The Rise and Fall of Securitization

Sergey Chernenko sergey.chernenko@fisher.osu.edu

The Ohio State University

Sam Hanson shanson@hbs.edu

Harvard University

Adi Sunderam asunderam@hbs.edu Harvard University

December 2013

Abstract

The rise and fall of nontraditional securitizations—collateralized debt obligations and mortgage-backed securities backed by nonprime loans—played a central role in the financial crisis. Little is known, however, about the factors that drove the pre-crisis surge in investor demand for these products. Examining insurance companies’ and mutual funds’ holdings of fixed income securities, we find evidence suggesting that both agency problems and neglected risks played an important role in driving investor demand for nontraditional securitizations prior to crisis. We also use our holdings data to shed light on the factors that drove the dramatic collapse of securitization markets beginning in mid-2007. Contrary to conventional crisis narratives, we find little evidence of widespread fire sales. Instead, our evidence is more consistent with the idea that a self-amplifying buyers’ strike drove the dramatic collapse of securitization markets.

We are grateful to Josh Coval, Robin Greenwood, David Scharfstein, Erik Stafford, and seminar participants at Harvard for helpful comments. Hanson and Sunderam gratefully acknowledge funding from the Division of Research at the Harvard Business School.

1

The rise and fall of nontraditional securitizations—collateralized debt obligations and

mortgage-backed securities backed by nonprime loans—was at the heart of the recent financial crisis.

Although traditional securitizations of prime mortgages, commercial mortgages, and consumer debt

had existed for decades, nontraditional securitizations grew exponentially from 2003 to mid-2007.

The primary market for nontraditional securitizations then collapsed in dramatic fashion in mid-2007,

and secondary market prices for these products fell precipitously throughout 2007 and 2008. These

price declines generated enormous losses for large financial intermediaries, ultimately imperiling

their soundness and helping to trigger a full-blown systemic crisis in 2008.

Despite general agreement on this broad narrative, surprisingly little is known about the

underlying factors that drove the pre-crisis surge in investor demand for these nontraditional

securitizations and the subsequent collapse of demand. While competing theoretical explanations

abound, there has been little detailed empirical work assessing these narratives. In this paper, we use

new micro-data on mutual funds’ and insurance companies’ holdings of fixed income securities to

shed light on the factors that drove the rise and fall of securitization.

Explanations of the rapid rise of nontraditional securitizations from 2003 to mid-2007 point to

specific investor types that drove growing demand for these products due to either “bad beliefs” or

“bad incentives.” Explanations that emphasize “bad beliefs” argue that some agents misunderstood

the risks of investing in securitizations.1 For instance, demand may have been driven by rating-

sensitive investors who were misled by the AAA ratings granted to many securitizations.

Alternatively, demand may have been driven by investors who neglected the risk of a large

nationwide downturn in house prices, and thus believed that a diversified exposure to subprime

mortgages was virtually riskless.

By contrast, “bad incentives” explanations for the securitization boom emphasize the role of

principal-agent problems between professional investors and their principals.2 According to these

narratives, investors took on tail risk by buying securitizations. Because these investors received

equity-like compensation, buying securitizations rationally maximized investors’ own gains while

ultimately destroying value for their principals. Agency explanations for the boom come in many

flavors, which differ primarily in where they locate the critical agency conflict within the financial

1 See, for instance, Coval, Jurek, and Stafford (2009), Gennaioli, Shleifer, and Vishny (2011). 2 See, for instance, Rajan (2005).

2

system. Some explanations emphasize agency problems between external capital providers and

managers at financial institutions. Other explanations emphasize agency conflicts between firm

management and individual firm employees (e.g., traders) who wanted to take excessive risks. By

studying both mutual funds and insurance companies, we are able to shed light on the role of both

types of incentive problems.

A key question in assessing these competing narratives is: which investors drove the boom

and bust in securitizations? For instance, in the case of “bad beliefs” explanations, it is natural to ask

which types of investors were most prone to these mistaken beliefs. Similarly, in the case of “bad

incentives” explanations of the boom, we want to know which types of investors were most prone to

these incentive problems and what factors exacerbated the incentive conflicts.

Investor heterogeneity also plays a key role in explanations of the collapse of the market for

securitizations from mid-2007 to 2009. A broad theoretical consensus points to fire sales—the forced

sale of securities by investors with high valuations to others with low valuations—as the key

amplification mechanism at work during the bust.3 In these explanations, binding leverage constraints

forced financial intermediaries, who were natural holders of complex securitizations, to sell to other

investors with lower valuations. The resulting decline in prices further tightened intermediary

leverage constraints, leading to further forced sales. A central, yet largely untested, prediction of

these theories is that the decline in prices was associated with increased trade between levered

intermediaries and other investors. In other words, prices declined because large amounts of forced

trading significantly changed the identity of the marginal holder. An alternative to the conventional

fire-sales view is that investors staged a “buyers’ strike,” refusing to trade due to adverse selection or

some other friction.4

In this paper, we examine data on investors’ holdings of securitizations to shed light on the

types of investors who drove the boom and bust. Specifically, we study quarterly fixed income

holdings of mutual funds and insurance companies from 2003 to 2010. These “real money”

3 See, for instance, Shleifer and Vishny (1992, 2011), Kiyotaki and Moore (1997), Lorenzoni (2008), Brunnermeier and Pedersen (2009), Geanakopolos (2010), Stein (2012), Brunnermeier and Sanikov (2012), and Dávila (2011) for theoretical treatments of the fire-sale amplification mechanism. 4 See, for instance, Dang, Gorton, and Holmstrom (2010), Hanson and Sunderam (2013), Milbrandt (2013), and Morris and Shin (2012).

3

investors—intermediaries with access to stable capital from household savers—are two of the largest

investors in credit assets, owning close to 30% of all corporate bonds and private securitizations.5

Throughout the analysis we distinguish between traditional securitizations—which were

backed by prime residential mortgages, commercial mortgages, and consumer debt—and

nontraditional securitizations that consisted of securitizations backed by nonprime home mortgages

and collateralized debt obligations. While traditional securitizations had existed for decades,

nontraditional securitizations were a far more recent innovation. And the rapid growth in

securitization markets during the 2003-2007 boom was concentrated in nontraditional products.6

We begin by presenting two key aggregate stylized facts about the boom in nontraditional

securitizations from 2003 to 2007. First, there was significant variation in credit spreads at issuance

on highly-rated securitizations backed by different types of collateral. Spreads on AAA-rated

traditional securitizations were negligible. By contrast, spreads on AAA-rated nontraditional

securitizations were much wider, in line with spreads on BBB-rated corporate bonds. This suggests a

more nuanced view of the boom than simple tellings of either bad beliefs or bad incentives theories,

which tend to lump all AAA-rated securitizations together. Regardless of whether demand from some

investors was elevated due to bad incentives or bad beliefs, it appears that there were other investors

who clearly differentiated between traditional AAA-rated securitizations and the nontraditional

AAA-rated structure products that were at the heart of the financial crisis.

Our second aggregate fact is that mutual funds and insurers as a whole were “below market

weight” in securitizations, particularly in nontraditional securitizations. This suggests that the

proximate surge in demand for structured finance products did not stem primarily from real-money

investors. Nonetheless, to the extent that real-money investors are affected by the same incentive

problems and expectational errors as other investors, cross-sectional patterns in their holdings can

help shed light on the drivers of the boom in securitization. Given that nontraditional securitizations

grew most explosively during the boom, we focus our analysis on these securitizations. A simple

5 According to the Flow of Funds, as of 2007Q2, total outstanding corporate and foreign bonds, which include privately-issued securitizations, were $10.88 trillion. Of this, mutual funds (excluding money market mutual funds) owned $0.84 trillion (7%), and insurers (both life and property and casualty) owned $2.13 trillion (20%). 6 It should be noted that both traditional and nontraditional securitization markets collapsed in the bust, with traditional securitization recovering more quickly and strongly.

4

variance decomposition underscores the role of investor heterogeneity: variation across investors, as

opposed to common variation over time, is crucial for understanding our panel of holdings data.

To flesh out the drivers of this investor-level variation, we study the relationship between a

host of investor characteristics suggested by theory and holdings of nontraditional securitizations

during the boom. For mutual funds, we focus on a key prediction of the incentives narrative:

managers facing stronger performance-flow relationship should take more risk. We find little

evidence of this. However, we find that inexperienced mutual fund managers invest significantly

more in these products than more experienced managers—despite the fact that inexperienced and

experienced managers faced similar performance-flow incentives. This is consistent with the idea that

inexperienced managers are more prone to bad beliefs. Moreover, we find a discrete shift away from

nontraditional securitizations for managers who were active during the market dislocations of fall

1998, consistent with theories of reinforcement learning. Finally, we find that team-managed mutual

funds invest far more in these products than individual-managed funds. This is consistent with the

social psychology literature on group polarization in risk-taking, which argues that social dynamics

can lead groups to take on beliefs that are more extreme than the beliefs of any single group

member.7 Overall, the results are consistent with the idea that bad beliefs played an important role in

driving mutual fund demand for nontraditional securitizations.

Turning to insurance companies, we find that holdings of nontraditional securitizations are

concentrated among insurers that are poorly capitalized and have low financial strength ratings. In

addition, insurance companies that manage their portfolios internally hold more nontraditional

securitizations than those whose portfolios are managed externally, and insurance companies

organized as mutuals tend to hold fewer nontraditional securitizations than those organized as stock

companies. These findings are consistent with the idea that incentive conflicts play an important in

role in explaining insurer demand for nontraditional securitizations. In summary, our evidence

suggests that both bad beliefs and bad incentives are likely part of the explanation for the pre-crisis

surge in investor demand for these products.

