Collapse of the Asset-Backed Commercial Paper Market

The Evolution of a Financial Crisis:

Collapse of the Asset-Backed Commercial Paper Market*

DANIEL COVITZ, NELLIE LIANG, and GUSTAVO A. SUAREZ

April 5, 2012

ABSTRACT

This paper documents “runs” on asset-backed commercial paper (ABCP) programs using a novel dataset of all transactions in the U.S. market during its severe contraction in 2007. We find that one-third of programs were run within weeks of the onset of the ABCP crisis and that runs, as well as yields and maturities for new issues, were related to program-level and macro-financial risks. The findings are consistent with the asymmetric information framework used to explain banking panics, have implications for the degree of risk-intolerance of commercial paper investors, and inform upon empirical predictions of recent papers on dynamic coordination failures.

Keywords: Asset-backed commercial paper, runs, financial crisis JEL Codes: G01, G10, G21

*

All authors are at the Federal Reserve Board. This paper represents the views of the authors and does not necessarily represent the views of the Board of Governors or other Federal Reserve staff. We are deeply indebted to two anonymous referees, Campbell Harvey (the editor), and an associate editor for detailed feedback that greatly improved the paper. We thank seminar participants at the Federal Reserve Bank of San Francisco, the Yale Conference on Financial Crisis Research, the IADB-LFN, the IMF, and the Bank of Canada, Viral Acharya, Franklin Allen, Adam Ashcraft, Markus Brunnermeier, William Dudley, Gary Gorton, Zhiguo He, Jeffrey Lacker, Eduardo Levi-Yeyati, Peter Lupoff, Gregory Nini, Philipp Schnabl, Jeremy Stein, and Wei Xiong for useful comments. We also thank Scott Aubuchon, Elisabeth Perlman, and Landon Stroebel for excellent research assistance.

Since the mid 1980s, banks have moved an increasing volume of assets off their balance sheets and funded them through asset-backed commercial paper (ABCP) programs, bankruptcy remote “paper companies” that issue short-term debt in the commercial paper market.1 Traditionally, ABCP programs financed receivables from nonfinancial companies, but over time they increasingly financed a wider range of assets, including highly rated mortgage- and other asset-backed securities. By the end of 2006, ABCP outstanding in the United States had grown to $1.1 trillion, larger than the amount of unsecured (non-asset-backed) commercial paper outstanding and a significant part of the U.S. shadow banking system.2

However, in the summer of 2007, ABCP outstandings began to plummet. The proximate cause of the contraction was mounting concerns about the default risk of subprime and other mortgages. Outstanding ABCP shrank by $190 billion, almost 20%, in August, and yields soared and maturities shortened for new issues. Outstanding ABCP fell by an additional $160 billion by the end of the year (see Figure 1). The deep contraction likely contributed to the broader financial crisis because banking institutions sponsored and provided liquidity and credit support to ABCP programs, and because securitization markets relied on ABCP for funding, and so were likely adversely affected by the contraction in ABCP.

[Figure 1 about here]

In this paper, we study the collapse of the ABCP market in 2007 in order to better understand the framework behind financial panics, as well as to improve our understanding of the risk intolerance of commercial paper investors, and to shed light on a distinguishing assumption in recent theories of coordination failures in short-term credit markets.

Our analysis exploits a rich data set based on all transactions and amounts of paper outstanding at ABCP programs in the U.S. market in 2007. The data are proprietary information from the Depository Trust and Clearing Corporation (DTCC) on the prices, quantities, and

maturities of almost 700,000 transactions by 339 ABCP programs, as well as weekly information on the maturity structure of program-level outstandings. These data are supplemented by hand-collected information from reports by major rating agencies on the type of program and the identities of the sponsors and liquidity providers, to create a dataset that is unparalleled in detail about ABCP programs.

The focus of our analysis is on the measurement and determinants of “runs” on ABCP

programs.3 A program is defined as entering a run during a week in which it does not issue despite having 10% or more of its outstandings scheduled to mature; the program continues in a run until it issues again. The empirical analysis of runs considers a rich set of potential determinants, including program risk characteristics, program type, sponsor type, and macro-financial variables. In addition, we conduct an empirical investigation of the yield spreads and maturities of new issues for programs not in a run.

