly reported …nancial information of the country, as well as meetings with 11

Such policy is similar for the three main CRAs: they usually decide to issue an unsolicited rating if it is useful to capital

markets, and if there is su¢ cient information that is publicly available to produce the rating. 12 France had a bitter reaction to CRAs in response to its downgrade by Moody’s on November 2012. Under new enacted rules, CRAs can only issue unsolicited ratings three times a year designated in advance by the CRA and may publicly release the change between the market close and one hour prior to the market open. 13 Interestingly, on February 24th 2011, S&P announced that it would start issuing sovereign ratings on the United States on an unsolicited basis. This was caused by S&P losing its rating agreement with the US government. 14 As of December 2012, other countries with unsolicited ratings were: Argentina, Australia, Cambodia, India, Japan, Singapore, Taiwan, and the US (S&P); Kenya, Mauritius, Nigeria, and Zambia (Moody’s); and Argentina, Bolivia, Australia, Ecuador, Mongolia, Turkey, Canada, Taiwan, Malaysia, Mozambique, Sri Lanka, Bulgaria, India, Estonia, Japan, Vietnam, Thailand, China, Singapore, Jamaica, New Zealand, Costa Rica, Dominican Republic, Czech Republic, Latvia, Lithuania, USA, Cyprus, Slovenia (Fitch).

7

government, central bank, and private sector representatives. The three CRAs generally notify the issuer of the rating and the outlook, providing a rationale for the decision. Fitch and S&P ratings can be appealed prior to their publication if new factual information is presented. Moody’s ratings cannot be appealed (Gaillard, 2012). In addition to notifying the issuer, CRAs typically signal in advance their intention to consider rating changes. They use negative review or watch noti…cations to indicate that the downgrade is likely within the next 90 days (IMF, 2010). Once the …nal decision has been made, CRAs have the obligation to inform the issuer of the rating, in advance of its publication. The advance copy time is of 12 hours in the EU (per the EU 2009 Directive15 ), and two hours in other regions. The information contained in these reports is con…dential and cannot be disclosed or released prior to its publication by the CRAs. In recent years, and because of the criticism that CRAs have been exposed to, ratings are usually disclosed to the public after markets close. Although CRAs are extremely opaque on the fees they charge governments for their service, the few disclosures available16 suggest that for Public Finance ratings, fees range from $50,000 to $200,000.

B

Short and Long Term; Local and Foreign Currency Ratings

S&P, Moody’s, and Fitch produce two distinct types of sovereign credit ratings: issuer ratings and debt ratings (referred to as senior unsecured ratings). Issue and issuer rating should only di¤er at the lowest rating levels. However, CRAs tend to assign debt ratings somehow automatically and based on preprescribed correlations, that is why they are of very little utility. At most rating levels, both tend to be equal (Vir Bathia, 2002). Additionally, for the same sovereign there may be several issuers with di¤erent ratings: for example, both the Kingdom of Spain and the Instituto de Crédito O…cial (the Public Lending Institute) are rated even though they both represent the same sovereign bonds. For each issuer, S&P, Moody’s and Fitch issue short-term and long-term ratings.17 They match the maturity of the corresponding bonds. Short-term ratings indicate the potential level of default within a twelve-month period. Fitch and Moody’s scales for short-term and long-term ratings are di¤erent. While long-term ratings follow the usual alphabetical scale from AAA to D, short-term ratings range from F1+ (best quality grade), then F1, F2, F3, B, C, and …nally D (Default).18 Moody’s system is somehow similar. Additionally, the three CRAs produce local- and foreign-currency ratings, depending on the currency of the corresponding issue. Local currency sovereign ratings re‡ect the CRA’s opinion on the capacity 15 16

http://www.esma.europa.eu/system/…les/L_302_1.pdf See, for instance: Standard & Poor’s Ratings Sewices U.S. Ratings Fees Disclosure, January 16, 2007. See also Becker

and Milbourn (2011) and Cornaggia et al. (2012). 17 Moody’s generates long- and short-term foreign currency ceilings as explained below, but not ratings. 18 Sometimes long-term ratings are downgraded while short-term ratings are not. On April 9, 2010, Greek long-term rating was downgraded to BBB- from BBB+, but the short-term rating was preserved at F2.

8

and willingness of a government to raise resources in its own currency to repay its debt obligations to bondholders on a timely basis (Moody’s, 2008). The foreign currency rating re‡ects the capacity of the government to mobilize foreign reserves to repay its debt on a timely basis. Moody’s and Fitch’s foreign currency ratings are constrained by the country ceiling (the highest rating that can be assigned to a foreign-currency denominated security issued by the country). Such country ceiling for foreign-currency bonds and notes is expressed on the same long-term scale. For instance, on February 2007, Portugal’s ceiling was increased from A1 to Aa3, which automatically meant a foreign currency rating upgrade of the same magnitude, while the local currency rating remained at Aa2. S&P’s tends to rate a sovereign’s local currency debt from zero to three notches above the sovereign’s foreign currency debt rating.19 The magnitude of the di¤erence depends on the development of the domestic capital market. In 2001, the local currency rating of sovereign issuers was, on average, one notch higher than the foreign currency rating (Gaillard 2012). The di¤erence has been declining over time. When the issuer is a member of a monetary union like the Euro, local and foreign currency ratings are identical. Only when membership in the Eurozone is uncertain, does S&P penalizes the foreign currency rating with one notch. This has not happen during the sample period even in the worse moments of the Euro bond crisis. Therefore, in our sample of bonds and sovereigns we do not distinguish between local and currency ratings.

III A

Sample Construction Bond Data

We …rst collect bond-speci…c information from Datastream. In particular, we select all government bonds issued by the 25 European countries available in Datastream. We drop bonds from Cyprus, Hungary, Latvia, Lithuania, Luxembourg, Poland, Slovak Republic and Slovenia because there is only one bond with data available in these countries. We collect price and yield data for the period June 2006 to October 2012 (which corresponds to the time span of the DataExplorers dataset), as well as spreads relative to the US T-Bond, volume and amount outstanding. Our sample includes bonds from countries that are members of the Euro (Austria, Belgium, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, the Netherlands, Portugal, and Spain). We also have bonds from European Union, non Euro members (Czech Republic, Denmark, Sweden and the U.K.), as well as two European countries which are not members of the EU (Norway and Switzerland). Prices are in dollars, as this is the reporting currency of the bond lending data. 19

http://www.standardandpoors.com/ratings/articles/en/ap/?assetID=1245227841398

9

B

Securities Lending Data

We use the DataExplorers European government bond daily buy-side …les between June 2006 and October 2012, which contain information regarding lending amount and fees for government bonds (central government and governmental institutions) issued in the 18 countries above. Before June 28, 2006, the DataExplorers information is only published with a weekly frequency. Data Explorers records daily bond loan information from about 100 participants representing approximately 85% of the OTC securities lending markets. Therefore we have information on the total value of the securities o¤ered and borrowed, the inventory of securities on loan held by institutions, the number of transactions happening every day as well as information regarding lending fees. For a detail description of the dataset, see Sa¢ and Sigurdsson (2011). DataExplorers makes transaction information available only to clients. To gain access to all …elds, clients must in exchange contribute their stock lending/borrowing transactions to the DataExplorers securities lending data universe . A subset is available to non-contributors (academics). Instead of transaction-by-transacion data, we have access to daily consolidated information (total demand on the day, and average lending fees for instance). Average lending fees for a given day are providing in di¤erent ways. First, DataExplorers provides us with the simple average buy side cost to borrow expressed as a fee, calculated from all transactions from the hedge fund peer group. Absolute fee data are however very scarce in the dataset. We have also compiled a value weighted average fee score (VWAF), which is computed from the average of all applicable loan fees weighted by loan value. Each average is then assigned to a fee bucket from 0 to 5, with 0 representing the least expensive fee bucket (general collateral) and 5 the most expensive (most expensive/special).1 Table 1 shows that our …nal sample contains 6; 052 bonds from 18 countries, for which we have daily data on securities lending, pricing and volume, and rating information (see below). [Insert Table 1] Most of the issues (28%) are German bonds, and 16% of bonds are French. There are also 459 bonds (8% of the sample) from the UK. Table 1 shows as well that 1; 123 bonds (19% of the sample) are PIIGS (Portugal, Italy, Ireland, Greece, and Spain) bonds. These are bonds frequently downgraded during the sovereign bond crisis. Speci…cally, 187 bonds in the sample (corresponding to Greece, Ireland, and Portugal) are downgraded to junk status at some point during the sample period. Table 2 reports the number of observations by country and year. We have 3; 418; 846 daily observations in total and 189; 936 on average per country. 1

For a good explanation of lending fees, see D’Avolio (2002). Specials can earn revenue of anything between 100 bps to

6000 bps or higher on an annualized basis (Securities Finance Trust Company, 2012)

10

[Insert Table 2] In Table 3 we report bond lending information. We compute a relative lending demand for an issue as the ratio of the value of current inventory on loan from bene…cial owners, divided by the total market value of the issue (both in dollars). Similarly, a measure of supply is the ratio of the active inventory value (current inventory available from bene…cial owners less what is deduced to be temporarily restricted from lending2 ) to the total market value of the issue. The ratio of lending demand to lending supply is referred to as utilization, and measures how much of the lending supply of a particular issue is indeed on loan. Active utilization sharpens this ratio by removing inventory which is not readily available to be borrowed. These three variables are in parentheses. We also report lending fee scores (0=cheap to lend/borrow; 5=expensive to lend/borrow or special )

[Insert Table 3] Its noteworthy that the average lending demand (bonds actually on loan) for all issues in our sample is 2:85%. This compares to an average on loan value of 0:58% for stocks in the Cohen et al. (2007) sample. Similarly, the average lending supply is 13:89%, and Sa¢ and Sigurdsson (2011) report 8% average weekly lending supply for international stocks. While this is certainly an indication that sovereign bonds are more heavily lent than stocks, most of these transactions could well be representing repo transactions, as we describe in detail in Section IV, and collateralizations of short sales of equities. This is also consistent with the UK (24:59%) and Switzerland (23:82%) displaying the largest lending supply in the sample. Notwithstanding, we observe that, while in 2006 issues of Germany, the UK, and Switzerland have higher utilization rates than issues of Greece, Portugal, and Ireland (the three countries whose bonds are downgraded to junk during the sovereign crisis), the situation is the reverse by 2012. Utilization of Greek issues increases from 33% to 63% in this period; the increase is from 15% to 46% for Ireland and from 41% to 52% for Portugal. These increases are the result of declines in both lending supply and lending demand, but with lending demand decreasing at a lower rate during the period. With respect to fees, lending becomes more expensive for Greek, Irish, and Portuguese issues. Note that DataExplorers classi…es fees into undisclosed buckets, so it is di¢ cult to give an absolute magnitude of these costs. In Table 4 we report fee scores by country, and show that only 0:1% of the observations in the sample correspond to special bonds. D’Avolio (2002) reports that the percentage of stocks on special in his sample is 9%. [Insert Table 4] 2

For example, where securities are held in too small parcels or have been restricted by the bene…cial owners.

11

Only 61% of Greek bonds, to cite a severely downgraded country, are cheap to lend/borrow, compared to a 95% of UK issues. No issues from Switzerland, the UK, or France are the in highest fee bucket, but 1:3% of Greek, 1:1% of Portuguese, and 0:3% of Irish issues are. To provide a comparison with the previous literature, we have computed the average fees per fee bucket, even though there are very few observations with available data on actual fees (2; 512 observations). These …gures are reported in the bottom row of the Table. Fees for cheap issues are 54 bps on average, and increase to 533 bps for the very expensive ones. If we consider as specials those issues with fees above 100 bps (which is equivalent to being in buckets 2 to 5), then the percentage of special issues in Greece, Ireland, and Portugal is 19, 22, and 19 respectively during the period.