Finally, we turn to investor behavior during the bust. We find little evidence of increased

transaction volume as prices of securitizations declined throughout late 2007 and 2008. Instead, 7 See, for example, Myers and Lamm (1976) and Isenberg (1986) in the social psychology literature and Bliss, Potter, and Schwarz (2008) and Bär , Ciccotello , and Ruenzi (2010) in the mutual funds literature.

5

trading in nontraditional securitizations declined through the boom from 2003 to 2007 and was

extremely low during the 2007 to 2009 bust. In contrast, trading in corporate bonds remained

relatively constant from 2003 to 2010. This suggests that the collapse of the secondary market for

these products may be better described as a buyers’ strike or market freeze than a fire sale.

A brief note on our approach is in order. First, the aim of this paper is not to advance one

specific narrative of the crisis. Indeed, it is unlikely that one theory alone can completely explain a

complex phenomenon like the rise and fall of nontraditional securitization. Second, our analysis is

largely descriptive. Lacking clean natural experiments, we are not able to definitively identify causal

effects. Instead, we pursue a series of descriptive tests that can help us begin to discriminate between

alternative explanations. We carefully consider factors that may confound our interpretations of these

tests, but ultimately our tests remain descriptive. Nonetheless, we believe they are a useful first step.

In summary, our goal in this paper is to document a series of robust stylized facts that can inform

future theoretical and empirical work on the crisis. In this way, our approach is similar to that of

Griffin, Harris, Shu, and Topalugua (2011), who study investor behavior in the dot-com boom and

bust.

The plan for the paper is as follows. Section I provides some background on traditional and

nontraditional securitizations. Section II outlines different theoretical narratives of the crisis and

develops a number of testable hypotheses. Section III explains the numerous data sources we use in

the paper. Section IV analyzes our holdings data to explore drivers of the boom, while Section V uses

our data to shed light on the mechanics of the bust. Section VI offers some concluding observations.

I. Background on Securitization

This section provides a brief background on securitization. Securitizations are created through

the process of “pooling”—assembling a pool of cash-flow-generating financial assets such as loans or

debt securities—and “tranching”—issuing claims of various priorities backed by that pool.

In the United States, the securitization of home mortgages dates back to the late 1960s and

early 1970s when various Government Sponsored Enterprises (GSEs), corporations that were

implicitly or explicitly backed by the US government, began issuing securities backed by residential

6

mortgage pools.8 Only “conforming” mortgages—loans to borrowers with high credit (FICO) scores,

with low loan-to-value ratios, below a Congressionally-approved loan size limit (non-“Jumbo”

loans), and with sufficient documentation—are eligible to be securitized in GSE mortgage-backed

securities (GSE MBS). The market for GSE MBS grew exponentially during the 1980s, and by the

early 2000s it had overtaken the market for US Treasuries to become the single largest debt market in

the US (Fabozzi 2005).

The late 1980s saw the advent of several types of private securitizations that were not

implicitly backed by the government. We broadly refer to these securitizations as “traditional

securitizations”.9 They include:

x Commercial Backed Mortgage Securities (CMBS): The securitization of commercial

mortgages into CMBS was pioneered by investment banks in the early 1980s. However,

this market only grew to prominence in the early 1990s when the US Government’s

Resolution Trust Corporation began relying heavily on CMBS securitizations to sell off

the commercial real estate loan portfolios of failed depository institutions. The CMBS

market grew rapidly throughout the 1990s and 2000s.

x Consumer Asset-Backed Securities (ABS): The securitization of credit card debt, auto

loans, and student loans began in the late 1980s. These markets developed rapidly

throughout the 1990s, and by 2000 over 30% of all non-mortgage consumer debt had been

securitized. However, the market for consumer ABS grew more slowly during the 2003-

2007 boom, and the fraction of consumer debt that had been securitized actually fell to

25% by 2007Q2.

x Prime non-GSE MBS: Beginning in 1977, investment banks began securitizing prime

“jumbo” home mortgages that conformed to all GSE MBS criteria other than size

(Fabozzi 2005; Gorton 2008).

8 Ginnie Mae issued its first MBS in 1968, and Freddie Mac followed in 1971. Fannie Mae did not begin issued MBS until 1981. Ginnie Mae is a government owned corporation and is explicitly guaranteed by the US government. Fannie Mae and Freddie Mac receive a variety of unique government benefits and have historically been perceived as being implicitly backed by the government. 9 The exact definitions of these different types of securitizations are not universally agreed upon. We typically hew closely to conventional investor categorizations. However, we sometimes make simplifications for the sake of clarity.

7

The boom in securitization from 2003 to 2007 prominently featured new types of

securitizations that developed after these traditional securitizations. We label the second set of

securitizations “nontraditional”. They include

x Nonprime RMBS: The securitization of subprime and Alt-A home mortgages began in

the mid 1990s (Lehman Brothers 2004). Alt-A mortgages are non-conforming due to high

loan-to-value ratios or a lack of sufficient documentation. Subprime mortgages are non-

conforming because the borrower has a low FICO score. The origination and

securitization of nonprime mortgages exploded during the boom from 2003-2007.10

x Collateralized Debt Obligations (CDOs): CDOs are securitizations backed by a

portfolio of fixed income assets. Depending on the underlying assets, CDOs are classified

as collateralized bond obligations (CBOs), collateralized loan obligations (CLOs), or ABS

CDOs. CBOs are collateralized largely by corporate bonds and were the most common

type of CDO in the late 1990s and early 2000s. However, this market grew slowly after

the prominent downgrades of several CBOs backed by high yield bonds in 2002. CLOs

primarily use unsecured loans to highly leveraged corporations (“levered loans”) as

collateral. CLO issuance boomed from 2003 to 2007. ABS CDOs use bonds from other

securitizations as the underlying collateral. ABS CDOs, particularly those using nonprime

RMBS as collateral, grew explosively over the course of the boom. Many of the most

dramatic losses by large financial intermediaries announced during 2007 and 2008 were

linked to ABS CDOs.11

Although our classification of securitizations into traditional and nontraditional is not universally

used in the academic literature, the classification is meaningful and also serves as a useful short-hand.

Furthermore, it is motivated by the historical evolution of these markets described above,

classifications widely employed by practitioners, as well as by the differences in new-issue pricing

that we discuss below. 10 See Fabozzi (2005), Gorton (2008), and Ashcraft and Schuermann (2008) for further background on nonprime MBS. Oddly, market participants sometimes refer to subprime mortgages as home equity loans (HEL). This is because the first subprime MBS securitizations in the mid-1990s actually contained second lien mortgages or home equity loans (Lehman Brothers 2004). Over time, these were replaced by first lien mortgages, but the name stuck. 11 See Shivdasani and Wang (2013) for further background on CLOs. See A. K. Barnett-Hart (2009) and Cordell, Huang, and Williams (2012) for more detail on ABS CDOs.

8

II. Narratives of the Rise and Fall

In this section, we discuss competing narratives of the boom and bust in nontraditional

securitization with an eye towards motivating our empirical tests. Our description of these narratives

and their predictions is deliberately stark. Our goal in this paper is not to provide definitive tests of

the different narratives. Instead, our goal is to document a series of robust stylized facts that can

inform future research about the mechanisms that drove the rise and fall of securitization.12 In

pursuing this goal, our approach is to document correlations, even if they reflect endogenous

relationships between investor characteristics and holdings of structured products.

A. Narratives of the Rise

Explanations of the boom in securitization can broadly be grouped into two categories: bad

beliefs and bad incentives.13 We describe each of these narratives in turn, outlining their empirical

predictions for prices and patterns in investor holdings.

A.1 Bad beliefs

In the context of the 2003-2007 securitization boom, bad beliefs explanations argue that

investors misunderstood the risks of investing in nontraditional securitizations. Essentially, investors

treated all AAA-rated securitizations as though they were virtually risk-free. For instance, Gerardi,

Lehnert, Sherlund, and Willen (2008) and Gennaioli, Shleifer, and Vishny (2012) argue that investors

neglected the risk of a substantial nationwide downturn in house prices and therefore believed that

diversified exposure to residential mortgages was almost riskless. Coval, Jurek, and Stafford (2010)

argue that investors focused on credit ratings which reflected expected losses, but neglected the

systematic risk embedded in securitizations. Ashcraft, Goldsmith-Pinkham, and Vickery (2010) and

Rajan, Seru, and Vig (2012) argue that investors and rating agencies failed to update their models of

12 Erel, Nadauld, and Stulz (2013) carry out a similar exercise for bank holding companies. However, they are not able to examine variation across the type of collateral backing the securitization. 13 We do not explicitly focus on explanations of the boom that emphasize the demand for safe, “money-like” assets (e.g., Caballero, 2009; Caballero and Farhi, 2013; Gorton and Metrick 2009, 2010). While these explanations are surely important, they can fall into either the bad beliefs or bad incentives categories depending on the reasons that investors perceived nontraditional securitizations to be money-like. Similarly, we largely bypass explanations focusing on the role of international demand for safe assets (e.g., Bernanke 2005, Bernanke, et. al. 2011, Caballero, 2009; Shin, 2012). Since our data only included US-based mutual funds and insurers, it is not particularly useful for addressing these questions.

9

loan default in the face of deteriorating loan underwriting standards, thus neglecting the risks

associated with lending to riskier households.

We begin with predictions about prices. Simple versions of the bad beliefs narrative for the

boom have strong predictions for security prices. Specifically, if the demand from investors who

neglected these risks was large enough, then the yield of all AAA-rated securitizations should have

been similar and close to those on default-free government bonds.

Bad Beliefs 1: Prices of AAA-rated nontraditional securitizations should have been highly homogeneous and close to risk-free rates.

We next turn to predictions about holdings. In bad beliefs narratives, a key question is which

types of investors were the optimists who bought large quantities of nontraditional securitizations

during the boom. In the simplest version of the bad beliefs narrative, all investors incorrectly assessed

the risks of securitizations. This suggests that most of the variation in our holdings data should be

driven by common variation over time.