The main empirical results are as follows. First, a substantial number of ABCP programs

experienced a run in the last five months of 2007. About 30% of programs were in a run within weeks of the onset of the ABCP crisis and nearly 40% of programs, more than 120 programs, were in a run at the end of 2007, and the odds of exiting a run were very low. Moreover, declines in outstandings at programs experiencing runs accounted for most of the drop in ABCP outstanding in 2007. Second, runs in the crisis were not random but instead were significantly more likely at riskier programs, based on observable program characteristics, program type, sponsor type, and macro-financial variables. Third, for the programs that could issue, yield spreads and maturities of new issues had explainable variation during the crisis, and the determinants were similar to those that help to explain runs.

These results are consistent with previous findings from studies of bank panics that runs

are caused by shocks with uncertain incidence in the cross section. This “asymmetric

information” framework was formalized first by Chari and Jagannathan (1988) to explain bank panics. Gorton (1988), Calomiris and Gorton (1991), and Calomiris and Mason (2003) provide evidence of this view by showing that bank panics were triggered by an observable macroeconomic shock, and were generally at weaker banks. Analogously, we find that runs appeared to be triggered by a macro-financial shock, and that they were more likely at “weaker” programs, such as those with weaker liquidity support, lower ratings, and weaker sponsors. Moreover, investors in ABCP likely knew little about the actual exposures of individual programs to subprime or other risky mortgages, in part because some sponsors viewed their portfolios as proprietary investment strategies. Further, our finding that the determinants of runs, yields, and maturities were similar provides additional evidence that runs were not random, which also supports the asymmetric information framework.

Our results also contribute to the literature that shows risk intolerance among commercial

paper investors. Calomiris, Himmelberg, and Wachtel (1995) find that the unsecured commercial paper market is restricted to very high quality firms. In addition, Calomiris (1995) finds that Penn Central’s failure in 1970 triggered declines in commercial paper issued by other institutions; Gatev and Strahan (2006) find similar adverse spillover effects on commercial paper issuers from Enron’s collapse in 2001.4 Further, Pennacchi (2006) describes money market mutual funds, the primary investors in commercial paper, as wanting to hold only very high quality assets to limit the risk of “breaking the buck.” A fund breaks the buck when its net asset value falls below $1 per share, in which case the fund’s commitment to redeem shares at $1 can trigger a run.5 Our finding of substantial numbers of runs on ABCP programs is also indicative of risk intolerance among commercial paper investors. However, our finding that spreads and maturities of new issues had explainable variation during the crisis suggests that commercial paper investors had a somewhat measured response to the risk of some programs.

Our results also have implications for theories of dynamic coordination failures involving short-term investors (see Acharya, Gale, and Yorulmazer (2011), He and Xiong (2011), and Brunnermeier and Oehmke (2012)). We are unable to test these theories directly because we cannot identify whether investors in one transaction are the same as investors in another.6 However, we do find evidence consistent with an important empirical prediction of Brunnermeier and Oehmke (2012), namely that debt maturities shorten when asset volatilities increase. In their model, greater volatility implies that more information is revealed about an issuer’s default probability at rollover dates, allowing short-term investors to extract rents from long-term investors. As a result, an issuer has an incentive to deviate from an equilibrium with long-term debt by issuing more short-term debt.7 Krishnamurthy (2010) documented that maturities in the commercial paper market shortened in the summer of 2007; the contribution of our analysis is to link this shortening to the weakness of programs, as well as to measures of spreads and volatilities in interbank funding markets.

The remainder of this paper proceeds as follows. In Section I we describe relevant

institutional details of the ABCP market, data, and summary statistics. Section II describes our definition and analysis of runs, and Section III presents our empirical analysis of maturities and yield spreads on newly issued ABCP. In Section IV, we discuss implications for bank balance sheets and securitization markets, followed by a conclusion in Section V.

I.

Institutional Details of the ABCP Market, Data, and Summary Statistics

A. Investor Information about Risks of ABCP Portfolios

Investors appeared to have little understanding of the credit quality of ABCP portfolios

leading up to the turmoil in August 2007. Indeed, Moody’s issued a report on July 20, just weeks before the crisis erupted, entitled “SIVs: An Oasis of Calm in the Subprime Maelstrom”

(Moody’s (2007)), suggesting little concern about the quality of assets. The possibility that investors had less than a complete understanding of the risks of ABCP is also suggested by a J.P. Morgan research note published on August 16 (JP Morgan (2007)), which noted that “ABCP is a complex investment that would take volumes to explain completely.”