C

Rating Data

In this Section we describe the collection process for ratings and downgrades. Section II explains that, for each sovereign and CRA, there is no single rating. First of all, ratings can be di¤erent for di¤erent bonds issued by the same country (which is especially true for di¤erent maturities). Moreover, S&P and Fitch produce short-term and long-term ratings, as well as local- and foreign-currency ratings and their combinations. Moody’s produces foreign currency ceilings, as well as long-term and short-term ratings. We obtain individual issuer rating information from S&P, Moody’s and Fitch. We con…rm with data at the issue level from S&P Rating Xpress that all bonds issued by a given country with similar maturity and currency are given the same rating. Moreover and as described above, in our sample countries there is no di¤erence in the sample period between local and foreign currency ratings.3 We are therefore con…dent that we can use only two ratings (short term and long term) for all ratings of a sovereign by the three CRAs. We subsequently match issues and ratings (long-term and short-term) by country depending on whether the maturity of the bond exceeds …ve years or not.4 We observe that CRAs consistently downgrade all bonds of a given country and similar maturity at once. We additionally obtain information on outlook changes for the three CRAs. For each of the rating agencies, we generate a set of rating variables. First we record the date at which any country’s rating (irrespective of whether short/long term or local/foreign currency) changes. Similarly for outlook changes. We distinguish between notch changes and junk downgrades5 (the latter only applicable to Greece, Ireland, 3

In the sample countries, the last episode when domestic and foreign currency Moody’s ratings di¤ered was August 1999,

when Sweden’s foreign currency rating was Aaa and the local currency rating was Aa1. In the case of Fitch, it was September 2000 (Denmark, with foreign rating AA+ and local rating AAA). 4 Note that within a particular bond, the relevant rating becomes the short term rating as soon as the date to maturity equals …ve years. 5 For S&P and Fitch, a downgrade to junk status means that the issuer ratings changes from BBB- to BB+. For Moody’s it means a change from Baa3 to Ba1.

12

and Portugal during the sample period). We classify downgrades (whether notch or grade changes) into unique (if there has not been a previous grade change by any of the three rating agencies in the previous thirty trading days) and not-unique. We also construct, from each bond and rating change, an indicator that equals one when, within the previous 30 days, there has been an outlook change. In order to accurately measure the date of a rating change, we use Factiva to check, for each bond rating change that we identify, when the announcement by the rating agency takes place (usually when markets are closed). Table 5 shows the entire history of rating episodes, including changes (or con…rmations) of investment outlooks by the three CRAs. [Insert Table 5] There is no CRA action between June 2006 and October 2012 in Denmark, Norway, Sweden, and Switzerland. In the Table we distinguish between notch downgrades (from AA+ to AA, from Baa2 to Baa3 for instance), and downgrades to junk status (from BBB- to BB+, from Baa3 to Ba1). Only three countries (Greece, Ireland, Portugal) are downgraded to junk in that period. Greece is downgraded by the three CRAs (…rst by S&P, and subsequently by Moody’s and Fitch), as it is Portugal (…rst by Moody’s, then by S&P and Fitch); Ireland is downgraded only by Moody’s. These are obviously the countries with most rating events, in addition to Spain and Italy. There are also very few upgrades, all in the Czech Republic. In Table 6 we consolidate rating events by CRA, and depending on whether they are notch downgrades, downgrades to junk status, or upgrades. [Insert Table 6] There are in total 62 notch downgrades, 7 downgrades to junk status, and 2 upgrades6 . Out of these, in 32 of the notch downgrades, and 5 of the downgrades to junk status, there is no previous rating event (whether another downgrade or an outlook change). Only in 58 of the 62 notch downgrades the downgrade itself is not a follow-up of a previous downgrade (in the last 30 days) by another CRA.

D

Additional Data

D.1

Benchmark Bonds

The risks associated with lending a sovereign bond depend, among other variables, on the liquidity of the issue. Unfortunately, trading volume data for the bonds in our sample are not widely available in 6

Because upgrades are extraordinary events during the sample period, and because they all correspond to a single country,

we do not provide a particular study of these in the paper.

13

Datastream (only 1; 828 issues). To classify bonds depending on liquidity, we follow an alternative strategy. For each country, debt agencies (Bloomberg and Reuters) select a short-term and a long-term reference bond (the “benchmark” bonds) which change over time and which represent the country’s solvency both in the short and the long run. Benchmark government bonds are the most actively traded issues of government bonds among all outstanding issues. From time to time and in accordance with established rules, government bonds of maturity nearest to the benchmark tenor are chosen as the benchmark issue to represent that tenor. We obtain the time-series list of benchmark bonds per country from Datastream. In fact, the country rating assigned to each country by the agencies is therefore the rating (short term or long term) that is assigned to the corresponding benchmark bond. Additionally, government-bond markets and credit-default swap markets follow closely the developments of benchmark bonds because they are used to determine a country’s spread vs. for instance the German (benchmark) bond. Generally, the benchmark bond is the latest issue within the given maturity band; consideration is also given to yield, liquidity, issue size and coupon. D.2

Credit Default Swap Data

For each country in the sample, we collect daily data on Credit Default Swaps referenced to the country, and for di¤erent maturities (6 months, 1, 2,..., 10, 20, and 30 years). For Switzerland and the Netherlands we do not have data on CDS of 20 and 30 year maturities. We then match each bond in our sample with CDS data depending on maturities. That is, we assign the nine-year CDS spread to a bond with a nine-year-to-maturity at the time of the observation. We match bonds with maturities between 10 and 15 years to the 10-year CDS; bonds with maturities between 16 and 25 years to the 20-year CDS; and bonds with maturities longer than 25 years to the 30-year CDS.7 DataExplorers (2011) has analyzed the relationship between securities lending costs and CDS spreads and found that sometimes CDS spreads lead securities lending costs, and sometimes it is the opposite.8 D.3

Other Variables

We include in our cross-sectional analysis several …nancial indicators at the daily frequency. These include the MSCI country stock market indices for the countries in the sample; the yield on the 10 year German benchmark bond; the euro-dollar exchange rate; and the euro overnight index average (EONIA).9 Many repo transactions have repo rates that are linked to an overnight index such as EONIA (for instance the 7

For Switzerland and Netherlands we match bonds with maturities of 10 or higher to the 10-year CDS. http://www.dataexplorers.com/sites/default/…les/Research%20Note%20%234%20EU%20Sovereign%20Bond%20Borrow%20Costs. 9 EONIA is a measure of the e¤ective interest rate prevailing in the euro interbank overnight market. It is calculated as 8

a weighted average of the interest rates on unsecured overnight lending transactions denominated in euro, as reported by a panel of contributing banks. See http://www.ecb.int/home/glossary/html/act4e.en.html#189

14

repo rate could be EONIA+10 bps). As we discuss below, the repo market is an important determinant of the sovereign securities lending market. As EONIA increases, securities lending is more costly (cash lending is more pro…table). On May 19th 2010, The German …nancial regulator, BaFin, announced unilateral regulation of credit derivatives which included a ban on naked short-selling and naked credit default swaps on Eurozone sovereign debt.10 Such a prohibition must have a severe impact in sovereign bond lending markets as any short sale of bonds from the Eurozone now requires a prior loan. However, BaFin could only impose the law in German exchanges. We therefore generate a dummy variable that equals one if the bond issuer is a Euro member and the observation is from after the prohibition date, and zero otherwise. A careful perusal

Supply Demand Utilization Fee Score Supply Demand Utilization Fee Score

Country is a Euro member

of the data (below) shows no clear pattern of e¤ects.

No

Before Ban 17.4% 5.5% 19.7% 15.6%

After Ban 17.3% 2.4% 14.8% 14.3%

12.0% 3.0% 18.7% 37.2%

14.2% 2.0% 15.0% 33.9%

Yes

Figure 1: Lending Supply and Demand Around the German Ban

IV A

Securities Lending of Government Bonds Securities Lending vs. Repo Transactions

A securities lending agreement consists of a lender transferring a security (equity, corporate bond, or government bond) to a borrower, who in exchange pays a lending fee. The borrower in turn posts collateral to guarantee the future delivery of the borrowed instrument, and the lender must pay interest on such a collateral (the lending fee is usually netted against such interest into what is called the rebate rate). Typically, because securities lending transactions are in the interest of and initiated by the borrower, they result in cash‡ows which are positive for the lender. That is why, with few exceptions, the rebate is negative (that is, the lending fee exceeds the interest rate on the collateral). This is usually not the case when the 10

New York Times, May 19 2010. The ban on naked short sales in certain shares and government bonds was introduced

(§ 30h of the German Securities Trading Act, WpHG) by means of the German Abusive Securities and Derivatives Trades Prevention Act from July 21, 2010.

15

downward price pressure on a security is so intense that the demand for borrowing securities is too low. The borrower will then sell the borrowed security in the market, establishing a short position. In the case of government bonds, the borrower could use the sovereign to guarantee a short position in another security (a corporate bond of a …rm from that country for example). Or can hold the sovereign for hedging purposes. In a securities lending transaction the lender can call the deal o¤ at any time by requesting the return of the security from the borrower, in which case the borrower will need to buy the security back from the market. However, until this happens, if the value of the security ‡uctuates in the market, so does the collateral that the borrower needs to post to the lender. Both the lending fee and the interest rate are contracted upon at the time of the transaction. Besides cash collateral, a bond lending transaction collateralized by other securities is also possible (securities lending against non-cash collateral). Bond loans which are not collateralized by cash are not very frequent (DataExplorers, 2011). They often form part of an arbitrage (between two similar bonds) or a hedge (for example, shorting the government curve component of a corporate bond to create a pure long exposure to credit spreads). In Asquith (2013), corporate bond loans which are collateralized by cash represent 99:6% of the sample.11

Figure 2: Anatomy of a Securities Lending Transaction When there is a negative sentiment regarding the value of the sovereign bond, lending demand will increase and so will do the lending fee that borrowers will be willing to pay. For securities lenders, an increase in the lending fee is an indication that the borrower is willing to sell the sovereign short for a pro…t, and therefore the value of the bond when the securities loan is returned will be lower. Note what would happen if a securities borrower receives a signal regarding a drop in the value of the sovereign bond (for instance because of an impending downgrade). The securities lending demand for that bond will increase, as well as the lending fee. The lending supply may increase (if the securities lender responds to the higher return on the loan) or decrease (because the negative signal conveyed by the 11

However, FSB (2012) reports that with regards to European sovereigns, about 50% of the lending transactions are

collateralized with cash (the percentage is only 20% for UK bonds).

16

increased lending demand decreases de expected value of the bond). The securities borrower will now sell the sovereign in the market to bene…t from the securities’ expected drop in price. The cost of securities lending (lending fee minus rebate) increases (possible making the sovereign special ) or decreases depending on the interaction between supply and demand. What makes government bonds peculiar is that there is a very similar transaction that results in almost identical payo¤s, but which is entirely initiated by the securities lender: a repo transaction.12 In a repo transaction (“repurchase agreement” or “repossession”), also known as “cash lending transaction”, securities lender is in fact a cash borrower. In this case the securities lender (cash borrower) receives cash after posting a government bond as collateral. The cash received may be equal to the market value of the collateral, but more typically the securities borrower (cash lender) discounts a haircut, which re‡ects the di¤erent liquidity of the collateral vs. cash. As the securities lender may now use such collateral, it has to pay interest to the securities borrower at the corresponding repo rate, which is typically lower than LIBOR given that it corresponds to a collateralized loan–and which is, in Europe, usually benchmarked to EONIA. Because the cash lending agreement is in the bene…t of the securities lender, typically cash ‡ows are bene…cial to the securities borrower. This is the main di¤erence between a repo transaction and a securities lending transaction.