Bad Beliefs 2: Under a simple version of the narrative, most variation in holdings of nontraditional securitizations should be common variation over time.

In more nuanced versions of this narrative, particular subsets of investors may have been

more susceptible to bad beliefs. For instance, Gennaioli, Shleifer, and Vishny (2012) argue that

unsophisticated investors were shocked by the swift decline of nationwide home price. Consistent

with this, anecdotal accounts suggest that demand from smaller, less sophisticated investors like

small insurance companies and pension funds played an important role in fueling the boom.14 Coval,

Jurek, and Stafford (2010) argue that rating-sensitive investors, and the rating agencies themselves,

were crucial because ratings measure expected loss rather than systematic risk. From a security

design perspective, the AAA-rated tranches of securitization are “economic catastrophe bonds” that

load heavily on tail risk and therefore maximize yield for a given expected loss.

In the context of mutual funds, the literature has argued that inexperienced managers may be

more prone to bad beliefs (e.g., Greenwood and Nagel, 2009; Malmandier and Nagel, 2011;

Campbell, Ramadorai, and Ranish, 2013). Similarly, the literature on group polarization in social

14 For instance, the town of Narvik, Norway suffered significant losses on CDO investments and appeared in the infamous stick figure “Subprime Primer” (http://www.nytimes.com/2007/12/02/world/europe/02norway.html). Springfield, MA also suffered similar losses (http://dealbook.nytimes.com/2008/01/28/in-this-city-cdo-spells-trouble/).

10

psychology argues that groups tend to make decisions that are more extreme than the initial

inclinations of its members. This suggests that team-managed mutual funds may take more extreme

positions in securitizations than individually-managed funds. Finally, the level of expertise of a

mutual fund family in evaluating fixed income securities may proxy for the accuracy of its beliefs.

Specifically, under the bad beliefs view, it seems natural to think that unsophisticated mutual funds

seeking “safe” assets would be large holders of securitizations, while more sophisticated funds would

correctly avoid these products.

Bad Beliefs 3: Mutual funds managed by less experienced managers and mutual funds managed by teams should have purchased more nontraditional securitizations.

A.2 Bad incentives

In contrast to bad beliefs narratives, bad incentives narratives of the boom emphasize agency

problems within the financial sector.15 In these explanations of the boom, sophisticated actors within

financial institutions correctly understood the risks of securitizations all along. However, due to

incentive frictions between these agents and their ultimate principals, these agents purchased more

securitizations than their ultimate principals would have if acting on their own behalf.

For prices, the bad incentives narrative has similar implications to the bad beliefs narratives.

In very simple versions of the narrative, given enough demand from financial institutions fraught

with agency problems, the prices of AAA-rated securitizations should have been highly homogenous.

Essentially, if principals treated all AAA-rated securities as though they were risk free, agents would

demand as much of these securities as possible, regardless of their true riskiness.

Bad Incentives 1: Under the bad incentives narrative, prices of AAA-rated nontraditional securitizations should have been highly homogeneous and close to risk-free rates.

Different versions of the bad incentives narrative locate the critical agency friction at different

points within the financial system. Some emphasize agency problems between external capital

providers and managers at financial institutions. For instance, the mutual fund literature typically

argues that agency problems between fund managers and investors stem from the relationship

between fund performance and future flows (Chevalier and Ellison, 1997, 1998). Since expected fund

inflows are a convex function of past fund outperformance, managers face incentives to take on 15 There is a large literature studying the incentives faced by loan officers in the origination process (e.g., Keys, Mukherjee, Seru, and Vig 2009). We focus on incentive problems with investors who ultimately purchased those loans.

11

significant risk to thus maximize their expected future compensation even if this does not maximize

expected fund returns. Thus, according to the bad incentives view, the stronger the performance-flow

relationship facing a fund manager, the larger her holdings of securitizations should be.

Bad Incentives 2: Mutual funds’ holdings of nontraditional securitizations should be correlated with the strength of the performance-flow relationship faced by funds.

Because insurance companies sometimes outsource the management of their portfolios to

external asset managers, they also offer a venue for studying this type of incentive problem. In this

kind of delegated asset management, the central conflict is between traders, who are compensated

based on risk-adjusted performance, and external capital providers. If capital providers cannot

perfectly observe risk-taking and therefore perfectly risk-adjust performance, traders may have

incentives to take excessive risks. This benchmarking problem may be particularly acute for highly-

rated securitizations since, almost by construction, they perform well outside of adverse, low-

probability economic states (e.g., Acharya, Pagano, and Volpin, 2013; Adrian and Shin, 2013).16

Bad Incentives 3: Under the bad incentives narrative, insurers that have externally managed portfolios should buy more nontraditional securitizations.

Agency problems between external capital providers and managers at financial institutions

can also take the form of risk-shifting in the tradition of Jensen and Meckling (1976). In these

narratives (e.g., Landier, Sraer, and Thesmar, 2011), managers acting on behalf of existing

shareholders bought securitizations to take on tail risk, increasing the expected payoff to equity,

while reducing total firm value at creditors’ expense.17,18 For instance, the capitalization levels and

ratings of insurance companies may reflect the strength of the agency conflict between their equity

holders and their creditors, implicit (i.e., the government) and explicit (i.e., policyholders).19

Specifically, less well-capitalized insurers may have stronger risk-shifting incentives, and may

therefore be more likely to hold nontraditional securitizations. Finally, insurance companies are 16 Ellul and Yerramilli (2012) provide evidence that banks with stronger risk management took less tail risk in the boom. 17 Beltratti and Stulz (2011), Fahlenbrach and Stulz (2011), and Erel, Nadauld, and Stulz (2013) argue that incentive conflicts between shareholders and managers have little power to explain the performance of banks during the crisis. 18 The “too big to fail” version of this story focuses on the government’s role as an implicit creditor of large financial institutions. Under this explanation, these institutions bought securitizations to rationally maximize the value of their implicit government guarantees. See e.g., Acharya, Cooley, Richardson, and Walter, 2009; Acharya and Richardson, 2009; Acharya, Schanbl, and Suarez, 2011; and Iannotta and Pennacchi, 2012. 19 The literature on risk-taking incentives in insurance companies is less well-developed than the comparable literature on mutual funds. Important exceptions include Becker and Ivashina (2013), who show that within rating insurers tend to hold riskier corporate bonds, and Koijen and Yogo (2013), who study regulatory capital arbitrage by insurers.

12

typically organized either as public companies with external shareholders or as mutuals, which are

owned by policyholders. These differences in organizational form also create differences in the

strength of the agency conflict between equity holders and creditors. In the case of mutual companies,

policy holders are effectively providing both equity and debt to mutual insurance companies,

potentially reducing the scope for risk-shifting.

Bad Incentives 4: Insurers with less capital and low ratings should buy more nontraditional securitizations. In addition, insurers organized as mutuals should buy less.

Note that both the bad beliefs and bad incentives narratives provide reasons why some

investors may have had elevated demand for nontraditional securitizations. The large literature on

limits to arbitrage initiated by DeLong et al (1990a) and Shleifer and Vishny (1997) suggests that

other investors may have been reluctant to aggressively lean against these demand shocks. This

combination of demand shocks and limited arbitrageur risk-bearing capacity would then explain why

these demand shocks raised equilibrium prices and led to a surge in issuance. Going further, rational

investors may even have had incentives to “ride” a bubble rather than leaning against it as in DeLong

et al. (1990b) and Abreu and Brunnermeier (2002).

B. Narratives of the Bust

In contrast to the strong disagreement about the drivers of the boom in securitizations, there is

wider consensus about the ensuing market collapse from mid-2007 to 2009. The frictionless null

hypothesis is that the decline in the prices of securitizations during the financial crisis was driven by

bad news about their fundamental values (i.e., news about discounted expected cash flows). If all

investors simultaneously revised their assessment of fundamental value, we would not expect the

collapse in prices to be associated with a surge in trading volume. By contrast, narratives of the bust

have typically emphasized amplification mechanisms that led prices to over-react to this bad news.

B.1 Fire sales

The most common narrative of the bust suggests that fire sales significantly amplified the

impact of this initial bad news on prices. In these explanations, binding leverage constraints forced

financial intermediaries—who were the natural holders of securitizations—to sell to other investors

with lower valuations. The resulting decline in prices further tightened leverage constraints, leading

to further forced sales. In other words, fire-sales theories articulate a natural amplification mechanism

13

whereby some initial moderately bad news could have ultimately had a severe impact on prices.20 A

critical feature of these explanations is that the decline in the prices of securitizations was associated

with increased trade between levered intermediaries and other investors. Put differently, these

explanations for the bust emphasize fire sales not fire holds. Prices declined because the identity of

the marginal investor changed as natural holders were forced to sell their holding to others.21

In this way, different explanations for the bust have very different implications for aggregate

trading patterns. The fundamentals-driven narrative suggests no reason for trading behavior to change

with the onset of the financial crisis, while the fire sales narrative suggests increased trade at the

beginning of the crisis.

Fire Sales 1: Under the fire sales narrative, transaction volume in nontraditional securitizations should have risen during the crisis.

In addition, we can examine the cross-section of our transactions data to further distinguish

between different narratives of the bust. For instance, fire sales narratives predict that mutual funds

suffering large outflows are forced to liquidate their structured product holdings. A related prediction,

from the large literature on limited arbitrage, suggests that externally managed funds may have been

forced to sell more of their risky holdings than internally managed funds due to frictions between

fund managers and external capital providers.

Fire Sales 2: Under the fire sales narrative, mutual funds facing outflows and insurance companies with externally managed portfolios should sell more nontraditional securitizations.