Some information on ABCP holdings, aggregated across programs, was available in mid

to late 2007. In particular, Moody’s reported in July that aggregate holdings of highly rated private-label MBS for certain types of programs were about one-quarter of program assets (Moody’s (2007)). However, programs viewed their specific holdings as proprietary investment strategies, prompting trade organizations representing securities dealers and investors—the Securities Industry and Financial Markets Association (SIFMA), the American Securitization Forum (ASF), and the European Securitization Forum (ESF)—to recommend improvements in disclosures of assets held in ABCP programs in September, more than a month after the crisis had erupted.

In contrast to information about specific assets, information about each program’s characteristics, type, and sponsor are available in a program’s prospectus. In addition, rating agencies prepare periodic (typically annual) program-level reports to update information in the prospectus. Of the largest rating agencies, Moody’s Investors Service is the most comprehensive in covering the ABCP market.

B. Program Characteristics

Liquidity and credit support. Liquidity support insures against broad market disruptions that might otherwise force a program to sell assets. At the end of July 2007, about 87% of programs had explicit liquidity support from at least one financial institution in the form of a bank back-up line. As an alternative, or in some cases a complement to liquidity support from a

financial institution, 24% of programs at that time issued paper with options that allowed them to extend the maturity of the paper past its due date for a fixed period of time at a pre-set penalty rate (Table I). This feature, in effect, requires investors to internalize the program’s liquidity risk, making it a weak form of support from an investor’s perspective. From the point of view of ABCP investors, the extendibility option implies a risk of holding an asset that cannot be easily liquidated, should the issuer exercise the option of extending the maturity of the paper. Money market mutual funds, the typical investors in commercial paper, are sensitive not just about eventual repayment but also about the timing of repayment, because these funds are exposed to withdrawals from their own short-term investors. In addition, SEC rules impose an upper limit on the average maturity of the portfolios of registered money market mutual funds.

[Table I about here]

Some programs also have credit support, a contractual commitment to support the program if its assets became impaired. Only 16% of programs had credit support at the end of July 2007 (Table I). All programs with credit support in our data also had liquidity support.

Rating. Nearly all ABCP programs are rated by nationally recognized statistical rating

organizations. Ratings reflect the ability of the program to pay in full and on time. Short-term prime ratings assigned by Moody’s Investors Service are P-1 (the highest), P-2, and P-3 (the lowest). The vast majority of ABCP programs carry a P-1 rating by Moody’s, because they are either secured by receivables and over-collateralized, secured by highly rated and presumably diversified pools of securities, or because they have contractual support features (Moody’s (2003)). Ratings generally determine the eligibility of paper for purchase by money market mutual funds.

C. Program Types

Mutli-seller and single-seller programs are the traditional and most common program types. Such programs are “bankruptcy-remote” conduits that issue ABCP backed by receivables and loans purchased from multiple firms or a single firm, where bankruptcy-remoteness implies that the assets of the program will be shielded from the bankruptcies of the firms that sell the assets to the program. At the end of July 2007, the ABCP market contained 98 multi-seller programs, 40 non-mortgage single-seller programs, and 11 mortgage single-seller programs (i.e., programs that primarily warehoused mortgages prior to their securitization); combined, these programs accounted for 58% of the market (Table I). Notably, single-seller programs were the most frequent users of extension options at that time, with more than 60% issuing extendible paper.

Securities arbitrage programs accounted for 13% of the market’s outstanding paper at the

end of July 2007. These programs purchase long-term, highly rated securities and are often sponsored by banks to avoid the regulatory capital charge that would be incurred if the assets were held on the bank’s balance sheet; the sponsor banks typically provide liquidity support.

Structured investment vehicles (SIVs) also fund highly rated securities, and accounted for

7% of the market at the end of July 2007.8 However, such programs, in contrast to the other types of programs, tend not to have explicit agreements for committed back-stop liquidity lines to cover the full amount of their short-term liabilities. Instead SIVs rely on “dynamic liquidity management” strategies, which involve liquidating assets to pay investors if needed. Specifically, unlike other program types, SIVs use mark-to-market accounting with liquidation clauses (or wind-down triggers) that transfer the control of the program to a trustee that could liquidate the SIV’s assets if its junior liabilities or assets drop in value.