Figure 3: Anatomy of a Repo Transaction When there is a negative sentiment regarding the value of the sovereign bond, securities lending supply will drop because the deteriorating quality of the collateral means a higher haircut and therefore more costly borrowing. Demand will subsequently decrease as well because securities borrower will be less willing to accept the low-quality bond as collateral. Ultimately securities lending cost (=minus repo rate) is very low. Note that a securities borrower with private negative information regarding the value of the sovereign will not enter a repo transaction, which represents accepting an overpriced bond as collateral. 12

A splendid guide to repo transactions is Euroclear (2009). The role of repo markets during the …nancial crisis is described

in detail in Gorton and Metrick (2012).

17

Finally, and of great importance for this paper, there are two main di¤erences between cash lending and securities lending transactions: 1. The collateral in a cash lending transaction is usually a high-quality (AAA-rated) bond. Although in practice repo transactions can be collateralized by lower-rated bonds, this is unusual. Collateral re-use (re-hypothecation) and collateral velocity, or the length of collateral re-use chains, can also be procyclical. According to Singh (2011), the length of “re-pledging chains”has shortened signi…cantly since the crisis. Immediately after the failure of Lehman Brothers, some securities lenders withdrew from the market entirely. However, many will only accept high-quality government bonds as collateral or cash collateral that they will reinvest at short maturities in high quality government bond repo, Treasury bills and/or in MMFs (FSB, 2012). 2. The tenor of the transaction is pre-determined, so there can be overnight repos, or term repos (typically ten days to one month). At the maturity of the security, the cash lender (security borrower) returns the collateral and receives cash+repo rate from the cash borrower (security lender).

B

Distinguishing cash from securities lending

Unfortunately, the database of securities lending transactions for government bonds includes both cash lending and securities lending transactions. Only the data that are provided to practitioners (transaction based) distinguish between the two. DataExplorers only provides consolidated data to academics (by day and security), which means that, for the same bond and in the same day there may be a repo and a bond lending transaction, that is reported jointly by consolidating volumes, and averaging fees and rebates. This explains why, as described in Section B, most of the transactions in the database correspond to German and UK bonds (AAA rated throughout the sample period). For these countries, most transactions must be repo transactions, that is, cash-lending transactions rather than securities-lending transactions. There are two potential strategies in order to isolate pure security lending transactions: one is to assume that only high-quality bonds (let say AA and better) will be used as collateral in repo transactions, meaning that the rest can be considered as pure securities lending. Even if they are, repos of lower-quality bonds become cheaper and easier to implement through the ECB, especially after 2011, as Nyborg (2011) has shown. The problem with such strategy is that it is biased towards observing more securities lending for say Portuguese or Greek bonds. Alternatively, we note that the rebate rate will be reported only when there is at least one securities lending transaction on the day. The histogram below plots rebate rates for the non-missing observations: In the sample, rebates are only available in 2; 512 issue-days, which leaves us with very few observations where the rebate is observable: 18

2007 2009 2010 2011 2012 Total

UK

SPAIN

PORTUGAL

NETHERLANDS

ITALY

IRELAND

GREECE

GERMANY

FRANCE

BELGIUM

AUSTRIA

year

Figure 4: Histogram: Rebate Rates (N=2,512)

16 1

2 2

6 101 101

38 348 348

3 3

2

13 13

0

6 94 80 80

7 7

102 70 70

3 56 85 85

0

Figure 5: Data Availability on Rebate Rates These are almost for sure securities lending transactions, but there are too few. We say almost surely, because on occasions, buyers will seek a speci…c security as collateral in the repo market. The security is said to go on special in the repo market when the repo rate becomes negative and therefore resembles a rebate rate. This happens if a certain security is in very high demand. Then the cash lender (securities borrower) will compete by o¤ering cheaper than their competitors, so the borrowing fee may be large enough to force the repo rate on that security so low as to make the di¤erence negative. The situation is not uncommon in equity repos and becomes more likely for all types of securities when interest rates in general are very low. However, this phenomenon is certainly impossible to happen for bonds in the higher risk category. To isolate pure security lending transactions, we use instead the following strategy: DataExplorers compiles a variable (DCBS in the dataset) de…ned as “Data Explorers Daily Cost of Borrow Score; a number from 1 to 10 indicating the rebate/fee charged by the agent lender based on Data Explorers proprietary

19

benchmark rate, where 1 is cheapest and 10 is most expensive”. This is similar to the fee score described earlier. The 10 cost buckets are not of the same size, with most of the observations falling into the …rst one, when it is not missing. For borrow scores greater than one, the bond is likely to be a special bond. Daily Cost of Borrow Score Cheap 1 2 3 4 5 6 7 8 9 Expensive 10 Missing Total

Observations 448,135 34,971 4,276 984 263 53 43 28 47 3 2,957,586 3,446,389

Percent 13 1.01 0.12 0.03 0.01 0 0 0 0 0 85.82 100

Figure 6: Distribution of Borrowing Cost In fact, the correlation between DCBS and the rebate is

0:77, suggesting that DCBS is high (higher

than one) in securities lending transactions. Therefore, we identify securities lending transactions as those for which the variable DCBS is 2 or more, which is at least available for 15% of the observations in the sample. The table below reports the available observations where the cost of borrowing index satisfy this condition.13 As it can be seen, the …nding that most observations correspond to Greece, Portugal and Ireland is consistent with the hypothesis that these are indeed securities lending transactions (there are only 23 bond/days in the UK).

Hence, when the cost of borrowing index is 2 or more, we deem the observation as special or “Expensive to Borrow”and should therefore considered a stock lending transaction, and a repo transaction otherwise. In case of German bonds, they also became special in the last years because of the ECB haircut policy (see Nyborg, 2011). Therefore we perform robustness checks (Section ??.A) to eliminate the impact of German bonds from our results. 13

There are also a signi…cant number of German bonds in the sample of "Expensive to Borrow" issues. The reason is that,

at the onset of the European crisis, the demand for German bond-collateralized repos was so high that repo rates became negative. See Financial Times, Nov. 28, 2011: http://ftalphaville.ft.com/2011/11/28/767721/the-german-bond-market-is-allabout-buy-and-hold/

20

Country AUSTRIA BELGIUM CZECH REPUBLIC DENMARK FINLAND FRANCE GERMANY GREECE IRELAND ITALY LUXEMBOURG NETHERLANDS NORWAY PORTUGAL SPAIN SWEDEN SWITZERLAND UK Total

2006 4 0 0 2 2 3 2 0 1 2 0 3 3 2 1 1 0 0

2007 5 0 0 0 0 10 11 7 0 15 0 3 4 3 2 0 0

2008 175 41 34 39 71 314 669 139 48 127 0 178 38 56 117 45 21 35

2009 30 0 3 9 1 57 214 5 0 4 0 29 21 3 10 4 0 0

2010 3 0 16 6 0 62 195 145 10 22 0 18 26 8 11 7 0 0

2011 13 9 16 0 34 62 358 513 204 24 0 10 20 84 56 5 0 0

2012 0 0 16 0 0 31 184 127 135 7 0 56 21 399 138 23 0 0

Total 230 50 85 56 108 539 1,633 936 398 201 0 294 132 556 336 87 21 35

26

60

2,147

390

529

1,408

1,137

5,697

Figure 7: Number of Observations with High Fees

V A

Results Univariate Results

In Table 7 we report initial results on securities lending data corresponding only to the three countries that are downgraded to junk status in the sample period (Greece, Ireland, and Portugal). Panel A of Table 7 reports average utilization rates in a window of t =

5 to t = +10 relative to the rating downgrade

announcement. We provide results by individual downgrade. There are six of those, as per Table 6, one of them simultaneous for two CRAs (Portugal was downgraded on January 13, 2012, by both S&P and Fitch). [Insert Table 7] Average utilization is 21:6% of the available lending supply …ve days before the downgrade announcement, and it increases to 23:37% one day before. It then reaches up to 25% at t = +3, and goes back to its original levels at date t = +8. Note that, because of settlement and disclosure dates, the securities lending activity reported at t + 3 is happening before the downgrade announcement. To get a better understanding of these …ndings, Panel B of Table 7 reports average utilization relative to non-downgrade dates. To that end, we construct, for every issue in our sample, a time-series average of utilization rates, excluding rating events, that is, 15 day periods around downgrade announcements (whether notch downgrades or downgrades to junk status). The time series average excludes observations before January 1, 2009, to exclude the abnormal behavior of securities lending around the Lehman crisis. We then compute the ratio of the observed utilization rates on rating periods, relative to the time-series averages of the variables. Our results show that, in the pre-downgrade period, average demand is 3%, 21

25.5% 25.0% 24.5% 24.0% 23.5% 23.0% 22.5% 22.0% 21.5% 21.0% -5

-4

-3

-2

-1

0

+1

+2

+3

+4

+5

+6

+7

+8

+9

+10

Figure 8: Average Utilization Around Downgrades to Junk Status 1%, and 9% higher than in no-rating periods in Greece, Ireland, and Portugal respectively. In the window [t = 0; t = +3], these percentages increase to 5%, 30%, and 8%. These results suggest that, even if the increase in lending activity when a downgrade is announced is not large relative to the previous …ve days, we …nd that securities lending is overall intensive in the few days before, at, and after a downgrade announcement.

B

Event Study around Bond Downgrades

Several papers have studied the impact of bond downgrades on stock markets. Brooks et al. report that sovereign rating downgrades have a negative impact on stock markets when they are announced (a one-day abnormal return of 197 basis points). Michaelidis et al. (2012), studying stock markets from 65 countries, …nd evidence that country stock indices moves before the public announcement of a sovereign rating downgrade, with a weaker reaction at the announcement and a mild correction after the event. Arezki, Candelon, and Sy have examined the spillover e¤ects of selected European sovereign rating downgrades during the 2007–2010 period. Their main …nding is that sovereign rating downgrades impact not only the …nancial markets in the country that was downgraded but also other euro area countries. These papers analyze the impact of downgrades on equity, not bond prices. In this section we perform an event study to analyze the impact of sovereign downgrades on issue prices. As market benchmarks, we use the Markit IBoxx benchmark indices for Euro-denominated sovereign bonds. These indices include investment grade, …xed income, sovereign issues in Euro. We obtain daily returns on the benchmarks from Datastream, for …ve di¤erent maturity groups: from 1 to 3, 3 to 5, 5 to 7, 7 to 10, and more than 10. We match each issue in our sample with the corresponding maturity benchmark, and estimate market model regressions of daily sovereign return on the benchmark, using daily observations for each bond and in the period June 2006-December 2007, excluding days t =

5 to t = +10

relative to any rating event, if there is one. We compute abnormal returns for a window of

22

5 to +10

relative to the downgrade event date with the estimates from the previous regressions, for all downgrades in the period January 2008-October 2012.14 Table. First we distinguish between notch downgrades (from AA to AA-, or from A to BBB for instance), and “junk” downgrades (from investment grade to junk). Note that junk downgrades have only happened for Ireland, Portugal, and Greece. These downgrades may correspond to any of the three rating agencies. Sometimes rating downgrades overlap, in which case we only consider the nearest future downgrade for each day. [Insert Table 8] In general we …nd that notch downgrades are associated to positive pre-announcement e¤ect (+0:11%, signi…cant at the 5% level), a negative and signi…cant e¤ect at the announcement day ( 0:19%, signi…cant at the 1% level), and an insigni…cant price impact afterwards. For most countries the CARs in the preannouncement period are positive (and signi…cant, except for Greece). Only Portugal ( at the 5% level) and Spain (

0:28%, signi…cant

0:38%, signi…cant at the 1% level) display negative returns.

CARs around downgrades to junk are reported for only three countries. We …nd insigni…cant preannouncement returns for Greece. In Ireland, the CAR over the …ve-day window is

1:86% (signi…cant at

the 5% level). Surprisingly, pre-announcement returns are positive and signi…cant for Portugal ( +0:11%). At the announcement day of a junk downgrade, Ireland and Portugal display negative returns ( and

0:84%

0:39%, both signi…cant at the 5% level or more); Greece’s announcement e¤ect is being insigni…cant.