B.2 Buyers’ strikes

A second explanation that has received less attention is that investors’ uncertainty about the

valuations of securitizations changed dramatically. In these narratives, investors staged a “buyers’

strike”, and prices entered a downward spiral characterized by a near absence of trade. Precipitous

price declines and a collapse of trading volume may be explained by a variety of frictional

amplification mechanisms. In Diamond and Rajan (2012), for instance, institutions that would 20 See Kiyotaki and Moore, 1997; Shleifer and Vishny, 1992, 2011; Geanakoplos, 2009; Brunnermeier and Pedersen, 2009; Stein, 2012 for fire-sale explanations. 21 Merrill, Nadauld, Stulz, and Sherland (2012) provide evidence that some insurers sold some nonprime residential mortgage backed securities at fire sale discounts. However, the aggregate volume of transactions they study is quite small. Ellul, Jotikasthira, Lundblad, and Wang (2012) exploit a difference in regulation between property and casualty insurers and life insurers to argue that mark-to-market accounting can exacerbate fire sales. Ellul, Jotikasthira, and Lundblad (2011) argue that regulation forced insurers into fire sales of corporate bonds before the crisis. Coval and Stafford (2007) study fire sales by mutual funds in equity markets.

14

otherwise sell assets at fire-sale prices instead refuse to do so due to a risk-shifting problem. By

contrast, Dang, Gorton, and Holmstrom (2010) and Hanson and Sunderam (2013) rely on adverse

selection mechanisms. In these models, bad news lowers prices and raises uncertainty about asset

valuations; as a result, the scope for adverse selection between informed and uninformed investors

becomes too large and trade collapses. Milbrandt (2012) presents a model where trade freezes

because financial intermediaries wish to avoid recognizing losses that tighten their capital constraints.

In the model, prices fall until the potential profits from transacting exceed the shadow cost of the

tightening capital constraints.

Buyers’ Strikes 1: Under the buyers’ strikes narrative, transaction volume in nontraditional securitizations should have collapsed during the crisis.

III. Data

The data we use in our analysis comes from a number of different sources which are

described below.

A. Mutual Fund and Insurance Holdings

Our primary data set is the Thomson Reuters eMAXX database of quarterly security-level

holdings of asset-backed securities by US-domiciled mutual funds and insurance companies.

Thomson Reuters obtains par value holdings data from regulatory filings—Schedule D for insurance

companies and Forms N-CSR(S) and N-Q for mutual funds—as well as directly from these firms.

Our sample period runs from 2003Q1 to 2010Q4, which is long enough to allow us to study changes

in the demand for securitizations during the boom as well as trading during the crisis. The sample of

firm-quarter observations conditions on having at least one asset- or mortgage-backed security in the

investment portfolio, including GSE MBS securities. Our cross-sectional results are therefore

conditional on having some exposure to securitizations. The firms that are missing from the data are

largely those that by regulation or choice never invest in securitizations.22

22 There are two exceptions to this general statement. The first is that some money market mutual funds (MMMFs) are included in the eMAXX data during the early part of our sample period, but disappear over time. Because of these coverage issues and because risk taking by MMMFs has been studied before (Kacperczyk and Schnabl, 2013; Chernenko and Sunderam, 2013), we exclude MMMFs from our analysis. The second exception is that separate accounts of insurance companies are included in the eMAXX data during 2003-2004 and 2009Q4-2010Q4, but not during 2005Q1-2009Q3. We therefore exclude separate accounts from the analysis.

15

B. Securities

We supplement our holdings data by collecting detailed security-level data from a number of

sources. To obtain comprehensive data on the collateral backing each security, we collect security

data from the three major credit rating agencies—Fitch, Moody’s, and S&P—and from Bloomberg.23

Some examples of collateral descriptions in these data are “MBS - Prime Jumbo”, “CDO - Cash ABS

- Mezzanine”, “ABS - Floorplans - Auto Dealer”, or “Subprime/Home Equity RMBS”. These

descriptions allow us to differentiate between traditional and nontraditional securitizations. To

simplify the analysis, we classify securitizations into six broad collateral types: (1) GSE MBS, (2)

CMBS, (3) consumer ABS, (4) prime private-label RMBS, (5) nonprime private-label RMBS, and (6)

CDOs. Our focus is on explaining the demand for the nontraditional securitizations that were at the

heart of the credit boom and crisis. We therefore calculate the share of nonprime RMBS and CDOs in

a given investor’s overall bond portfolio and label it nontraditional share.

In addition to information on collateral, we collect data on the coupon or spread at issuance

from Moody’s and Bloomberg, the initial credit rating from Fitch, Moody’s and S&P, and the time

series of outstanding amounts from Bloomberg.

C. Mutual Funds

We also collect data on the characteristics of the investors in our eMAXX data. We obtain

mutual fund characteristics from the CRSP Mutual Fund Database. We construct a fund-level link

between eMAXX and CRSP. We first match funds based on the spelling distance between their

names and then manually check both matched and unmatched observations. We are able to link 84%

(2,355 out of 2,813) mutual funds in eMAXX to CRSP.

Our cross-sectional analysis excludes index funds since they follow a passive investment

strategy. We also exclude funds that invest solely in government securities since the mandates of

these funds routinely restrict them from investing in private securitizations. Finally, we exclude the

handful of equity and municipal bond funds that show up in our data because they account for a very

23 eMAXX provides some information on the collateral backing each security, but the information is not granular enough for our purposes. The most common value of the collateral variable, accounting for 68.5% of all security-level observations, is Single Family Mortgage Loans. This is a very broad category that mixes agency securities together with private-label subprime RMBS.

16

small fraction of aggregate securitization holdings and invest a very small share of their portfolio in

securitizations.

From Morningstar we obtain data on the portfolio managers of all US mutual funds, including

their age, educational background, and their start and end dates managing different funds. We use

birth year when available and educational background information to infer each manager’s age. We

measure manager’s experience using the first time we observe them managing a mutual fund.

D. Insurance Companies

We obtain insurance company data from AM Best. The AM Best data includes information on

insurers’ assets, capitalization, income, realized and unrealized capital gains, insurance premia, and

AM Best’s financial strength rating for individual insurers. We are able to match 84% (3,630 out of

4,302) of insurance companies in eMAXX to AM Best. Overall, our data on insurance companies is

similar to our data on mutual funds, except that we do not have biographic data on insurer portfolio

managers.

IV. The Rise

A. Aggregate Facts

In this section, we use our data to shed light on the drivers of the boom in securitization between

2003 and 2007. As discussed in Section III, the two main explanations we explore are those based on

bad incentives and those based on bad beliefs. We start by examining aggregate time-series facts

about the boom, before turning to more detailed cross-sectional analyses.

A.1 Aggregate issuance during the boom

We begin by providing some background statistics on the explosive growth of the market for

securitizations between 2003 and 2007. Figure 1 shows the dramatic rise and fall of issuance of

securitizations. Issuance of nontraditional securitizations almost quadrupled from $98 billion in

2002Q4 to $420 billion at the peak in 2006Q4. By comparison, issuance of traditional securitizations

roughly doubled from $103 billion in 2002Q4 to $200 billion at its peak in 2007Q2. Panel B shows

that out of traditional securitizations, consumer ABS was roughly stable, and most of the growth took

place in CMBS. Panel C shows that the growth of nontraditional securitizations was dominated by

nonprime RMBS, while CDO issuance was concentrated in 2005-2006.

17

A.2 Primary market pricing during the boom

As this rapid growth in issuance and outstanding amounts was taking place, there was

surprisingly little adjustment in the prices of securitizations. In Panel A of Figure 2, we plot the

quarterly time series of average credit spreads for AAA-rated securitizations backed by different

types of collateral from 2003Q1 to 2007Q4. We compute value-weighted averages of new-issue

spreads, restricting attention to the spreads on floating rate notes indexed to 1-, 3-, and 6-month

LIBOR to avoid benchmarking issues.

Panel A shows that during the boom, credit spreads on AAA-rated nontraditional

securitizations were significantly wider than those on other comparably-rated credit assets, including

AAA-rated traditional securitizations and corporate bonds. The data reject the conventional view that

investors were willing to buy any AAA-rated asset at nearly risk-free rates. The difference in spreads

between nontraditional AAA-rated products and traditional AAA-rated bonds, while modest in

absolute terms, is economically meaningful relative to typical spreads on AAA-rated assets. Indeed,

spreads on AAA-rated nonprime RMBS, CMBS, and CDOs were closer to those on BBB-rated

corporate bonds than to those on AAA-rated corporate bonds. During the height of the boom from

2004Q1 to 2007Q2, AAA-rated consumer ABS spreads averaged 9 bps, AAA-rated nonprime RMBS

spreads averaged 22 bps, and AAA-rated CDOs spreads averaged 34 bps. This can be compared with

corporate bond spreads that averaged 4 bps (over LIBOR) for AAA-rated bonds and 37 bps for BBB-

rated bonds.

Panel B of Figure 2 shows similar variation in the prices of BBB-rated securitizations. BBB-

rated tranches of traditional consumer ABS securitizations largely traded in line with BBB-rated

corporate bonds. By contrast, BBB-rated nonprime RMBS, CMBS, and CDO tranches traded in line

with BB-rated and B-rated corporate junk bonds. Specifically, from 2004Q1 to 2007Q2, spreads on

BBB-rated consumer ABS averaged 81 bps, spreads on BBB-rated RMBS averaged 179 bps, and

those on BBB-rated CDOs averaged 231 bps. This compares with corporate bond spreads that

averaged 37 bps (over LIBOR) for BBB-rated bonds, 131 bps for BB-rated bonds, and 231 for B-

rated bonds.