Collateralized debt obligations (CDOs) fund at least part of their senior tranche in the

commercial paper market. While similar to SIVs in terms of their assets, CDOs do not actively manage their liabilities; they tend to rely instead on full liquidity lines from financial institutions. CDOs accounted for about 4% of the market at the end of July 2007.

Hybrid programs combine features of securities arbitrage and multi-seller programs, and accounted for about 8% of the market at the end of July 2007. Other programs not classified elsewhere accounted for another 10%.

D. Sponsor Types

Sponsors of ABCP programs decide which assets to purchase and how to finance them.

Sponsors may directly provide liquidity or credit support to their programs, or contract separately for such support. In July 2007, Large U.S. banks (those with more than $500 billion in assets in mid-2007) sponsored mostly multi-seller programs (Table I). With the salient exception of Citigroup, no large U.S. banks were substantially involved in sponsoring the SIV segment of the market in July 2007. Small U.S. banks sponsored a very modest share of the market. Foreign banks sponsored a substantial share of ABCP, about 30% of programs and, relative to domestic banks, were more likely to sponsor securities arbitrage programs. Nonbank institutions, such as mortgage lenders, finance companies, and asset managers, sponsored roughly 55% of active programs in July 2007. Nonbank sponsors can contract with commercial banks for full liquidity support, and may also utilize extendibility features, dynamic liquidity management techniques, or simply offer less than full liquidity support (for example, in the case of SIVs).

E. Data and Summary Statistics

Our raw data include all transactions in the U.S. ABCP market in 2007: 693,762 primary market transactions (new issues) by 339 programs over 251 trading days. These data are from the Depository Trust and Clearing Corporation (DTCC), the agent that electronically clears and settles both directly and dealer-placed commercial paper. The issues in the sample are discount instruments paying face value at maturity. For each transaction, DTCC provides the identity of the issuer, the face and settlement values of the transaction, and the maturity of the security. Using these data, we calculate implicit yields on new overnight paper (maturity of 1-4 days) paid by issuers using standard money market conventions (annualized yields are calculated under the assumption of a 360-day year). We calculate overnight risk spreads as the ABCP rate less the target federal funds rate, an overnight lending rate for banks set by the Federal Open Market Committee. We also obtain from DTCC a separate weekly file that contains program-level information on the amount and maturity distribution of outstandings. Further, we supplement the DTCC data with hand-collected information on program type, credit ratings, liquidity features, and sponsor identity from various reports written by Moody’s Investors Service.

As total outstanding ABCP plunged by nearly 30% from August to December 2007 (Figure 1, panel A), different program types were not hit equally hard (Table II, panel A). Outstandings at multi-seller programs fell only about 10% from July to December, but outstandings at SIVs fell about 80%, and mortgage single-seller programs virtually disappeared. These dramatic declines in outstandings are consistent with the possibility that investors were intolerant to risk and that paper issued by certain program types may have had some risk. The risk of paper issued by certain program types may have reflected relatively weak program characteristics, a possibility that is explored below in Sections II.C and II.D.

[Table II about here]

Programs that were not run issued paper with shorter maturities and higher spreads than in the earlier part of 2007. For example, overnight ABCP yield spreads over the target federal funds rate across all program types soared to an average 47 basis points in August, and remained high and volatile through the end of the year (Table II, panel B, and Figure 1), up from monthly averages of between 2 and 6 basis points in the first seven months of 2007. While the jump in spreads was evident across all program types in August, spreads for single-seller and SIVs continued to escalate in subsequent months, while spreads on multi-seller programs narrowed relatively slightly until year-end pressures intensified.9

In addition, the average maturity of new-issue paper dropped to about 23 days on average

in the last five months of 2007, from 31 days on average in the first seven months of the year (Figure 1, panel C). Although all program types experienced notable declines in the average maturity of new issues, single-seller programs that specialized in mortgages and SIVs that continued issuing experienced more pronounced drops in their average maturity of new issues (Table II, panel C).10

II.

A. Defining Runs

Analysis of Runs

In traditional bank runs, depositors withdraw demand deposits from commercial banks.