Post-downgrade, bond prices drop in Greece (

2:28% in a ten-day window) and Portugal (

1:43%), both

signi…cant at the 1% level, and increase in Ireland ( +2:27%, signi…cant at the 10% level).

C

Cross-Sectional Regressions

C.1

Variables

In this table I regress stock lending variables on natural explanatory variables. For the remaining tables, the endogenous variables are de…ned as follows: Supply measure: de…ned as the active inventory value of bene…cial owners (in dollars), divided by the total market value of the securities. The Active inventory value is the current inventory available from bene…cial owners less what is deduced to be temporarily restricted from lending, e.g., where securities are held in too small parcels or have been restricted by the bene…cial owners. That is, the demand variable measures how much of a bond is available for loan. 14

One di¢ culty of the event study is that it is very di¢ cult to de…ne the "pre-event" dates, because downgrades overlap

and follow each other over time. That is why we follow the strategy of estimating market model regressions using 19 months of data at the beginning of the sample, and eliminating those when estimating abnormal returns.

23

Demand measure: de…ned as the value of current inventory on loan from bene…cial owners, divided by the total market value of the securities. It represents how much of a security is currently taken on loan. Active Utilization: a ratio of the previous two, Demand value as a % of the realistically available supply (BO On Loan Value / Active BO Inventory Value) Lending Fee. This is very scarce in the data, typically available only for 20% of the sample. As explained in Section B, the lending fee is expressed as an index that varies between one (cheapest to borrow) and …ve (most expensive to borrow). We estimate regressions of the form: yt+1 =

+

n X

j xjt

(1)

j=1

where y is one of the variables below, and xj , j = 1; :::n is a set of explanatory variables, including credit rating downgrade indicators. The reason why we date securities lending variables in t + 4 is that, in Europe, securities lending transactions can be settled up to day T + 3 relative to the loan date. Moreover, DataExplorers discloses its data with a one-day delay. This means that observations at any day t indeed represent securities lending transactions that may have occurred between days t

4 and t

1. We are

conservative enough to choose a one-day delay to make sure that the reported securities lending transactions we use have happened before a sovereign bond downgrade being announced at t = 0, but not known to the market yet. The set of controls xj includes: CDS spread in percent, as a measure of the riskiness of the bond. The relationship between CDS spreads and securities lending activity has been analyzed in several papers (see Gorton and Metrick, 2012). Daniels and Jensen (2005) analyze the relationship between corporate CDSs and corporate bonds. They …nd that credit rating is a signi…cant determinant of both CDS spreads and credit spreads for investment grade issues and especially for non-investment-grade issues, but not the reverse. In our paper we use CDS spreads as a determinant of securities lending decisions even though there may be a simultaneous relationship between the CDS market and the securities lending market. This could happen if securities lending leads bond prices and these cause trading in the CDS market. There is supporting evidence in the literature for a simultaneous relationship: see for instance Ammer et al. (2007), and Chan-Lau & Kim (2004). That is why in our regressions we lag the CDS spread by one day. In the corporate CDS market, Aktug et al (2001), Zhu (2006), and Blanco et al. (2005)

24

have shown that CDS spreads lead bond spreads.15 Interest Yield of the bond at time t; Bond Daily Returns, computed from Datastream bond clean prices (which take into account coupon payments); EONIA (Euro OverNight Index Average), which is a measure of the cost of money and therefore of the cost of borrowing the stock; Euro-Dollar Exchange Rate, as an indicator of European-wide risk; MSCI Country Stock Market Index. These variables are explained in Section III.D.D.3 Ban on German Shorts dummy. On May 19th 2010 Germany prohibited shorting on German bonds, and on Germany territory. The prohibition de…nitely had an impact on stock lending activity of these bonds (and therefore on the repo market) although, because of the geographical scope of the prohibition, the impact was only mild. We consistently …nd that the German ban increases utilization rates, probably because most of the lending activity in Germany moved to London. Note in any case that German bonds are generally repoed, not lent. The daily average utilization in all bonds, to control for average stock lending activity in the market which responds to market-wide changes in investor sentiment. The Germany Benchmark 10-year Bond yield A dummy that equals one if the bond is a benchmark bond (see Section III.D.1). Bond maturity. We construct three dummy variables, one for each maturity (short term, medium term, and long term bonds), and estimate the e¤ect of the last two in the regressions. C.2

Determinants of Lending Supply and Demand

Table 9 reports the …rst set of cross-sectional regressions. These show that the previous-day CDS spread is a signi…cant determinant of both securities lending demand and supply. A 100 bps increase in CDS spreads reduces the lending inventory by 0:3% of the total capitalization of the bond (signi…cant at the one percent level), and the lending demand by 0:05% basis points (signi…cant at the 10 percent level). The impact on utilization is signi…cant, although economically meaningless as well (less than one bps). Additionally, the higher riskiness of the issuer makes securities borrowing more expensive.16 A signi…cant determinant of lending demand is the overnight lending rates in the Eurozone: a 100 bps increase in EONIA increases lending demand by 0:23%, and thus increases utilization by 1% (these results 15

Note in any case that, because the securities lending transactions are dated at t+1, they may represent actual transactions

occurring at t 1 as well. 16 Economically, the magnitude of this coe¢ cient is di¢ cult to gauge becase lending fees are consolidated into fee buckets which are undisclosed by DataExplorers.

25

are signi…cant at the …ve and one-percent levels, respectively) and lending fees by 10 bps. The impact of EONIA on lending demand is consistent with cash lending (and therefore securities borrowing) becoming more pro…table when interest rates increase. [Insert Table 9] Finally, we …nd that securities lending demand is higher for short-term and medium-term bonds (signi…cant at the one-percent level). The rest of variables (except for the average utilization of the day) are not statistically signi…cant. In the subsequent Tables we will always control for these variables. However we do not report their corresponding coe¢ cients because their e¤ects do not change either quantitatively or qualitatively. C.3

Bond Rating Indicators

We construct several dummy variables to isolate securities lending transactions in the period before rating downgrade announcements. Our …rst dummy variables are two indicators (Pre-Notch Downgrade and PreDowngrade to Junk ) that equal one in days t =

5 to t =

1 relative to a notch downgrade or downgrade

to junk status, respectively. Additionally, we create similar dummy variables for each of the three CRAs we study. Sometimes downgrades of either type happen so close in time that observations overlap. In that case the relevant downgrade is the earlier one.17 Even though most downgrade announcements happen after market closures, we constraint the pre-downgrade period to day t = during market hours. Also, we do not go beyond day t =

1 to avoid the possibility of disclosures

5 to make sure that all our securities lending

transactions are not known by the market at the time of the bond downgrade.18 C.4

Securities Lending Before Downgrade Announcements

In Table 10 we show the volume of securities lending prior to sovereign downgrades. We …rst construct a dummy variable (“Expensive to Borrow ” indicator), as we discuss in Section IV.B, and we interact this dummy with the downgrade indicators. We …rst …nd that, for issues of any type, both securities lending demand and supply is signi…cantly lower ( 0:5% and 0:19% respectively) during the …ve days preceding a downgrade to junk status. Note that because these are average daily e¤ects, the compounded e¤ect is economically important (the increase in utilization in the total pre-downgrade period is 1% of the total 17

For instance, Greece was downgraded by Moody’s on July 25, 2012, and by S&P on July 27, 2012. In that case, the

dummy "Pre-Notch Downgrade" equals one on July 26, but not on July 25. However, the "Pre-Notch Downgrade by S&P" equals one on July 25. 18 We in any case perform robustness tests on di¤erent event windows. See Section ??.C.

26

value of the issue). Such e¤ect is obviously concentrated among the PIIGS countries, as they are the only one downgraded during the sample period. However, the utilization of expensive-to-borrow or special bonds (which are most likely used in securities lending transactions) is signi…cantly higher (at the one-percent level). [Insert Table 10] The panel regressions also show a signi…cant increase in fees, reasonably following a demand increase. This result is statistically signi…cant for downgraded PIIGS bonds, both relative to all the other bond-days in the sample, as well as relative to all other PIIGS, non-downgrade days. As we discuss in Section IV.B, some of the expensive issues may correspond to Germany, especially in the later part of the sample. The last columns of Table 10 con…rm that it is only for downgraded countries that utilization rates increase. With respect to notch downgrades there is also a signi…cant increase in utilization rates, but not due to more securities lending. Instead, we …nd signi…cant drops in securities lending demand in the …ve days preceding the downgrade: using all observations, the drop is of 1:64% per day, and 1:94% in the subsample of PIIGS countries. C.5

The Impact of News

So, far, we have found that investors have anticipated downgrades of European issues, especially those downgrades to junk debt su¤ered by Greece, Ireland and Portugal. Nashikkar and Pedersen (2007) report similar results using a sample of US corporate bond loan data from one of the world’s largest custodian banks. They …nd signi…cant lending activity already 20 days before a downgrade to speculative grade. We now entertain three di¤erent explanations for these …ndings. The …rst hypothesis is that rating agencies use …nancial markets information to form the decision to downgrade a European issues. If investors possess valuable information regarding the solvency of a sovereign, and this information is incorporated into market prices, then bond yields and prices, as well as securities lending decisions, will be public signals that rating agencies can easily use. Section V.B shows that, on average, abnormal returns prior to notch downgrades are slightly positive and signi…cant; abnormal returns prior to junk-downgrades are insigni…cant on average and, by country, positive and signi…cant in Portugal, negative and signi…cant in Ireland, and negative but insigni…cant in Greece. It is therefore di¢ cult to reject the hypothesis that the market does not anticipate European bond downgrades–at least one can say that the only downgrade that was anticipated by the market was that of Ireland (by Moody’s, on July 2011). More importantly, it turns out that securities lending information is not observable by the market at the time of the downgrade, as we argue above. Therefore, such hypothesis is rejected by the data. 27

The second hypothesis is that investors anticipate sovereign downgrades. This can happen if CRAs signal their decisions with anticipation (for example through warnings or outlook revisions), or if CRAs actions are predictable using publicly available information. We call this hypothesis the public information hypothesis. Nashikkar and Pedersen (2007) also argue that increased short-selling before a downgrade could be an indication of (1) private information regarding the bond downgrade; (2) public information not fully re‡ected in prices; and (3) investors’hedging reason to short in connection with the downgrade. Their evidence is consistent with (2) and (3). The third hypothesis is that investors receive private information–from CRAs or the governments involved–before the public announcement of a sovereign downgrade (remember that the decision to downgrade has to be communicated to the issuer at least 12 hours before the public announcement). We call this hypothesis the private information hypothesis. Michaelidis et al. (2012) …nd that stock markets anticipate sovereign downgrades. They argue that their results seem more consistent with information leakages about the content and timing of the pending announcement. The reason is that most of the signi…cant pre-downgrade abnormal returns are concentrated among markets with low institutional quality. The last two hypotheses should lead to the same …ndings, namely that securities lending activity increases before bond rating downgrade. This is so because, in the presence of an impeding downgrade, sovereign securities lending supply increases–as a response of …nancial institutions expecting negative returns–and lending demand increases–as short sellers target the a¤ected country’s debt and borrow securities in the market. However the public information hypothesis predicts that, the more public information regarding the impending downgrade there is in the market, the more pronounced such e¤ect will be. [Insert Table 11] We try to test these two hypotheses in two steps. We …rst identify those downgrades that are not subsequent downgrades following an earlier one, at least during the last 30 days. We call these new downgrades. Greece was downgraded by Fitch on October 22, 2009, and later on December 8, also by Fitch. Subsequently S&P downgraded Greece on December 16, 2009, and also by Moody’s on December 22, 2009. The …rst two downgrades by Fitch are new under our characterization (because the previous downgrade on a Greek issue had happened); all the others are not new. To shed additional light to the distinction between the public and private information hypotheses, we also consider those downgrades that do not follow a previous warning (change to Negative Outlook ) by the rating agency (in the previous 30 days). Warnings are usually released in pre-speci…ed dates (typically quarterly) by the rating agency. There are a few notch downgrades that are new, but only …ve 28