At the same time, we find little difference in the cross-sectional dispersion of spreads within a

given rating and collateral class. For instance, from 2004Q1 to 2007Q2 the cross-sectional standard

deviation of spreads among AAA-rated consumer ABS averaged 10 bps, whereas the cross-sectional

18

dispersion for AAA-rated CDOs averaged 12 bps. Thus, the bulk of variation in credit spreads within

a given rating was across collateral classes as opposed to within collateral class.24

The very wide spreads on nontraditional securitizations are an underappreciated fact about the

boom, and one that creates problems for simple versions of both bad beliefs and bad incentives

explanations. As discussed above, if bad beliefs or bad incentives affect the demand for all

securitizations similarly, then we would expect the prices of securitizations to be quite homogenous.

For instance, taken literally, in Gennaioli, Shleifer, and Vishny’s (2012) neglected risks model of the

boom, all investors should have treated all securitizations as a perfect substitute for riskless

Treasuries. Instead, the price variation we observe suggests that the marginal buyer of nontraditional

securitizations believed these assets were significantly more risky than traditional securitizations.25

Moreover, the observed variation in spreads lines up with ex post loss rates, at least directionally.

Figure 3 plots average spreads on different asset classes during the boom, against their cumulative

loss rates in the bust as reported by Moody’s (2012). The figure shows a positive correlation,

indicating that the securities investors regarded as more risky during the boom did in fact suffer more

losses during the bust. This does not mean that investors got prices “right” during the boom. Given

the ex post losses, investors should have been charging spreads an order of magnitude larger.

However, it does show that investors correctly understood the ordering of risk across products.

Of course, this does not rule out bad beliefs or bad incentives as key drivers of the boom. It

merely suggests a more complex story in which agents disagreed about the risks of nontraditional

ABS, but investors with more conservative valuations had limited risk-bearing capacity.

A.3 Who bought securitizations during the boom?

The variation in prices documented above suggests that data on investor holdings are critical

for understanding drivers of the boom. They suggest important heterogeneity in investor attitudes

towards different types of securitizations, making it important to know which intermediaries were

driving demand. Relatively little is known about which intermediaries bought subprime RMBS and

24 Thus, our results do not contradict those in Adelino (2010), who finds that, in the cross-section of AAA-rated subprime RMBS, new issue credit spreads have very little predictive power for future downgrades or default. His findings within an asset class and rating are consistent with our between asset class findings. 25 It seems plausible that the pricing differences between traditional and nontraditional securitizations were, to a significant degree, about expected losses as opposed to differences in risk premia. Since all highly-rated securitization tranches are like “economic catastrophe bonds” which only incur losses in systematically bad times, it is difficult to appeal solely to risk premia to explain the differences between traditional and nontraditional ABS pricing.

19

CDOs during the boom. To some extent, this gap in our knowledge reflects shortcomings in

reporting: prior to the crisis, regulatory filings for banks and insurers did not readily distinguish

between securitizations and traditional corporate bonds. Fortunately, we can use our eMAXX data to

begin to fill in these holes in our understanding.26

Tables 1 and 2 report estimates of the outstanding quantity of various private credit securities

and the estimated holdings of insurers and mutual funds from 2003 to 2010. In Table 1, we report

holdings of all private credit securities, which include corporate bonds, foreign bonds, and privately

issued ABS. In 2003Q2, insurers owned 26% of private credit securities which trended down to 20%

by 2010Q4. Mutual funds owned 7% of private credit securities in 2003Q2 which trended up to 11%

by 2010Q4. Thus, our data capture two of the largest investors in credit assets. We use these numbers

as a benchmark for assessing the overall weight of insurers and mutual funds within private credit

securities markets.

We next report holdings of traditional consumer ABS and CMBS. Both insurers and mutual

funds were underweight traditional consumer ABS. However, insurers were consistently overweight

CMBS, holding approximately one-third of outstanding CMBS from 2003 to 2010. This is consistent

with insurers’ long-standing expertise in commercial real estate. According to the Flow of Funds,

insurers’ market share of all commercial mortgages, not including CMBS holdings, averaged 28%

from 1952 to 2000.27

Table 2 shows the evolution of outstanding amounts and holdings for nontraditional

securitizations—private RMBS and CDOs/CLOs. As shown in Table 2, outstanding RMBS and

CDOs/CLOs surged during the boom. However, both insurers and mutual funds were lightening up

on subprime RMBS, alt-A RMBS, and CDOs/CLOs exposure from 2003 to 2007. Specifically,

comparing insurers’ market shares in Panel B with their share of all private credit securities in Panel

26 Previous estimates, put together in the midst of the crisis, include Lehman Brothers (2007, 2008). 27 According to the National Association of Insurance Commissioners (2012), insurers invest in commercial real estate “for three primary reasons: (1) to increase the level of asset type diversification in their investment portfolio; (2) to match long-term assets with long-term liabilities, because commercial mortgage loans are generally long-term with fixed interest rates and almost always include prepayment (call) protections; and (3) to minimize credit losses, because, based on historical experience, commercial mortgage loans have had modest realized credit losses.” See http://www.naic.org/capital_markets_archive/121026.htm for further background.

20

A, we see that insurers became more underweight RMBS and CDOs as issuance boomed from 2003

to 2007Q2.28 We see similar patterns for mutual funds.29

This suggests that demand for these assets was not primarily coming from the regulated,

unlevered sector. If regulatory arbitrage was driving the surge in investor demand, it was not the

desire of insurance companies to arbitrage their capital regulations in order to hold high-yielding

securities. Indeed, looking across collateral types, the share of real money investors appears to be

negatively correlated with the new issue credit spreads shown in Figure 2. Moreover, the distribution

of ratings for insurance company and mutual fund holdings, shown in Figures 4 and 5, shows that

they were not tilting their portfolios towards tranches with a particular rating. Approximately 80% of

their holdings are rated AAA, which is roughly the share of issuance volume that was AAA-rated.

The one exception is holdings of CDOs, where only 40% of holdings are AAA-rated. This suggests

that regulatory arbitrage was not creating demand for AAA-rated CDOs among insurers and mutual

funds.

Though the evidence suggests that mutual funds and insurers did not drive the growth of

securitizations, they were still important holders of these products, with about $90 billion in total

holdings as of 2007. In the next sections, we turn to more detailed cross-sectional determinants of

insurance company and mutual fund holdings. To the extent that these investors were affected by the

same incentives and biased beliefs as other investors, cross-sectional patterns in their holdings should

shed light on drivers of the boom in securitization more generally.

B. Mutual Funds

In this section we examine portfolio weights of nontraditional securitizations in the

investment portfolios of mutual funds, before turning to insurance companies in the next section.

B.1 Variance Decomposition

We start by reporting the results of a simple variance decomposition of the share of

nontraditional securitizations in a fund’s bond portfolio. The results in Table 3 suggest that variation

28 The National Association of Insurance Commissioners also finds that insurer holdings of CDOs are quite low. See http://www.naic.org/capital_markets_archive/110218.htm. 29 The finding that mutual funds and insurers held very few CDOs is consistent with data on publically announced structured finance write-downs as of January 2009. In short, with the exception of AIG, US-based life and P&C insurers announced few CDO write-downs. Furthermore, few stand-alone commercial banks had large CDO write-downs: almost all bank CDO write-downs were in banks with a broker-dealer. See http://www.docstoc.com/docs/45701126/Creditflux-writedowns-Microsoft-Excel-Spreadsheet_-78-kb-xls.

21

across funds, as opposed to common variation over time, is crucial for understanding mutual fund

holdings of securitizations. Specifically, common variation over time can explain less than 2% of

total variation in the portfolio share of nontraditional securitizations. Fund fixed effects, on the other

hand, explain more than 60% of total variation.

We might expect funds within the same fund family to face similar incentives and to have

access to similar information. As a result, fund investment decisions are likely to be correlated within

a family. Columns 5-7 confirm this intuition. Fund family fixed effects can explain about 30% of

total variation in nontraditional share. Nevertheless, in large families, those with at least 10 funds,

there is significant variation across different funds that belong to the same family, suggesting that

heterogeneity across individual portfolio mandates and managers plays an important role. The R2 of

family fixed effects for these funds is only 11%.

Overall, the results of this variance decomposition underscore the role of investor

heterogeneity. To assess the bad beliefs and bad incentives narratives of the boom, we therefore

examine the cross-sectional patterns in mutual fund holdings of nontraditional securitizations.

B.2 Mutual Funds versus Variable Annuities

To begin our analysis, we focus on a key prediction of the incentives narrative: managers

facing strong performance-flow relationships should take more risk. Our ideal experiment would hold

fixed fund managers’ information and vary their incentives. This suggests comparing the investment

decisions one manager makes for two different funds with different performance-flow sensitivities.

Of course, different funds managed by the same manager are likely to have different investment

objectives and restrictions, which would confound the comparison. We would like to isolate cases

where the same manager is in charge of multiple funds with the same investment objective and

benchmark but different performance-flow sensitivities.

Variable annuities provide an excellent laboratory to approximate this ideal experiment.

Variable annuities are a type of insurance contract whose value depends on the performance of the

investment options selected by the purchaser of the insurance policy. The underlying investment

options are typically mutual funds. These variable annuity funds are regulated and structured just like

regular mutual funds under the Investment Company Act of 1940. The only difference is that variable

annuity funds are available only to purchasers of variable annuities. Because they are not widely

available and sold as part of an insurance product, variable annuity funds are likely to face different

22

performance-flow relationships, and therefore risk-taking incentives, than regular mutual funds.

Table 4 shows that variable annuity funds face lower performance-flow sensitivities than

regular mutual funds. CRSP’s coverage of variable annuities does not start until 2008Q3.30 As a

result, we can only look at the performance-flow relationship after the crisis. Specifically, we use

monthly data for the January 20009 to June 2013 period. We include all bond and hybrid funds and

do not limit the sample to funds in our eMAXX data. Comparing the coefficients on past returns

interacted with regular and variable annuity fund dummies, performance-flow relationship is 2-3

times stronger in regular mutual funds than in variable annuity funds. Given such large difference in

the performance-flow relationship, the incentives story would predict that regular funds would

engage in more risk taking than comparable variable annuity funds.