We define a run on a commercial paper program analogously as occurring when short-term creditors refuse to roll their positions. To characterize such an action as a run requires that short-term creditors are moving ahead of other economic agents, as with depositors trying to withdraw their funds before other depositors (as in the numerous studies of depositor runs), or firms drawing on lines of credit before banks cut the lines (as in Ivashina and Scharfstein (2010), Campello, Graham, and Harvey (2010), and Campello, Giambona, Graham, and Harvey (2011)).

This notion of moving ahead of others is plausible in short-term credit markets, as the last investor to roll its position may end up with payment delays and credit losses.

To measure runs on ABCP programs, we define program i as being in a run in any

period t in which it has more than 10% of its outstanding paper scheduled to mature but does not issue.11 The program is also considered to be in a run if it was defined as being run in the prior period and does not issue in the current period. That is, programs remain in a run state until they issue. More formally:

Maturingit

0.1 and Issuanceit 01 if Outstandingit

Runit , (1)

1 if Run= 1 and Issuance0i(t 1)it

0 Otherwise

In our analysis, t is a particular week because our data on program outstandings, used to

measure the need to issue, are available only weekly. The condition that maturing paper is more than 10% of outstandings is intended to capture the need to issue. The condition that issuance is zero is intended to capture the inability to issue. The zero-issuance condition makes our definition of runs conservative in the sense that programs that issue even a small amount relative to the amount of maturing paper, perhaps at very high cost, will not be classified as being in a run. Another reason that our definition of runs may be conservative is that it classifies a program as not in a run, even if the paper was issued in a non-arms-length transaction to the program’s sponsor. Unfortunately, our data set does not contain the identity of investors, so we are unable to adjust for the possibility of non-arms-length transactions.

One additional limitation of our measure is that it may classify a program that is unable to issue in a given week as not being in a run, if the program has less than 10% of its outstanding paper scheduled to mature in that week and was not classified as being in a run in the prior week.

We address this issue by dropping observations for which the following three conditions hold: issuance during the week is zero, less than 10% of the program’s paper is scheduled to mature over the week, and the program was not in a run in the prior week. Importantly, all our results are both qualitatively and quantitatively similar if we include these observations.12

To our knowledge, no other empirical analyses of runs use transaction-level data on investor or depositor withdrawals. Recent studies define runs by the change in banks’ deposits or wholesale liabilities, which reflects the net effect of inflows and outflows by various depositors and investors. Shin (2009) discusses retail and wholesale runs at Northern Rock in 2007; Oliveira, Schiozer, and Barros (2011) investigate determinants of runs at Brazilian banks in 2008; and Iyer and Peydró (2011) examine the effects of fraud at a major Indian bank on runs at banks that were connected through inter-bank deposits. De Graeve and Karas (2010) use an ex ante definition of bank runs in Russia, as a supply shock in which deposit outflows are greater at uninsured banks than insured banks, but like the other papers, they only observe the net change in deposits. Gorton and Metrick (2012) define a run in the repo market when investors increase spreads and haircuts, but they do not have transaction-level data for repos.

B. ABCP Runs and Events in Money Markets in 2007

Runs on ABCP programs mounted quickly in August 2007. Runs, as defined in equation

(1), were quite low in each week from January to July of 2007, but then shot up in August as the financial market turmoil erupted (see Figure 2). Starting in August, the percent of ABCP programs experiencing a run each week climbed sharply through September to above 30% of all ABCP programs, and by the end of 2007, more than 40% of programs were in a run. As a result, after no ABCP program had defaulted for many years (from at least 2001 to July 2007), two programs defaulted in August, accounting for 2% of outstandings, and an additional three

programs defaulted by December. Similarly, although two programs had extended before August, extensions did not escalate sharply until August, and an additional 19 programs were extended by year-end. The total share of ABCP outstanding that defaulted reached about 3% by the end of 2007.

[Figure 2 about here]

The programs we identify as experiencing runs between July and December 2007 accounted for a substantial portion of the decline in ABCP outstanding depicted in Figure 1. ABCP outstanding at programs that experienced a run between July and December 2007 dropped 81% between July 25 and December 26, 2007. By contrast, ABCP outstanding at programs that did not experience a run decreased 2% over the same period. Taking into account the relative share of outstandings of programs that were run, roughly 95% of the decline in ABCP between July and December 2007 can be attributed to decreases at programs that experienced runs.