junk downgrades that qualify as new downgrades, as shown in Table 3: one by S&P (Portugal, January 13, 2012), three by Moody’s (Greece, June 14, 2010; Ireland, July 13, 2011, and Portugal, July 5, 2011), and one by Fitch (Portugal, January 13, 2012). We therefore interact four indicators: a pre-downgrade indicator, the Expensive to Borrow dummy, a control for New downgrades, and an Outlook Change indicator. We report regression results in Table 11. Our …rst …nding is that, in the whole sample, lending supply is lower for special bonds. Consistent with this …nding, utilization rates and fees are higher (signi…cantly at the 1% level). However, there is a signi…cant additional reduction in lending supply in the …ve days that precede a bond downgrade to speculative grade: lending demand drops, and borrowing fees increase. Note that we …nd signi…cantly lower fees before the announcement of a notch downgrade. In the case of new junk downgrades with no prior warnings (no downgrade or outlook revision in the prior 30 days), we …nd strong evidence that lending supply and demand increase, resulting in a lower cost of borrowing. In the whole sample, lending supply increases by 0:39% (signi…cant at the 1% level) in the pre-downgrade period19 ; and lending demand increases by 0:11% (signi…cant at the 1% level). For the whole …ve-day period, these coe¢ cients imply that lending supply and demand increase by 2% and 0:5%, respectively. As a result, there is a signi…cant increase in utilization rates (up 0:11%). Lending fees are lower as well: they are 0:7 buckets lower in the pre-downgrade period, which represents a estimated increase in fees of 67 bps. [The table below reports all observations with available data on lending fees, classi…ed into fee buckets.

Fee Bucket Lending Fee (bps)

0 54.7

1 85.9

2 166.1

3 222.2

4 492.2

5 533.3

Figure 9: Average Lending Fees by Fee Bucket The average di¤erence in fees between to consecutive buckets is 95 bps. The shift in lending demand and supply curves that we identify is equivalent to a SOUT shift (using Cohen et al.,2007 terminology), which happens in their case when stocks see their loan fee fall but the loan quantity rise. Cheaper borrowing makes it possible for more investors to enter the market, thus relaxing a previously existing shorting constraint. They argue that SOUT predicts an immediate downward price adjustment. Table 8 reports negative abnormal returns in the post-downgrade period (which is also the time at which securities lending information is released to the market). Our results are consistent across subsamples. They are strong among Euro countries, and also among PIIGS countries. In the subsample that only includes Greek, Irish, and Portuguese issues, the supply and demand increases ( +2:09% and 0:37%) are signi…cant at the 1% level, as well as the reduction in fees ( 19

For new downgrades with no prior warnings, the full pre-rating e¤ect is the sum of the signi…cant coe¢ cients of the four

Pre-Junk Period dummies

29

0:8 buckets). The signi…cant increase in utilization rates ( 0:12%, signi…cant at the 5% level) is consistent with Figure 8. Overall, our …ndings suggest that most of the abnormal securities lending activity prior to bond downgrades in Europe is concentrated among rating announcements that surprise the market. Indeed, we do not …nd that, when the market receives public signals of an impeding downgrade (either a prior downgrade by another rating agency, or a more straightforward negative outlook on the sovereign), there is signi…cant securities lending activity (there is indeed a drop in lending demand of about

0:29%). This is quite

intuitive because the public signal should have impacted securities lending at the time of the news release, thus making any future trading on public information unpro…table. Therefore, our results provide support against the public information hypothesis. In line with Nashikkar and Pedersen (2007), the remaining explanation for our results–that traders possess private information regarding an impeding downgrade–is not per se an indication of illegal insider trading20 in the securities lending market. Traders may well be using public information that it is ignored by the market; we cannot rule out that information about securities lending transactions are privately obtained by CRAs. Henry et al. (2011) study whether short sellers anticipate corporate bond rating downgrades. They …nd signi…cant short selling activity during the one year before the downgrade. They …nd no evidence of increased short selling in the days close to the downgrade announcement, which contradicts the hypothesis that short sellers possess private information. However, they …nd that short selling is more severe for …rms which went through a previous downgrade, which they interpret as evidence that short sellers are informed traders who anticipate rating downgrades. The question remains as why the market does not use the information contained in short selling activity. 20

Section C, Article 3 of Regulation (EC) No 1060/2009 of the European Parliament and of the Council of 16 September

2009: “Credit rating agencies shall ensure that persons referred to in point 1: [...] (b) do not disclose any information about credit ratings or possible future credit ratings of the credit rating agency, except to the rated entity or its related third party;” Similarly, the Securities Exchange Act of 1934, § 240.17g-4 states: “Prevention of misuse of material nonpublic information. (a) The written policies and procedures a nationally recognized statistical rating organization establishes, maintains, and enforces to prevent the misuse of material, nonpublic information pursuant to section 15E(g)(1) of the Act (15 U.S.C. 78o– 7(g)(1)) must include policies and procedures reasonably designed to prevent: [...] (3) The inappropriate dissemination within and outside the nationally recognized statistical rating organization of a pending credit rating action before issuing the credit rating on the Internet or through another readily accessible means.”

30

C.6

Results by Rating Agency

We further explore the robustness of our results by analyzing the market for borrowing sovereign bonds depending on which CRA downgrades a particular issue. We split our dummy variables into three di¤erent categories, one per CRA, and present results in Table 12 [Insert Table 12] Let us …rst focus on junk downgrades. The …rst set of regressions shows a joint estimation for all three CRAs. Because the Fitch and S&P junk downgrade of Portugal happened the same day, Fitch coe¢ cients cannot be estimated. In the Table below we report total e¤ects from regressions 5 to 16, which we compute by adding the coe¢ cients of the two downgrade dummies per CRA. All coe¢ cients are signi…cant at the 5% level or better:

S&P Moody's Fitch

Supply 1.282 2.670 0.029

Demand Utilization Fees 0.262 0.163 -1.092 -0.080 0.412 -0.334 0.236 0.143 -0.930

Figure 10: Total E¤ects of Junk Downgrades We …nd consistent results with the ones in the previous Section. Lending demand increases consistently across CRAs, while lending demand slightly decreases prior to Moody’s downgrades, and increases elsewhere. Utilization is consequently higher. The shifts in the demand and supply curves resulting in lower lending fees are again consistent with supply-driven securities, cheaper lending transactions. In the case of notch downgrades, results are qualitatively di¤erent. In the case of S&P, we …nd marginally weak increases in securities lending demand (1:54% increase, signi…cant at the 10% level), which may suggest some market anticipation. However, in the case of Fitch and Moody’s downgrades, both demand and supply are lower in the pre-downgrade period, which is inconsistent with market anticipation.

VI

Robustness Tests

In this section we test the robustness of our previous results by performing additional tests which involve: removing potential contamination of German bonds’repo transactions in the later part of the sample; excluding the pre-2009 period from the sample; and reducing the event window to only pre-rating downgrade announcement days.

31

A

Excluding Germany

In Section IV.B we discuss that after 2009, German bonds became special because of an extraordinarily higher demand from market participants. Such demand was the consequence of the European Central Bank policy regarding collateral and repo haircuts. As we have shown, the result was extremely high borrowing costs and negative rebates for German bonds, despite their low riskiness. Because our panel regressions compare the securities lending demand and supply for sovereigns that su¤er a downgrade, comparing them to non-downgraded bonds (potentially German), our results may be overstating the magnitude of the e¤ect of the rating event. [Insert Table 13] Table 13 reports a similar regression to Table 12, but without the observations corresponding to German issues. The number of observations drops from 1; 294; 307 to 832; 563 because German bonds represent the largest part of our original sample. In any case, we do not …nd qualitative di¤erences in our results.

B

Excluding the pre-2009 period

As a second robustness test, we exclude the observations from the year 2008 and before. One reason is that most of the bond downgrades in our study, and certainly all of the downgrades to junk status, happen after 2008. The second reason is that the short selling activity in government bonds during 2008 is intense and therefore many sovereigns, including those of AAA-rated countries, went into special, so it is again di¢ cult to benchmark our results against the normal behavior of securities lenders and borrowers. [Insert Table 14] Table 14 presents the results.

C

Di¤erent Event windows

There is no evidence other than the one coming from market participants that the T + 3 settlement window is fully exploited by securities lenders/borrowers, and certainly DataExplorers does not provide detailed transaction and settlement information. Therefore, and to take a conservative approach, we have also estimated our regressions with securities lending variables dated at t + 1. That is, we still consider the DataExplorers reporting delay of one day, which means that the pre-rating period ends at t = 0 relative to the downgrade announcement. This ensures that all securities lending transactions have for sure happened before the downgrade. [Insert Table 15] 32

Table 15 reports this results, which are still consistent with our overall picture of demand and supply increases, with a signi…cant reduction in lending fees.

D

Other Tests

We have additionally checked the consistency of the results to extending the event window before day t

5.

Note that, because downgrade events overlap sometimes, it is less appropriate in our view to attribute an increase in securities lending demand or supply to a longer pre-event window. We have also estimated panel regressions using only the subsample of PIIGS countries, as well as only the subsample that only includes Greece, Ireland, and Portugal. Finally, we have estimated the regressions separately for notch and junk downgrades. The results (not reported) do not represent a signi…cant qualitative change with respect to our major …ndings. We have also used a coarser de…nition of new downgrades, assuming that a downgrade is a surprise only when there are no public signals (a previous downgrade or an outlook revision) within the last 90 days before its public announcement. One problem with such de…nition is that we are left with only two new downgrades, which correspond both to the same country (Portugal). We …nd in any case consistent results.

VII

Conclusions

Our paper presents strong evidence of securities lending market anticipation of sovereign bond dowgrades in Europe during the period 2008

2012. We construct a sample of bonds from 18 European countries

which includes members of the Euroland; EU, non-euro members, as well as non-EU countries. We compile the history of bond downgrades (notch downgrades as well as downgrades to speculative grade) for the sovereigns in our sample and complement it with issue-, as well as issuer-speci…c information. We then focus on the securities lending market. We obtain from DataExplorers the complete set of securities lending transactions for 6; 052 sovereign bonds, with information on securities lending demand and supply, as well as borrowing costs. We …nd that, in the …ve days preceding a junk downgrade, securities lending demand and supply increase signi…cantly with respect to the non-downgrade periods for the same issue. We rule out that CRAs use this information to form their decision regarding the downgrade because of the special T + 3 + 1 settlement and reporting period for securities lending transactions in Europe. This means that a securities lending transaction occurring two days before the public announcement of the downgrade is only known to the market one day after the announcement. Therefore, the only explanation that remains for our …ndings are that some investors anticipate the downgrade. However, we …nd that the increases in lending demand 33

and supply are only signi…cant for downgrades that surprise the market, i.e. downgrades that are not preceded by a previous downgrade or an outlook revision within the past 30 days. The di¤erence in results between these two types of downgrades is evidence against investors using publicly available information to bene…t from securities lending.