We therefore study a sample of fund pairs where one fund is a variable annuity fund, and its

paired fund is a regular mutual fund managed by the same portfolio manager. To construct this paired

sample, we start with a sample of 328 variable annuity funds in our eMAXX data as of 2003Q4. We

then manually search for a regular mutual fund that is offered by the same fund adviser.

For example, Hartford High Yield HLS Fund is a variable annuity fund in our data. It is

offered by Hartford Life Insurance Company and its affiliates and is managed by Hartford Investment

Management Company (HIMCO). We search the list of regular mutual funds managed by HIMCO

for a regular fund with the same investment objective and find one—the Hartford High Yield Fund.

According to their prospectuses, the two funds have the same investment goal—high current income

with growth of capital a secondary objective—and investment strategy. The text describing the

investment goals and investment strategies of the two funds are virtually identical. The two funds are

also managed by the same team of three portfolio managers. The only difference between the two

funds is that Hartford High Yield HLS Fund serves as an “underlying investment option[s] for certain

variable annuity and variable life insurance separate accounts of Hartford Life Insurance Company

and its affiliates.”

We are able to find matching funds for 158 variable annuities (about one half of the original

sample). A handful of regular funds serve as matches for multiple variable annuities. Figure 6 shows

the time series of nontraditional share for matched pairs. The number of fund pairs peaks in 2003Q4,

30 CRSP does have some variable annuities before 2008Q3. But their number jumps by about an order of magnitude in 2008Q3. Furthermore, before 2008Q3, just four insurance companies account for 90% of all variable annuities.

23

the point in time as of which we actually form matched pairs, and declines over time. Nontraditional

shares track each other quite closely. Figure 6 does not control for differences in size which could in

principle affect nontraditional share. However, controlling for size does not affect our results.

Despite significant differences in the performance-flow relationship, Figure 6 shows that

managers make similar investment decisions in their regular and variable annuity funds. The fact that

we see such similar behavior despite meaningful differences in incentives points to the importance of

information and beliefs. One caveat is that in addition to facing weaker performance-flow

relationship, variable annuity funds also experience somewhat less volatile fund flows.31 Having

more stable funding could have made variable annuity funds more willing to invest in securitizations,

offsetting the effects of facing weaker incentives. The difference in the volatility of fund flows is

smaller, however, than the difference in the performance-flow relationship. Controlling for objective

and size, the volatility of fund flows into variable annuities is lower by around 7%. This is about 13%

of the average annualized volatility of 51%.

B.3 Manager Experience

Manager experience is another characteristic that may affect the strength of the performance-

flow relationship facing a mutual fund. Career concern stories emphasize the importance of strong

performance for young inexperienced managers. For instance, Chevalier and Ellison (1999) also find

that the relationship between fund flows and past performance is stronger in younger funds, and that

the termination of younger managers is more sensitive to performance than the termination of older

managers. If younger fund managers face stronger performance-flow incentives we might expect

them to tilt their portfolios towards higher-yielding, riskier securitizations. In this section, we

examine this prediction in more detail.

In our analysis, we use two alternative proxies for manager experience, both from

Morningstar. Our first proxy is age. Morningstar provides birth year information for about 20% of

managers in our data. Otherwise, we impute the age of managers based on their graduation

information (Chevalier and Ellison, 1999). Specifically, we have information on the year fund

managers received their undergraduate, masters, MBA, and PhD degrees. We assume that managers

receive these degrees when they are 22, 27, 27, and 29 years of age. These numbers correspond to the

31 Based on data for the January 2009 – June 2013 period.

24

median age of mutual fund managers receiving these degrees in our data. Our imputed age measure is

available for about 60% of fund-quarter observations in our sample.

Our second proxy for experience is the number of years each manager has been managing

mutual funds. Using the start and end dates for fund-manager pairs in our Morningstar data, we

calculate the first time each manager appears in our data. We then compute the manager’s experience

as the difference between December 31, 2004 and this date.

Of course, the experience of the fund manager is potentially endogenous to the fund’s

investment strategy. For example, a fund that intends to invest more heavily in complex

securitizations might want to hire younger portfolio managers, or more managers, to analyze these

securities. Such endogeneity concerns are one reason we measure manager experience and the

number of portfolio managers as of a single point in time.32 To the extent that the boom in

securitization was largely unanticipated in 2004, then if we measure experience as of 2004 it is less

likely that funds chose the experience of their managers in response to the boom.

Table 5 reports summary statistics for mutual fund managers in our data. The average

manager is 45 years old and has 8.5 years of experience managing mutual funds. In addition to

experience, we also contrast the behavior of individually-managed and team-managed funds. About

64% of funds in our data are team-managed. The median fund has 2 managers.

In Figure 7 we plot the average nontraditional share of funds managed by experienced versus

inexperienced managers. The plot on the left splits managers based on their experience managing

mutual funds. Both experienced and inexperienced funds start with about a 3% portfolio weight in

nontraditional securitizations. Starting in 2004, all managers increase their nontraditional share, but

inexperienced managers increase their nontraditional share by significantly more than experienced

managers. The nontraditional share of inexperienced managers peaks at around 8% while that of

experienced managers peaks at around 5%.

The plot on the right splits managers based on their age instead of experience. Here the

picture is less clear. Although younger managers tend to have higher nontraditional shares, especially

around the peak of the credit boom, the difference is smaller at around 1%. Overall, the difference in

results highlights the importance of experience managing mutual funds as opposed to age per se.

32 Greenwood and Nagel (2009) use the same approach.

25

We examine the association between experience and nontraditional share more rigorously in

Table 6. Except for 2009, where it is close to zero, the coefficient on experience is negative and

statistically significant. Consistent with Figure 7, the coefficient is largest in magnitude during 2007.

At that point, the nontraditional share of experienced managers was 3.4% less than the nontraditional

share of inexperienced managers. This effect is economically large compared to the average

nontraditional share of 6.1%.

These splits are consistent with the idea that incentives motivated young managers to take on

more risks. However, the incentives narrative rests on the premise that young managers face stronger

performance-flow relationships. The existing literature shows that this is the case for equity mutual

funds. However, Table 7 shows it is not the case for the bond mutual funds in our data. Consistent

with prior literature, we find that fund flows respond strongly to past performance. However, when

we interact past performance with experience, we do not find any differences in the performance-

flow relationship for young and inexperienced managers on one hand and old and experienced

managers on the other hand.

If not differences in incentives, what explains the tendency of young managers to invest in

nontraditional securitizations? It could be that experience directly affects manager’s beliefs. This idea

is explored by Greenwood and Nagel (2009), Malmandier and Nagel (2011a, 2011b), and Campbell,

Ramodarai, and Ranish (2013). For example, Greenwood and Nagel (2009) find that young managers

were more heavily invested in technology stocks at the peak of the technology bubble, and that they

were more likely to exhibit return-chasing behavior. In the context of bond markets, managers who

have not managed through a credit cycle, experienced sharp movements in interest rates, or sudden

unwinding of carry trades, might be less likely to fully appreciate the risks involved in investing in

securitizations.

We find evidence consistent with this in Table 8. The effect of experience seems to be highly

nonlinear—in our data it appears to matter most when a manager has at least 7-8 years of experience.

These are the managers who were active during the market dislocations of fall 1998—which saw

turbulent fluctuations in credit spreads following Russian default in August which endangered a

highly leveraged hedge fund called LTCM—and perhaps learned about risk in that episode.33 One of

33 At the same time, it is possible that the crisis was a period during which investors learned a lot about the ability of fund

26

the key results of career concerns models is that learning about skill takes place early on. Here we

find the opposite result—little learning appears to take place for the first seven years. Overall, the

nonlinearity in the effect of experience on nontraditional share is more consistent with managers

learning about risk than with investors learning about manager ability.

B.4 Team-managed Funds

The size of the management team is another dimension along which funds may differ in their

propensity to act on biased beliefs. The social psychology literature on group polarization suggests

that social dynamics can lead groups to take on beliefs that are more extreme than the beliefs of any

single group member. For example, if the initial tendency of individual group members is to take on a

risky gamble or to engage in an altruistic act, then following group discussion, members will take on

even more extreme beliefs and attitudes.

The social psychology literature has explored two main explanations for the group

polarization phenomenon. The informational influence or persuasive argument theory suggests that a

kind of non-Bayesian updating tends to occur in groups. Essentially, group discussion tends to elicit

arguments in favor of the initially preferred alternative, and group members do not adjust for the fact

that they are exposed to a subset of information and possible arguments.34 According to the

interpersonal comparison explanation, group members perceive the group ideal to be more extreme

than their individual positions. In an attempt to gain acceptance and to be perceived more favorably

by the group, members end up exaggerating their personal preferences.

Figure 8 shows that team-managed funds tend to have higher nontraditional share than

individually managed funds. Nontraditional share of experienced individuals is virtually flat

throughout our sample period. Inexperienced individuals appear to increase their nontraditional share

somewhat early on in 2003 and 2004, but otherwise their holdings of securitizations are mostly flat

throughout the boom. Inexperienced individuals do appear to dramatically reduce their holdings in

early 2009. In contrast, team-managed funds and, in particular, inexperienced team-managed funds

stand out. These funds more than double the share of their portfolios allocated to nontraditional

managers active at the time. According to career-concerns stories, prior learning about manager ability would result in weaker performance-flow relationship and weaker incentives for experienced managers to invest in securitizations. Although career concerns and learning about manager ability could potentially explain the importance of being active during 1998, they are inconsistent with experience not mattering for managers with less than 7 years of experience. 34 This is related to the idea of “persuasion bias” explored in DeMarzo, Zwiebel, and Vayanos (2003).

27

securitizations. Once the crisis starts, their holdings fall rapidly back to the 2003-2004 levels. Table 9

formalizes these results in a regression setting. Team-managed mutual funds have 2.7% higher

nontraditional share in 2007 than individually managed funds.