To assess our identification of runs, we evaluate the likelihood that a program exits a run. Quick exits from runs would seem inconsistent with the intuitive notion that a run is an absorbing state in which a program is essentially shut out of the market. The estimated unconditional hazard rate over time of the probability that a program in a run would leave the run state is represented by the dotted line in Figure 2. In the first seven months of the year, the estimated hazard rate is high on average, and generally ranges from around 20 to 50%, suggesting that the few identified runs during that period may not have been “true” runs in the sense of the programs being unable to subsequently issue new paper. In contrast, the estimated hazard rate fell notably in early August, and then declined to near zero by the end of the year, suggesting that the many identified runs from August to December were indeed runs.

The proximate cause of the runs was mounting concerns about exposures of ABCP programs to subprime mortgages. In early August, BNP Paribas halted redemptions from three

affiliated money market mutual funds, announcing that it could no longer value the holdings of U.S. subprime MBS held in the funds (Table III). The European Central Bank (ECB) immediately announced that it would supply reserves as needed to promote stability, which totaled $130 billion on August 9, and the Federal Reserve made a similar announcement on August 10. The spread of LIBOR over overnight index swap (OIS) rates, an indicator of banks’ willingness to lend to one another, shot up (Figure 3, Panel A).

[Table III about here] [Figure 3 about here]

It is worth noting, however, that while concerns about subprime mortgages appeared to precipitate turmoil in the ABCP market, the evolution of such turmoil through August and over the remainder of the year seemed not to coincide precisely with shifts in sentiment about subprime mortgages. Indeed, the return on the AAA-rated tranche of the ABX, which had been negative, turned positive in the second half of August (Figure 3, Panel B), even as the number of ABCP programs in runs continued to accumulate. Both the LIBOR-OIS spread and ABX return are highlighted and discussed in detail in Gorton and Metrick (2012) as measures of broad market stress in the fall of 2007.

C. Cross-Sectional Regressions of the Probability of Experiencing a Run

To analyze the determinants of runs, we first estimate a cross-sectional probit model for

the latent probability of program i experiencing a run in any week in the sample period on program characteristics, program type, and sponsor type.13 More formally:

Pr(Runi 1) F hProgram characteristicshi jProgram typeji kSponsor typeki ,

hjk

(2)

where F denotes the cumulative distribution function of a standard normal variable.

The model is estimated separately for two time periods: The first is from February to

July 2007, before spreads ballooned and outstandings plummeted (the pre-crisis period); the second is from August through December 2007, the period of market turmoil (the crisis period).14 The possibility that coefficients on the program variables might change with the crisis is suggested by Martinez-Peria and Schmukler (2001). We use standard errors that are robust to cross-sectional correlations.

The first group of explanatory variables in the specification control for program

characteristics. Specifically, Program characteristichi denotes characteristic h for program i. The first characteristic variable is Extendibility, which equals 1 for programs that have the option to extend the maturity of their paper at the issuer’s request. Extensions are a weak form of liquidity support, as investors are essentially absorbing the liquidity risk. A second program characteristic variable, Number of liquidity providers, proxies for the strength of support in the event of a rollover disruption. A third characteristic variable, Lower rating, equals 1 for programs rated below P-1 by Moody’s Investors Service in the month before the beginning of the sample period (January 2007 for the pre-crisis period and July 2007 for the crisis period). A fourth program characteristic, Credit support, equals 1 when programs have contractual commitments from financial institutions to support the program in the event of asset impairment. A final program characteristic variable is Initial average maturity of outstandings, defined as the average maturity of a program’s outstanding commercial paper in the month prior to the beginning of the respective sample, as programs with shorter-term liabilities might be more susceptible to runs.

The specification also includes controls for program type, as investors might have looked to these indicators as broad signals of potential exposures to subprime mortgages. Program typeji equals 1 if program i is type j and is 0 otherwise. The set of j program types includes

multi-sellers, non-mortgage single-seller conduits, mortgage single-seller conduits, securities arbitrage programs, SIVs, and CDOs; hybrids and other are the omitted group.

The third set of variables controls for the type of sponsor. Sponsor typeki, equals 1 if program i is sponsored by an institution of type k and is 0 otherwise. The set of k sponsors includes Small U.S. bank sponsor, Non-U.S. bank sponsor, and Nonbanking sponsor; the omitted category is large U.S. banks.