34

References Rahmi Erdem Aktug, Geraldo Vasconcellos, and Youngsoo Bae, 2011, “The Dynamics of Sovereign Credit Default Swap and Bond Markets: Empirical Evidence from the 2001-2007 Period,” working paper. Arezki, R., B. Candelon, and A. Sy, 2010, “Sovereign Ratings News and Financial Markets Spillovers,” IMF Working Paper. Asquith, Paul, Andrea S. Au, Thomas R. Covert, Parag A. Pathak, 2013, “The Market for Borrowing Corporate Bonds”, Journal of Financial Economics, 107, 155-182. Becker, Bo, and Todd Milbourn, 2011, “How did increased competition a¤ect credit ratings?”, Journal of Financial Economics 101(1), 493–514. Blanco, R., Brennan, S., Marsh, I.W., 2005, “An empirical analysis of the dynamic relation between investment grade bonds and credit default swaps.” The Journal of Finance 60, 2255-2281. Cantor, Richard, 2004, “An introduction to recent research on credit ratings”, Journal of Banking & Finance 28, 2565–2573. Chan-Lau, J.A., Kim, Y.S., 2004, “Equity prices, credit default swaps, and bond spreads in emerging markets,” IMF Working Paper. WP/04/27. Bongaerts Dion, K.J. Martijn Cremers, and William N. Goetzmann, 2012, “Tiebreaker: Certi…cation and Multiple Credit Ratings”, The Journal of Finance 67, 113-152. Boot Arnoud, Todd Milbourn, and Anjolein Schmeits (2006), “Credit Ratings as a Coordination Mechanism, Review of Financial Studies” 19(1), 81-118. Brooks, R., R. Fa¤, D., Hillier, and J. Hillier, 2004, “The national market impact of sovereign rating changes,” Journal of Banking and Finance, 28, 233–250. Cohen, Lauren, Karl B. Diether, and Christopher J. Malloy, 2007, “Supply and Demand Shifts in the Shorting Market”, Journal of Finance 62, 2061–2096. Cornaggia, Jess, Kimberly J. Cornaggia, and John E. Hund, 2012, “Credit Ratings across Asset Classes: A A?”, working paper. D’Avolio, Gene, 2002, “The Market for Borrowing Stock”, Journal of Financial Economics 66, 271-306. Daniels, K.N. and M.S. Jensen, 2005, “The E¤ect of Credit Ratings on Credit Default Swap Spreads and Credit Spreads,” Journal of Fixed Income, 14, 16-33. 35

DataExplorers, “EU Sovereign Bonds: Borrow Costs and CDS Spreads”, Research Note #4, 8 August 2011. Du¢ e, Darrell, 1996, “Special Repo Rates.” Journal of Finance 51(2), 493-526. Du¢ e, Darrell, Nicolae Gârleanu, and Lasse H. Pedersen, 2002, “Securities Lending, Shorting, and Pricing”, Journal of Financial Economics 66, 307–339. Euroclear, “Understanding Repos and the Repo Markets”, March 2009. Financial Stability Board, “Securities Lending and Repos: Market Overview and Financial Stability Issues”, April 27, 2012. FitchRatings, The Ratings Process–Special Report, December 2011 Gaillard, Norbert, “A Century of Sovereign Ratings”, Springer, 2012, 206 pp. Gorton, Gary, and Andrew Metrick,2012, “Securitized banking and the run on repo”, Journal of Financial Economics 104, 425–451. Henry, Tyler R., Darren J. Kisgen, and Julie Wu, 2011, “Do Equity Short Sellers Anticipate Bond Rating Downgrades?”, working paper. House of Lords, European Union Committee, “Sovereign Credit Ratings: Shooting the Messenger? (Credit Rating Agencies - Oral evidence and associated written evidence)”, July 21 2011. Kaplan, Steven, Tobias Moskowitz, and Berk Sensoy, 2013, “The E¤ects of Stock Lending on Security Prices: An Experiment”, Journal of Finance, forthcoming. Michaelides, Alexander, Andreas Milidonis, George Nishiotis, and Panayiotis Papakyriacou, 2011, “Sovereign Debt Rating Changes and the Stock Market,” working paper. Moody’s Global Sovereign, Rating Methodology, September 2008. Nashikkar, Amrut, and Lasse H. Pedersen (2007), “Corporate Bond Specialness”, working paper. Nyborg, Kjell G., 2011, “The Euro Area Sovereign Debt Crisis: Secure the Debt and Modify Haircuts”, Swiss Finance Institute Occasional Paper Series N 11 –01. Sa¢ , Pedro A.C., and Kari Sigurdsson, 2011, “Price E¢ ciency and Short Selling”, Review of Financial Studies 24(3), 821-852.

36

Securities Finance Trust Company, “Securities Lending Best Practices, A Guidance Paper for US Mutual Funds”, 2012. Standard & Poor’s RatingsDirect, How We Rate Sovereigns, March 2012. Vir Bhatia, Ashok, 2002, “Sovereign Credit Ratings Methodology: An Evaluation”, IMF Working Paper WP/02/170. Zhu, H., 2006, “An empirical comparison of credit spreads between the bond market and the credit default swap market,” Journal of Financial Services Research 29, 211-235.

37

Country AUSTRIA BELGIUM CZECH REPUBLIC DENMARK FINLAND FRANCE GERMANY GREECE IRELAND ITALY LUXEMBOURG NETHERLANDS NORWAY PORTUGAL SPAIN SWEDEN SWITZERLAND UK TOTAL

Bonds in  Datastream 251 173 59 75 123 1008 1975 106 33 657 3 531 195 72 434 271 384 508 6858

Bonds in  Dataexplorers +  Datastream 208 154 57 61 113 958 1704 96 28 540 3 497 172 63 396 226 317 459 6052

Table 1.  Description of  the  Sample. The Datastream sample includes  bonds  from  18  European  countries  with  data available  on  price,  yield,  and  volume.  The  DataExplorers sample  of  bonds  is  based  on  the  European  government bond daily buy‐side files. The sample period is June 2006 to October 2012.  

11 12 14 18 18 16 10 28

231 271 292 282 243 256 212 540

8,146 19,544 28,155 26,858 24,469 22,856 15,168 145,196

7,722 14,639 17,156 17,744 21,009 21,249 14,326 113,845

1664 4,423 5,994 6,747 7,824 9,307 6,638 42,597

2111 4,866 6,853 7,337 7,182 7,168 4710 40,227

2537 5,782 6,611 6,791 9,519 13,610 9455 54,305

Total

43 56 57 53 59 60 34 96

UK

137 699 795 912 1,058 1,109 890 1,704

SWITZERLAND

ITALY

170 368 441 518 539 555 441 958

SWEDEN

IRELAND

22 29 34 38 53 81 58 113

SPAIN

GREECE

17 26 33 42 31 34 26 61

PORTUGAL

GERMANY

15 24 30 32 37 43 37 57

NORWAY

FRANCE

63 69 79 92 96 104 83 154

NETHERLANDS

FINLAND

71 106 146 130 111 114 92 208

LUXEMBOURG

DENMARK

Total

CZECH REPUBLIC

2006 2007 2008 2009 2010 2011 2012

BELGIUM

Year

AUSTRIA

Number of Bonds per Year and Country

2 1 2 2 2 3

57 131 248 286 310 327 216 497

39 81 55 61 73 88 72 172

22 25 27 30 37 48 33 63

52 126 178 215 221 268 214 396

87 101 85 94 93 116 85 226

28 136 137 137 180 211 119 317

149 202 207 190 172 174 109 459

1,214 2,462 2,860 3,131 3,333 3,606 2,733 6,052

Number of Observations per Year and Country 2006 2007 2008 2009 2010 2011 2012 Total

18,789 15,866 54,735 94,464 82,094 170,198 89,029 167,056 109,147 191,777 110,831 217,326 74,725 149,015 539,350 1,005,702

5,418 12,189 12,006 10,408 11,636 9,914 2231 63,802

1302 2536 2,912 3,440 3,811 2,984 1,786 18,771

26,056 51,985 54,648 50,436 48,846 51,716 35,151 318,838

46 261 425 518 386 1,636

6,860 16,536 41,080 52,436 58,914 59,108 35,281 270,215

4659 10,181 11,277 10,606 13,418 17,369 12016 79,526

2,731 5,487 5,577 6,580 8,078 9,215 5,460 43,128

5,957 19,437 30,728 38,432 43,300 44,494 35,182 217,530

9,814 17,666 16,675 14,928 16,976 20,491 13631 110,181

3,018 16,578 30,437 28,416 30,577 32,813 21,247 163,086

Table 2. Number of available observations.Number bonds per year and country (Panel I) and  number  of  observations  per  year  and  country  (Panel  II).  The  sample  includes sovereign issues  from  18 European countries  in the period June 2006 to October 2012, with  price  and  yield  information  available  in  Datastream,  and  securities  lending information available in DataExplorers.  

17,604 140,254 38,822 389,870 32,251 554,698 29,943 567,448 27,398 634,306 27,676 678,645 17,217 453,625 190,911 3,418,846

2006

2007

2008

2009

2010

2011

2012

Average

AUSTRIA

Lending Demand Lending Supply Active Utilization Fee Score (0‐5)

2.92% 8.54% 39.01% 0.1

3.42% 11.96% 44.73% 0.5

4.18% 14.08% 39.02% 0.6

2.13% 11.82% 25.54% 0.3

2.57% 14.58% 25.11% 0.2

3.23% 20.40% 26.90% 0.4

2.90% 18.08% 23.92% 0.2

3.06% 14.50% 30.71% 0.4

BELGIUM

Lending Demand Lending Supply Active Utilization Fee Score (0‐5)

5.04% 12.02% 41.35% 0.2

5.16% 10.41% 52.74% 0.4

4.57% 10.64% 46.33% 0.4

1.70% 9.55% 30.42% 0.2

1.95% 10.08% 35.11% 0.1

1.79% 8.79% 31.87% 0.3

1.14% 9.29% 25.16% 0.3

2.61% 9.80% 36.13% 0.3

CZECH REPUBLIC

Lending Demand Lending Supply Active Utilization Fee Score (0‐5)

0.42% 1.51% 28.77% 0.5

0.39% 4.46% 30.74% 0.2

0.35% 5.33% 15.09% 1.2

0.38% 4.93% 13.00% 0.5

0.47% 4.90% 20.36% 0.5

0.39% 7.58% 23.71% 0.6

0.52% 6.96% 20.34% 0.7

0.42% 5.68% 20.87% 0.6

DENMARK

Lending Demand Lending Supply Active Utilization Fee Score (0‐5)

1.64% 4.83% 27.46% 0.1

1.37% 7.97% 29.46% 0.5

0.77% 8.75% 23.05% 0.9

0.39% 7.22% 14.36% 0.5

0.84% 9.18% 23.92% 0.2

0.80% 10.58% 23.26% 0.2

0.95% 10.17% 24.21% 0.2

0.86% 8.79% 23.10% 0.3

FINLAND

Lending Demand Lending Supply Active Utilization Fee Score (0‐5)

0.90% 13.48% 33.23% 0.2

3.40% 19.16% 42.78% 0.3

6.08% 23.20% 46.87% 0.8

3.18% 15.43% 25.73% 0.3

2.06% 17.47% 29.22% 0.3

1.21% 18.23% 30.58% 0.5

1.20% 17.07% 29.01% 0.2

2.05% 17.97% 33.09% 0.4

FRANCE

Lending Demand Lending Supply Active Utilization Fee Score (0‐5)

2.68% 6.62% 46.39% 0.1

3.29% 10.68% 45.73% 0.5

3.48% 12.95% 42.11% 0.5

2.29% 11.91% 37.11% 0.3

2.15% 12.65% 35.11% 0.2

2.00% 14.07% 31.79% 0.3

1.72% 13.61% 30.59% 0.2

2.45% 12.58% 36.64% 0.3

GERMANY

Lending Demand Lending Supply Active Utilization Fee Score (0‐5)

6.33% 9.87% 55.75% 0.1

4.68% 14.48% 41.87% 0.4

3.56% 13.40% 37.31% 0.5

2.19% 10.45% 31.73% 0.4

2.18% 11.85% 33.03% 0.2

2.38% 17.33% 33.69% 0.4

2.41% 15.62% 33.20% 0.2

2.77% 13.89% 34.95% 0.3

GREECE

Lending Demand Lending Supply Active Utilization Fee Score (0‐5)