These results are consistent with the idea that team-managed funds take on exposure to

nontraditional securitizations due to bad beliefs. Moreover, the results in Table 7 show that team-

managed funds do not face stronger performance-flow relationships. Thus, differences in incentives

are unlikely to explain our results.

Overall, our analysis of the cross section of mutual fund holdings of securitizations is more

consistent with bad beliefs narratives of the boom than bad incentives narratives. The securitizations

holdings of mutual funds do not appear to vary with the strength of the performance-flow

relationships they face. However, funds managed by teams and young individuals take more risk,

consistent with bad beliefs narratives. Of course, this evidence is simply suggestive, not definitive.

Moreover, mutual funds may not face the type of incentive problems relevant for bad incentives

narratives of the boom. Therefore, we next turn to our data on insurance company holdings.

C. Insurance Companies

Insurance companies are the single largest institutional holder of credit securities. Moreover,

they are large regulated financial intermediaries, so their holdings may give us insight into the

behavior of other such intermediaries including banks and broker dealers.

C.1 Variance Decomposition

We again start by a simple variance decomposition of the share of nontraditional

securitizations in an insurers’ portfolio. As with mutual funds above, the results in Table 10 suggest

that variation across insurers, as opposed to common variation over time, is crucial for understanding

insurer holdings of securitizations. Specifically, common variation over time can explain less than

1% of total variation in the portfolio share of nontraditional securitizations. Insurer fixed effects, on

the other hand, explain more than 58% of total variation.

Among insurers, life insurers account for roughly 18% of corporate bond holdings, while

property and casualty (P&C) insurers account for 3%. Figure 9a shows that P&C insurers tend to

invest less in nontraditional securitizations than life insurers, though the time-series patterns are

similar for both types. Holdings fall in 2003, rise from 2004-2008, and then fall afterwards. The

difference in levels between P&C and life insurance holdings of nontraditional securitizations is in

28

line with the general tendency of P&C insurers to have safer portfolios. For instance, according to the

2007Q2 Flow of Funds, Treasuries and Agencies made up 17% of the total assets of P&C insurers. In

contrast, Treasuries and Agencies made up 10% of total life insurance assets. These differences can

be explained by differences in regulations and the demand for liquidity of P&C and life insurers.

Specifically, P&C insurers have a stronger precautionary demand for liquidity because the payouts on

their policies tend to be lumpier and harder to predict than the payouts on life policies. Indeed, as we

will see in Section V.B, P&C insurers tend to transact more than life insurers, particularly in their

holdings of GSE MBS.

C.2 Financial Strength

We now turn to more detailed analysis of cross-sectional determinants of insurance company

holdings. Insurance companies offer an interesting laboratory for testing different versions of the bad

incentives view. There are a number of frictions that could generate agency problems within

insurance companies. For instance, equity holders are residual claimants on the values of insurers’

asset portfolios net of policyholder claims. Thus, insurers’ capitalization levels and financial strength

ratings may reflect the strength of the agency conflict between their equity holders and their creditors.

To combat this problem, regulation requires insurance companies to maintain minimum levels of

capital on a risk-adjusted basis. That is, like banks, insurers with riskier portfolios are subject to

higher capital requirements. However, to the extent that these regulations are imperfect, less well-

capitalized insurers may have stronger risk-shifting incentives, and therefore be more likely to hold

nontraditional securitizations.

To test this hypothesis, in Figure 9c we split insurers by their capital-assets ratios as of 2003.

Since capital-assets ratios are correlated with size, we define high-capital insurers to be those with

above median capital-assets ratios within a given size decile (based on total assets). We hold the

classification constant over time to ensure that the results are not driven by a compositional shift in

the set of insurers that have low versus high capital ratios.

Figure 9c shows insurers with low and high capital ratios initially have similar holdings of

nontraditional securitizations in 2003. Over the course of the boom, both types of insurers increase

their portfolio allocations to securitizations. However, insurers with low capital ratios increase their

portfolio weights far more than insurers with high capital ratios. This is consistent with the notion

that these poorly capitalized insurers engaged in risk-shifting. Tables 11 and 12 formalize this

29

evidence in regressions. Table 11 shows that the level of nontraditional share is negatively correlated

with insurer capital ratios, with the magnitude of the effects growing over time. The effect is

economically significant. The standard deviation of the capital-assets ratio is around 20%, so that a

one standard deviation increase in capitalization is associated with a 2.4% (percentage points) lower

nontraditional share in 2007, relative to a mean nontraditional share of 2.2%. Panel B of Table 11

shows that the results also hold in differences. Between 2003 and 2007, insurers with low capital

ratios increase their nontraditional share more than insurers with high capital ratios.

Of course, the capital-assets ratio is a crude proxy for insolvency risk, which is the ultimate

driver of risk-shifting behavior. For instance, insurers may have a similar risk of insolvency, but

optimally have different capital because their policy liabilities have different risks. To address this

concern, we obtain data from A.M. Best on the financial strength of insurers. A.M. Best is a credit

rating agency dedicated to the insurance industry. Its credit ratings are based on a quantitative and

qualitative evaluation of an insurer’s ability to meet its ongoing policy and contract obligations.

In Figure 9b, we split insurers into those with high and low financial strength ratings as of

2003. To ensure that the results are not driven by size, we again define highly-rated insurers to be

those with above median ratings within a given size decile. The figure shows a pattern similar to our

splits based on capital ratios. Low-rated insurers have initial portfolio allocations to nontraditional

securitizations similar to those of high-rated insurers. However, over the course of the boom, low-

rated insurers increase their nontraditional shares more than high-rated insurers. Table 12 formalizes

this evidence in a regression setting. Again, the magnitudes are economically and statistically

significant. In 2007, highly-rated insurers have nontraditional shares 2.3 percentage points lower than

low-rated insurers, relative to a mean nontraditional share of 2.4%. Overall, although not conclusive,

the evidence is consistent with the notion that financially weaker insurance companies engaged in

risk-shifting by purchasing nontraditional securitizations.

C.3 Organizational Form

Insurers are organized as either stock companies with external shareholders or mutuals which

are owned by policyholders. These differences in organizational form may create differences in the

strength of the agency conflict between equity holders and other claimants. Specifically, in the case

of mutual companies, policy holders effectively provide both equity and debt, which may reduce the

scope for risk-shifting.

30

To examine this difference, Figure 9d compares the nontraditional shares of mutual insurance

companies versus non-mutuals. Note that here we cannot adjust for size in the way we did when

examining insurer capitalization, since insurers are either mutuals or not. Nevertheless, Figure 9d

shows that mutual insurance companies tend to have lower nontraditional shares than non-mutuals

and this difference increases over the 2003-2007 boom in securitization. Table 13 shows that this

difference is not statistically significant. However, the magnitudes are still economically meaningful.

Mutuals have about a nontraditional share about 1% lower than non-mutuals of the same size,

consistent with the notion that agency conflicts between external shareholders and insurance

company policyholders could affect holdings of nontraditional securitizations.

In addition to variation in their organizational form, insurers also vary in the way they manage

their investment portfolios. Specifically, insurance companies sometimes outsource the management

of their portfolios to external managers. The difference between internally managed portfolios and

externally managed portfolios may shed light on the impact of agency frictions on portfolio

allocations to nontraditional securitizations. On one hand, external managers may have incentives to

take excessive risk due to performance-based compensation. On the other hand, internal risk

management may bring its own set of agency problems.

To examine this difference, Figure 9f compares the nontraditional shares of internally

managed versus externally managed insurance portfolios. Again, we cannot adjust for size here since

portfolios are either internally managed or not. Nevertheless, Figure 9f shows that internally managed

portfolios tend to have higher nontraditional shares than externally managed portfolios during the

2003-2007 boom in securitizations. This is consistent with the idea that externally managed portfolios

are more subject to stricter limits on portfolio allocations, possibly due to agency conflicts. Table 14

shows that the difference in levels is not statistically significant. However, the magnitudes of the

point estimates are still economically meaningful. Externally managed portfolios have about a

nontraditional share about 1% lower than internally managed portfolios for insurers of the same size.

In summary, our analysis of insurance companies suggests that incentive problems may have

played an important role in driving their demand for nontraditional securitizations.

V. The Fall

Overall, our evidence suggests that both bad beliefs and bad incentives played important roles

31

in driving the boom in securitization. We now turn to the collapse of the market for securitizations

from 2007 to 2009.

A. Aggregate Evidence

As discussed in the hypothesis development section, the conventional macro-finance view of

the collapse is that there were significant fire sales of securitizations by leveraged intermediaries

during this period. This view suggests that we should observe high transaction volume during the

crisis, as the leveraged sector sells securitizations to the unlevered investors in our data. In contrast,

narratives of a “buyers’ strike” or market freeze suggest there should be little transaction volume

during the crisis.

Figure 10 shows the time series of insurer transactions by collateral type. We plot turnover,

the absolute value of transactions (i.e. buys + sells) normalized by total holdings. Thus, if an insurer

sells its entire existing portfolio and buys a portfolio of new securities, turnover will be 200%. We

take a four-quarter moving average to smooth the data series. As reference points, we also plot

turnover for GSE MBS, which are in our eMAXX data, and corporate bonds.35

The figure shows that, during the boom, securitizations generally traded less than GSE MBS

and more than corporate bonds. Turnover in nonprime RMBS and traditional consumer ABS is close

to turnover in GSE MBS pre-crisis, while CMBS and CDOs trade substantially less. As the crisis sets

in, trade in all types of structured falls substantially. In contrast, turnover of GSE MBS and corporate

bonds remains relatively constant throughout the crisis. Overall, there is little evidence that real

money investors bought large amounts of securitizations in the crisis. There was little trade, and

overall par dollars of these products held by mutual fund and insurers fall during the crisis.36

Figure 11 shows similar patterns for mutual funds. For mutual funds, we do not directly

observe transactions; we only observe changes in holdings from one quarter to the next. We therefore

impute transactions by adjusting the change in the holdings of each security for the change in the

security’s overall outstanding amount due to defaults and prepayments. We scale sales and purchases

35 We obtain data on insurance company corporate bond transactions from Mergent FISD. The underlying source of the data is the same set of regulatory filings that eMAXX collects our securitizations transaction data from. 36 The fall in par dollars held does not reflect trade. It instead reflects prepayments and defaults. When the collateral in a securitization prepays or defaults, the (par) dollar holdings of an investor in the securitization naturally fall. The crisis was associated with both defaults and mortgage prepayments. We use data from Bloomberg on total CUSIP-level amounts outstanding to verify that the decline in holdings is driven by prepayments rather than trade. Insurers and mutual funds hold a constant fraction of the amount outstanding, but the total outstanding declines due to defaults and prepayments.