The results from the cross-sectional regressions are shown in Table IV.15 A first finding

from these regressions is that runs in the crisis period (column 2) were significantly more likely at programs with relatively weak characteristics. In particular, in the crisis period, the estimated marginal effect of Extendibility is positive and significant at the 1% level, and the marginal effect for Number of liquidity providers is negative and significant at the 1% level, indicating that runs were more likely at programs with weaker liquidity support.16 In addition, the effect of Lower rating was positive and significant in the crisis period, indicating that programs perceived to be weak were more likely to be run. In the pre-crisis period (column 1), the set of significant program characteristics is similar to that in the crisis period, though the estimated marginal effects are smaller in absolute magnitude and less significant.17 The marginal effect of Initial average maturity of outstandings is insignificant in both periods.

[Table IV about here]

Runs also seemed more likely at program types with likely exposures to subprime mortgages, and again the effects are generally stronger in the crisis period than in the pre-crisis period. In particular, the marginal effect of Multi-seller was negative and significant at the 1% level in the crisis period, suggesting fewer concerns about diversified conduits with little or no mortgage holdings; the corresponding effect in the pre-crisis period was significant at the 5% level and notably smaller in absolute magnitude. The marginal effect of Mortgage single-seller

is positive though insignificant in the crisis period, and near zero in the pre-crisis period; the insignificance of this effect in the crisis period is somewhat surprising and likely reflects the high correlation between Mortgage single-seller and Extendibility. In addition, the estimated effects for Structured investment vehicles and CDOs, categories of programs with exposure to subprime mortgages, are positive and significant in the crisis period; only the marginal effect for CDOs is significant in the pre-crisis period.18 The estimated effects for Securities arbitrage programs, which also tended to fund subprime mortgages, were insignificant in both periods.

Further, the cross-sectional regression results provide some evidence that runs during the crisis were more likely at programs with arguably weaker sponsors. In particular, the marginal effect for Small U.S. bank sponsor in the crisis period is positive and significant at the 10% level, while the estimated effects for other sponsor types are insignificant. Before the crisis, the only significant sponsor-type effect is that for Non-U.S. bank sponsor, which is significant at the 1% level and negative.

Overall, the results from the cross-sectional regressions indicate that runs were not

random, but instead were significantly more likely at riskier programs, with risk measured based on observable program characteristics, program type, and sponsor type. Moreover, the estimated effects of various determinants of runs are larger and more significant during the crisis, suggesting that investors make stronger distinctions across programs in periods of greater uncertainty.

D. Panel Regressions of the Probability of Experiencing a Run

We next examine the determinants of runs using a probit model of the latent probability of program i experiencing a run in week t. The panel specification allows the inclusion of one

additional program characteristic variable and four macro variables in the estimations. The resulting specification is as follows:

Pr(Runit 1)

F gProgram characteristicsh'it jProgram typeji kSponsor typeki lMacro variableslt .

gjkl

(3)

The set of program characteristics, g, differs from h in equation (2) in that Lower rating varies by week in equation (3), and equation (3) also includes the weekly CDS spread of the main liquidity provider, a measure of the perceived risk of the main liquidity provider. This variable should capture investors’ views about the ability of the main liquidity provider to meet its obligations to support the program. The main liquidity provider is defined as the one that contributes the highest percentage of committed liquidity lines and that also provides at least 20% of the lines.

As in the cross-sectional analysis, we estimate the model separately for the pre-crisis and crisis period. For each period, we estimate two specifications: in the first, the set of macro variables, l, in equation (3) includes the LIBOR-OIS spread and its volatility; in the second specification, l consists of ABX return and its volatility. A higher and more volatile LIBOR-OIS spread should reflect greater concerns about the ability of banking firms to access short-term funding in interbank markets. Declines in the ABX return could reflect investors’ views about the deterioration in the asset quality and solvency of ABCP programs since they were perceived to be exposed to subprime mortgage assets. We cluster standard errors at the program level to account for the likely correlation in errors within a particular program across time.19

The results from the panel regressions are shown in Table V. Looking first at the

coefficients on program characteristics suggests again that runs were associated with weaker contractual liquidity support and lower ratings. In the crisis period (columns 3 and 4), the

本文来源：https://www.bwwdw.com/article/t1pi.html