3.70% 13.52% 32.96% 0.1

4.04% 14.99% 37.50% 0.6

3.52% 14.62% 31.03% 0.9

1.62% 9.48% 22.39% 0.4

0.90% 4.54% 32.09% 0.8

0.50% 3.29% 42.77% 1.8

0.09% 1.01% 62.66% 3.3

2.23% 9.57% 33.39% 0.8

IRELAND

Lending Demand Lending Supply Active Utilization Fee Score (0‐5)

0.62% 5.70% 14.85% 0.1

2.14% 12.63% 32.32% 1.0

2.60% 10.13% 35.36% 1.0

1.22% 9.62% 20.16% 0.2

1.44% 8.34% 25.60% 0.5

1.43% 5.15% 42.20% 1.4

0.70% 2.17% 46.46% 1.5

1.56% 8.25% 31.74% 0.9

ITALY

Lending Demand Lending Supply Active Utilization Fee Score (0‐5)

2.10% 9.38% 32.50% 0.1

2.63% 9.55% 43.84% 0.2

2.35% 9.60% 39.60% 0.3

1.29% 6.62% 29.30% 0.2

0.92% 5.93% 23.16% 0.1

0.56% 6.97% 19.49% 0.2

0.37% 7.26% 21.00% 0.2

1.48% 7.85% 31.12% 0.2

LUXEMBOURG

Lending Demand Lending Supply Active Utilization Fee Score (0‐5)

0.02% 4.82% 7.45% 0.0

0.38% 3.99% 19.16% 0.0

0.94% 7.88% 18.27% 0.2

0.76% 7.24% 12.86% 0.6

0.40% 7.97% 7.17% 0.2

0.66% 6.99% 13.56% 0.4

NETHERLANDS

Lending Demand Lending Supply Active Utilization Fee Score (0‐5)

8.47% 14.38% 52.53% 0.2

9.90% 20.52% 56.98% 0.5

5.42% 18.71% 40.64% 0.6

2.49% 15.90% 26.36% 0.3

2.23% 16.46% 23.00% 0.2

2.51% 21.08% 26.36% 0.4

2.46% 16.61% 25.31% 0.3

3.32% 17.97% 29.81% 0.4

NORWAY

Lending Demand Lending Supply Active Utilization Fee Score (0‐5)

1.71% 9.07% 26.24% 0.3

1.95% 15.64% 22.74% 0.6

1.79% 15.43% 25.90% 0.8

0.68% 10.27% 17.00% 0.6

0.83% 8.87% 16.08% 0.3

0.81% 8.82% 21.78% 0.3

0.38% 9.09% 22.82% 0.2

1.06% 10.78% 21.30% 0.4

PORTUGAL

Lending Demand Lending Supply Active Utilization Fee Score (0‐5)

2.17% 8.82% 41.19% 0.3

3.38% 11.76% 48.60% 0.4

3.55% 14.23% 41.30% 0.7

0.86% 10.83% 18.51% 0.2

1.03% 9.14% 27.91% 0.4

0.54% 9.70% 30.46% 1.2

0.31% 6.33% 52.33% 2.2

1.49% 10.12% 34.81% 0.7

SPAIN

Lending Demand Lending Supply Active Utilization Fee Score (0‐5)

2.69% 9.75% 43.62% 0.2

3.18% 12.73% 47.37% 0.5

2.59% 11.62% 39.44% 0.6

1.09% 12.25% 27.45% 0.3

1.23% 13.03% 24.65% 0.3

0.99% 13.23% 20.38% 0.3

0.59% 10.58% 25.89% 0.7

1.54% 12.30% 29.70% 0.4

SWEDEN

Lending Demand Lending Supply Active Utilization Fee Score (0‐5)

2.42% 10.51% 31.92% 0.1

2.04% 10.87% 35.82% 0.4

2.41% 10.14% 40.86% 0.6

1.53% 9.81% 24.74% 0.4

1.34% 11.25% 24.46% 0.1

0.75% 14.60% 23.38% 0.1

1.09% 12.96% 24.17% 0.1

1.60% 11.62% 29.07% 0.3

SWITZERLAND

Lending Demand Lending Supply Active Utilization Fee Score (0‐5)

12.20% 19.83% 51.78% 0.0

11.51% 22.01% 52.59% 0.0

10.06% 21.20% 47.36% 0.3

3.58% 16.06% 23.51% 0.2

3.68% 22.75% 23.08% 0.2

4.66% 34.22% 24.89% 0.2

2.46% 29.53% 21.69% 0.1

6.22% 23.82% 33.67% 0.2

UK

Lending Demand Lending Supply Active Utilization Fee Score (0‐5)

11.55% 32.40% 41.90% 0.0

14.57% 38.38% 54.64% 0.0

11.43% 29.66% 59.20% 0.1

5.98% 19.77% 47.58% 0.1

5.75% 20.64% 43.36% 0.0

4.03% 19.17% 39.81% 0.1

3.66% 18.73% 40.66% 0.1

7.71% 24.59% 47.16% 0.1

Total

Lending Demand Lending Supply Active Utilization Fee Score (0‐5)

4.24% 11.85% 40.08% 0.1

4.87% 14.79% 44.38% 0.3

4.26% 14.44% 40.31% 0.5

2.23% 11.65% 29.92% 0.3

2.15% 12.59% 29.69% 0.2

2.09% 15.96% 29.61% 0.4

1.89% 14.49% 29.48% 0.3

2.85% 13.89% 33.53% 0.3

Table 3. Bond Lending Information by Country. We compute a relative lending demand for an issue as the ratio of the value of  current  inventory on  loan  from  beneficial  owners,  divided  by  the  total  market  value  of  the  issue  (both  in  dollars).  Lending  supply is the ratio of the active inventory value (current inventory available from beneficial owners less what is deduced to be  temporarily restricted from lending) to the total market value of the issue. Active Utilization is the ratio of lending demand to  lending supply is referred to as active utilization. The fee score is the value weighted average fee score, which is computed  from the average of all applicable loan fees weighted by loan value. Each average is then assigned to a fee bucket from 0 to 5,  with  0  representing  the  least  expensive  fee  bucket  (general  collateral)  and  5  the  most  expensive  (most expensive/special). Data on bonds’ market capitalization is from Datastream. The sample includes sovereign issues from 18 European countries in  the  period  June  2006  to  October  2012,  with  price  and  yield  information  available  in  Datastream,  and  securities  lending information available in DataExplorers.  

0 AUSTRIA BELGIUM CZECH REPUBLIC DENMARK FINLAND FRANCE GERMANY GREECE IRELAND ITALY LUXEMBOURG NETHERLANDS NORWAY PORTUGAL SPAIN SWEDEN SWITZERLAND UK Total Average Fee (bps)

Value Weighed Average Fee Score (0=Cheap, 5=Expensive) 1 2 3 4 5 Above 1

All

76.4% 79.5% 55.4% 79.6% 72.4% 77.3% 76.8% 61.3% 52.0% 87.0% 62.5% 73.4% 78.4% 62.8% 72.2% 84.5% 87.9% 95.3%

16.8% 16.1% 37.6% 13.4% 20.4% 16.7% 16.9% 19.4% 26.1% 8.7% 37.5% 19.8% 11.9% 18.2% 19.0% 9.3% 9.3% 4.0%

1.5% 1.6% 3.3% 3.0% 1.7% 1.1% 2.2% 4.6% 8.8% 2.4% 0.0% 2.7% 4.8% 7.5% 4.1% 3.3% 2.5% 0.6%

5.1% 2.7% 2.4% 2.8% 5.1% 4.7% 3.4% 11.9% 10.2% 1.2% 0.0% 3.9% 2.3% 8.3% 3.4% 1.9% 0.3% 0.0%

0.1% 0.1% 0.3% 1.2% 0.4% 0.2% 0.7% 1.6% 2.6% 0.7% 0.0% 0.3% 2.5% 2.2% 1.2% 1.0% 0.0% 0.1%

0.0% 0.0% 1.0% 0.0% 0.0% 0.0% 0.1% 1.3% 0.3% 0.1% 0.0% 0.0% 0.0% 1.1% 0.0% 0.0% 0.0% 0.0%

6.8% 4.4% 6.9% 7.1% 7.2% 6.0% 6.3% 19.4% 21.8% 4.4% 0.0% 6.9% 9.7% 19.0% 8.8% 6.1% 2.8% 0.7%

8,209 7,181 763 2,181 2,393 26,676 45,369 2,927 1,645 15,399 72 12,898 2,012 2,375 7,077 2,768 6,703 16,512

129,008 79.1% 54.70

24,197 14.8% 85.90

3,587 2.2% 166.10

5,343 3.3% 222.20

896 0.5% 492.20

129 0.1% 533.30

9,955 6.1%

163,160

Table  4.  Lending  Fees by  Country and  Year. The  table  reports  the  percent  of  observations  in  the sample falling  into each of the six fee buckets computed  by DataExplorers. The last column reports total observations by country. Third‐to‐last row reports total by fee bucket. The fee score is the value weighted average fee score, which is computed from the average of all applicable loan fees weighted by loan value. Each average is then assigned to a fee bucket from 0 to 5, with 0 representing the least expensive  fee  bucket  (general  collateral)  and  5  the  most  expensive  (most  expensive/specialThe sample  includes  sovereign  issues  from  18  European  countries  in  the  period  June  2006  to  October 2012,  with  price  and  yield  information  available  in  Datastream,  and  securities  lending  information available in DataExplorers.  

71.44

Standard&Poors Date

Rating

Outlook

Moody's Rating

Outlook

Fitch Rating

Outlook

Upgrade

Stable

AUSTRIA Dec/05/2011 Jan/13/2012 Feb/13/2012

Negative Downgrade Negative

BELGIUM Jan/13/2009 Oct/07/2011 Dec/05/2011 Dec/16/2011

Stable Negative Negative Downgrade

Negative

CZECH REPUBLIC Oct/02/2007 Mar/04/2008 Dec/08/2008 Jun/04/2010 Aug/24/2011

Upgrade Stable Positive Downgrade

FINLAND Dec/05/2011

Negative

Dec/05/2011 Jan/13/2012 Feb/13/2012

Negative

FRANCE Downgrade Negative

GERMANY Dec/05/2011

Negative

GREECE Jan/11/2007 Mar/05/2007 Oct/20/2008 Jan/14/2009 Feb/25/2009 May/12/2009 Oct/22/2009 Oct/30/2009 Dec/08/2009 Dec/10/2009 Dec/15/2009 Dec/16/2009 Dec/22/2009 Jan/11/2010 Feb/02/2010 Feb/18/2010 Mar/12/2010 Apr/09/2010 Apr/22/2010 Apr/27/2010 May/10/2010 Jun/14/2010 Dec/17/2010 Dec/21/2010 Jan/14/2011 Mar/07/2011 Mar/29/2011 May/09/2011 Jun/01/2011 Jun/13/2011 Jul/25/2011 Jul/27/2011

Positive Positive Stable Downgrade

Stable Stable Downgrade

Negative Negative

Downgrade

Negative

Downgrade

Negative

Junk

Negative Negative

Downgrade

Stable

Negative

Downgrade

Stable Stable Stable Downgrade

Negative

Stable Stable Stable Stable Downgrade Downgrade Junk Junk

Downgrade Downgrade Downgrade

Stable Negative

Downgrade

Negative

Downgrade

Negative Negative

Downgrade

Developing

Negative Negative Negative

Downgrade

LUXEMBOURG Dec/05/2011

Negative

Dec/05/2011

Negative

NETHERLANDS ITALY Oct/19/2006 May/21/2011 Jun/19/2011 Oct/04/2011 Dec/05/2011 Jan/13/2012 Feb/13/2012 Jul/13/2012

Downgrade Negative Downgrade Negative Downgrade Downgrade Downgrade

Negative Negative

Standard&Poors Date

Rating

Outlook

Moody's Rating

Outlook

Fitch Rating

Outlook

IRELAND Jan/30/2009 Mar/06/2009 Mar/30/2009 Apr/08/2009 Apr/17/2009 Jun/08/2009 Jul/05/2009 Nov/04/2009 Jul/19/2010 Aug/24/2010 Oct/05/2010 Oct/06/2010 Nov/23/2010 Dec/09/2010 Dec/17/2010 Feb/02/2011 Apr/01/2011 Apr/14/2011 Apr/17/2011 Jul/13/2011 Jan/14/2012 Aug/02/2012

Negative Negative Downgrade Downgrade

Negative

Downgrade

Stable

Downgrade

Negative

Downgrade

Stable

Negative Downgrade Downgrade

Negative

Downgrade

Stable

Downgrade Negative Downgrade

Negative Downgrade

Downgrade Downgrade

Negative

Negative Negative

Negative Negative Downgrade Junk

Negative Negative

Negative Negative

PORTUGAL May/01/2007 Jan/21/2009 Sep/03/2009 Oct/29/2009 Mar/24/2010 Apr/27/2010 May/05/2010 Jul/13/2010 Dec/21/2010 Dec/23/2010 Jan/21/2011 Mar/16/2011 Mar/24/2011 Mar/29/2011 Apr/01/2011 Apr/05/2011 Jul/05/2011 Jan/13/2012 Feb/13/2012

Stable Downgrade

Stable Negative Negative Downgrade

Negative

Downgrade

Negative

Downgrade

Negative

Downgrade

Negative

Downgrade Downgrade

Negative Stable Negative

Negative Downgrade Downgrade Downgrade

Negative

Negative Stable Downgrade Junk

Negative Negative

Junk

Junk Downgrade

SPAIN Jan/15/2009 Jan/19/2009 Apr/28/2010 May/28/2010 Jun/30/2010 Sep/30/2010 Dec/15/2010 Mar/04/2011 Mar/10/2011 Jul/29/2011 Oct/18/2011 Dec/05/2011 Feb/13/2012 Jun/07/2012 Jun/13/2012 Oct/10/2012

Downgrade Downgrade

Negative Stable Downgrade Downgrade

Negative Downgrade Downgrade

Negative Negative Negative

Negative Downgrade Downgrade Downgrade Downgrade

Negative

UK Feb/13/2012

Stable

Negative Stable Negative

Negative

Table 5. History of Rating Downgrades. For each country and rating agency, the table reports the dates of all rating events in the period June 2006 to October 2012.  For  S&P  and  Fitch,  a  downgrade  to  junk  status  means  that  the  issuer ratings changes from BBB‐ to BB+. For Moody's it means a change from Baa3 to Ba1. Any other downgrade is described as “Downgrade”. “Issuer Rating” refers to both  foreign  and local currency;  long‐term  and short‐term; as well  as issue ratings, which in our sample change all at the same time.  

Standard&Poors

Moody's

Fitch

All CRAs

Notch  Downgrades

Downgrades to  Junk

Upgrades

…of which no event in previous 30 days

27 25

2 2

1 1

…of which no outlook revision in prior 30 days

15

1

1

Events …of which no event in previous 30 days

21 20

3 3

1 1

…of which no outlook revision in prior 30 days

11

3

1

Events …of which no event in previous 30 days

14 13

2 2

0 0

…of which no outlook revision in prior 30 days

6

1

0

Events

62

7

2

…of which no event in previous 30 days

58

7

2

…of which no outlook revision in prior 30 days

32

5

2

Events

Table 6. History of Rating Downgrades by rating agency. For each country and rating agency, the table reports the type of downgrades for the period June 2006 to October 2012. For S&P and Fitch, a downgrade to junk status means that the issuer  ratings  changes  from  BBB‐  to  BB+.  For  Moody's  it  means  a  change  from  Baa3  to  Ba1.  Any  other  downgrade  is described as “Downgrade”. “Issuer Rating” refers to both foreign and local currency; long‐term and short‐term; as well as issue ratings, which in our sample change all at the same time. “No event in previous 30 days” means that no other rating agency releases a credit rating for the corresponding country in the prior 30 days. “No Outlook Revision in prior 30 days” means that no other rating agency releases an Outlook (even if it is a confirmation of a previous outlook) for the corresponding country in the prior 30 days.  

Panel A: Utilization Days to Downgrade Country GREECE GREECE GREECE IRELAND PORTUGAL PORTUGAL

Agency S&P  Moody's Fitch Moody's Moody's S&P ‐ Fitch

Average

 

Year 2010 2010 2011 2011 2011 2012

Month May June January July June January

‐5 19.51% 23.69% 21.90% 28.71% 17.27% 18.60%

‐4 20.12% 23.86% 21.84% 29.39% 18.70% 24.60%

‐3 20.43% 20.68% 21.82% 25.19% 21.16% 24.44%

‐2 21.18% 20.66% 22.30% 29.24% 16.34% 23.97%

‐1 22.06% 20.80% 21.61% 36.25% 18.03% 21.22%

0 21.52% 23.02% 21.62% 33.42% 19.25% 21.42%

+1 21.71% 22.40% 21.44% 41.36% 18.76% 19.58%

+2 20.18% 20.94% 23.06% 41.65% 18.95% 21.13%

+3 21.00% 19.50% 22.32% 45.41% 20.38% 21.42%

+4 22.90% 22.43% 24.32% 41.39% 17.04% 16.73%

+5 21.00% 22.68% 24.23% 46.48% 16.51% 17.17%

+6 21.71% 19.93% 28.94% 38.20% 15.02% 18.55%

+7 20.20% 19.40% 26.60% 29.50% 15.16% 19.39%

+8 20.14% 18.53% 27.53% 29.30% 14.53% 22.91%

+9 20.58% 19.34% 27.10% 27.72% 13.13% 23.03%

+10 20.49% 21.97% 24.60% 36.33% 12.94% 22.17%

 

21.61%

23.08%

22.29%

22.28%

23.33%

23.37%

24.21%

24.32%

25.01%

24.14%

24.68%

23.72%

21.71%

22.16%

21.82%

23.08%

Moody's S&P ‐ Fitch

From t=+4 to t=+10

IRELAND PORTUGAL ALL

S&P  Moody's FITCH Moody's

From t=0 to t=+3

Agency GREECE ALL

From t=‐4 to t=‐1

Average Utilization (Relative to  Time Series Average)

1.03 0.99 1.05 1.06 1.01 1.09 0.97 1.20

1.05 1.02 1.05 1.07 1.30 1.08 1.05 1.10

1.08 0.99 1.00 1.25 1.15 0.96 0.84 1.08

Table 7. Bond Lending Information by Country around Bond downgrades. We report variables related to sovereign securities lending for the three countries which are downgraded to junk during the  sample  period  (Greece,  Ireland,  and  Portugal).  We  report  these  data  by  rating  event.  On  January  13,  2012,  Portugal  was  downgraded  to  junk  status  both  by  Standard  &  Poors  and  FitchRatings. Panel A reports Active Utilization in a. Panel B reports a multiple of Average Utilization and Average Demand relative to the corresponding time‐series averages, computed with all  country observations between January 1, 2009 and October 31, 2012. To compute time‐series averages, we compute, for each country, the average utilization and demand per bond, excluding a window of [‐5,+10] days relative to the announcement any downgrade of the country’s debt, and then calculate the average across all bonds in the country.   Average Demand is the ratio of the  value of current inventory on loan from beneficial owners, divided by the total market value of the issue (both in dollars). Active Utilization is the ratio of lending demand to lending supply is referred to as active utilization. Lending supply is the ratio of the active  inventory value  (current inventory available from beneficial owners less what is deduced to be temporarily  restricted  from lending) to the total market value of the issue. Data on bonds’ market capitalization is from Datastream. The sample includes sovereign issues from 18 European countries in the period  June 2006 to October 2012, with price and yield information available in Datastream, and securities lending information available in DataExplorers.  

Notch Downgrades t=‐5 to t=‐1 Country

Number  of Bonds

CAR

t=0 p‐value

CAR

***

(0.0371) (0.0000) (0.0186) (0.0004) (0.7767) (0.0168) (0.0000) (0.0379) (0.0007)

0.0168% 0.0703% 0.2934% 0.1602% ‐2.0330% ‐0.6757% ‐0.4071% ‐0.7319% ‐0.1497%

**

(0.0129)

‐0.1945%

AUSTRIA BELGIUM CZECH REPUBLIC FRANCE GREECE IRELAND ITALY PORTUGAL SPAIN

84 77 36 422 72 22 222 54 320

0.1241% 0.4046% 0.0802% 0.0788% 0.0585% 0.9822% 0.7645% ‐0.2761% ‐0.3764%

**

Total

1309

0.1053%

*** ** ***

** *** **

t=+1 to t=+10 p‐value

CAR

***

(0.6265) (0.2600) (0.0060) (0.0000) (0.0000) (0.0144) (0.0006) (0.0023) (0.0010)

‐0.1254% 0.8141% 0.3445% ‐0.2476% ‐6.9389% 1.1441% 2.3449% ‐1.0443% ‐0.5903%

***

(0.0000)

‐0.2%

*** *** *** ** *** ***

p‐value (0.0157) (0.0000) (0.0025) (0.0000) (0.0000) (0.0339) (0.0000) (0.0065) (0.0000)

** *** *** *** *** ** *** *** ***

(0.1157)

Junk Downgrades t=‐5 to t=‐1 Country GREECE IRELAND PORTUGAL

Number  of Bonds 51 11 41

CAR ‐0.1882% ‐1.8635% 1.0055%

t=0 p‐value

** ***

Total 103 0.1081% *, **, *** Denotes significance at the 10%, 5%, and 1% levels

CAR

(0.4752) (0.0377) (0.0082)

0.1611% ‐0.8394% ‐0.3944%

(0.6340)

‐0.1669%

t=+1 to t=+10 p‐value

** ***

CAR

p‐value

(0.2729) (0.0496) (0.0029)

‐2.28% 2.27% ‐1.43%

***

***

(0.0006) (0.0551) (0.0030)

(0.1024)

‐1.4499%

***

(0.0004)

Table 8. Cumulative Abnormal Returns (CARs) around sovereign bond downgrades. We estimate market model regressions of daily sovereign bond return on a benchmark, for  all  bonds  in  the  sample,  and  in  the  period  June  2006  to  December  2007.  We  then  compute abnormal returns for a window of ‐5 to +10 relative to the downgrade event date  with  the  estimates  from  the  previous  regressions,  and  for  all  events  during  the  period  January  2008‐October  2012.  The  table  reports  average  CARs  across  bonds  in  a given country. As market benchmarks, we use the Markit IBoxx benchmark indices for Euro‐denominated  sovereign  bonds.  These  indices  include  investment  grade,  fixed  income, and sovereign bond issues in Euro. We obtain daily returns on the benchmars  from Datastream, for five different maturity groups: from 1 to 3, 3 to 5, 5 to 7, 7 to 10,  and more than 10. We match each issue in our sample with the corresponding maturity  benchmark.  Bond  return  data  are  from  Datastream.  The  sample  includes  sovereign  issues from 18 European countries in the period June 2006 to October 2012, with price and  yield  information  available  in  Datastream,  and  securities  lending  information available in DataExplorers.  

*

CDS spread in percent Interest Yield Bond Daily Return Eonia (Euro OverNight Index Average) Euro‐Dollar Exchange Rate MSCI Country Stock Market Index Ban on German Shorts (Y/N) Average Utilization All Bonds Germany Benchmark Bond 10Y Reference Bond (Y/N) category==Medium‐Term category==Short‐Term Intercept R‐squared N. of cases * p