32

by average holdings during the previous and current quarters, and then take a four-quarter moving

average to smooth the data series. The figure shows that the turnover of GSE MBS is about twice as

high as the turnover of non-agency asset-backed securities. The turnover of nonprime RMBS trends

down over time from around 35% in late 2004 to around 10% by the end of the sample period. The

decline in nonprime RMBS turnover is especially pronounced during the crisis: it falls from 27% in

2007Q2 to 14% in 2008Q4. The turnover of CMBS and CDOs is relatively stable before the start of

the crisis in the second half of 2007 and experiences similarly large declines during the crisis. Finally,

the turnover of GSE MBS and traditional consumer ABS picks up during the crisis, potentially

consistent with funds choosing to transact in these more liquid securities.

These aggregate facts are inconsistent with simple fire sales stories like Kyotaki and Moore

(1997), Geanakoplos (2009), Brunnermeier and Pedersen (2009), Shleifer and Vishny (1992), and

Stein (2012). In most fire-sales narratives, outsized declines in prices are mediated by high trade

volume. Unless the demand curve for securitizations was very inelastic, the tiny volume of secondary

trade we observe seems insufficient to significantly depress equilibrium prices.37 Overall, given that

both prices and trading volume drop substantially, the buyers’ strike narrative seems to be more

consistent with the data.

One issue that arises with our analysis is that we do not observe the full universe of holders or

transactions. Thus, we can only make direct statements about transactions where at least one party is

either an insurance company or mutual fund. However, as discussed above, our data covers an

economically significant portion of the universe of securitizations, and insurers and mutual funds

make up a large fraction of total credit investor assets. One possibility is there were wide-spread fire-

sales that led to a reallocation of securitizations within the class of levered intermediaries, but not

between leveraged and unlevered intermediaries. However, to the extent that fire sales generally

result in the transfer of security holdings from one broad class of financial intermediary (levered

investors) to another (unlevered investors), it is natural to think that our data should capture at least

one side of the transaction. 37 For instance, one way to generate an extremely inelastic demand curve in a fire-sale setting would be to posit two types of investors: natural buyers with high valuations and unnatural buyers with very low valuations. Suppose the supply of the risky asset is 100 units, all initially held by natural buyers with high valuations. Now, if some shock forces natural buyers to sell even 1 unit of their holdings, unnatural buyers with low valuations will become the marginal holders and price will plummet precipitously. Of course, this is a knife-edge result. It relies on the risk-bearing capacity of natural buyers being just enough to hold the entire asset supply before the shock.

33

Overall, the evidence appears most consistent with the notion that the collapse of the

secondary market for securitizations was driven by a buyer’s strike rather than a fire sale. In models

of buyers’ strikes, price declines spiral without any trade. While collapse happens instantly in the

models discussed in Section II, one can imagine the mechanisms spiraling more slowly over the

course of the crisis in practice. For instance, in adverse selection driven models, increases in the

uncertainty of uninformed investors might dampen trading, in turn causing uncertainty to rise further.

B. Cross-sectional Evidence

We next turn to cross-sectional analyses of our transactions data. While the aggregate data

show that the total volume of transactions was small during the bust, there may still be cross-sectional

evidence that certain insurers or mutual funds were forced to sell during bust.38

We begin by analyzing insurance company transactions in Table 15. The table shows that

larger insurers tended to buy more nontraditional securitizations during the boom, consistent with

their higher holdings of those products. There is some evidence that larger insurers also tended to

increase their purchases in the crisis, but the magnitude of the effect is quite small. Otherwise, there

is little evidence that insurers were forced to sell nontraditional securitizations during the bust. If

anything, there is evidence that stronger insurers with higher financial strength ratings sold more.

These are unlikely to be forced sales. Insurers appear to be much more active in transacting in the

GSE MBS market. Large insurers are net sellers of GSE MBS in the boom and become increasing net

sellers during the crisis.

We next turn to mutual fund transactions in Table 16. The table shows that mutual funds

experiencing inflows tend to make net purchases of nontraditional securitizations during the boom, a

trend that does not change in the crisis. However, mutual funds experiencing outflows during the

boom do not tend to sell their nontraditional securitizations. This trend continues in the crisis—there

is no evidence that mutual funds suffering outflows are forced to sell their nontraditional

securitizations. Mutual funds with high nontraditional share of course purchased more nontraditional

securitizations during the boom. These funds ceased their purchases in the bust, but there is no

evidence that they actively sold off their holdings during the crisis.

Like insurers, mutual funds appear to be much more active transacting in the GSE MBS 38 Indeed, Merrill, Nadauld, Stulz, and Sherland (2012) provide evidence that P&C insurers sold some nonprime residential mortgage backed securities at fire sale discounts during the crisis.

34

market. The sensitivity of GSE MBS purchases to inflows and outflows is much higher than the

sensitivity of nontraditional securitization purchases. When mutual funds have inflows, they tend to

increase their GSE MBS purchases. When they suffer outflows, they tend to sell GSE MBS.

Moreover, these sensitivities increase in the crisis.

Overall, the results are consistent with the Scholes (2000) model of liquidity management.

Mutual funds and insurers that faced liquidity needs transacted in the most liquid securitization

market, the market for GSE MBS. If there were forced sales in any secondary markets for

securitizations, it seems likely that those forced sales would take place in the market for GSE MBS.

Indeed, some have argued that the Federal Reserve’s first quantitative easing program was so

effective because it provided liquidity in the market for GSE MBS at a time when there was high

demand for liquidity in that market.39 Our results are consistent with this notion.

VI. Conclusion

Nontraditional securitizations—nonprime RMBS and CDOs—were at the heart of the recent

financial crisis. The demand for these securities fed the housing boom during the early and mid-

2000s, while declines in their prices during 2007-2008 generated large losses for financial

intermediaries, ultimately imperiling their soundness and triggering a full-blown crisis. We have

many anecdotal narratives and theoretical models of the rise and fall of nontraditional securitizations,

but little systematic empirical evidence. Using novel micro-data on insurance companies’ and mutual

funds’ holdings of both traditional and nontraditional securitizations, this paper begins to shed light

on the factors that drove the rise and fall of securitization.

We document a number of robust stylized facts that should inform future theoretical and

empirical work on the crisis. First, insurance companies and mutual funds as a whole were below

market weight in securitizations, in particular nontraditional securitizations. In 2007Q2, right before

the start of the crisis, insurance companies and mutual funds held 27.3% of all private credit

securities but only 4.3% of nontraditional securitizations. This suggests that the proximate surge in

demand for structured products did not stem from these real-money investors. 39 See Sack (2009), who argues that one reason the Federal Reserve’s first quantitative easing program had a “large impact on MBS rates, is that the market began from a point of substantial spreads—ones that were well above market norms. These wide spreads could have reflected poor liquidity and an elevated liquidity premium on these securities.” Krishnamurthy and Vissing-Jorgensen (2013) make a similar argument about the first round of Fed asset purchases.

35

Second, heterogeneity across security classes and investors is key to understanding the crisis.

While spreads on AAA-rated traditional securitizations were negligible, spreads on AAA-rated

nontraditional securitizations were much wider, in line with spreads on BBB-rated corporate bonds.

This suggests a more nuanced view of the boom than the simple tellings of either bad beliefs or bad

incentives theories, which tend to lump all AAA-rated securitizations together. There were some

investors who understood and differentiated between the AAA tranches of traditional and

nontraditional securitizations. Simple variance decomposition underlines the importance of investor

heterogeneity: variation across investors as opposed to common variation over time is crucial for

understanding our holdings data.

The behavior of mutual funds is more consistent with bad beliefs than with bad incentives.

We find little evidence that funds facing stronger performance-flow relationship took on more risk by

investing in nontraditional securitizations. Despite not facing stronger performance-flow relationship,

inexperienced mutual fund managers invested significantly more in these products than did more

experienced managers. This is consistent with the idea that inexperienced managers are more prone to

bad beliefs. Supporting this idea is a discrete shift away from nontraditional securitizations by

managers who were active during the market dislocations of fall 1998.

Turning to insurance companies, we find that holdings of nontraditional securitizations are

concentrated among insurers that are poorly capitalized and have low financial strength ratings. In

addition, insurance companies that manage their portfolios internally hold more nontraditional

securitizations than those whose portfolios are managed externally, and insurance companies

organized as mutuals tend to hold fewer nontraditional securitizations than those organized as stock

companies. These findings are consistent with the idea that incentive conflicts play an important in

role in explaining insurer holdings of nontraditional securitizations. In summary, our evidence

suggests that both bad beliefs and bad incentives are likely part of the explanation for the pre-crisis

surge in investor demand for these products.

In contrast to multiple narratives of the boom, explanations of the bust center on a single

narrative of fire sales—the forced sales of securities by investors with high valuations to investors

with low valuations. A central, yet largely untested, prediction of fire sales models is that the decline

in prices was associated with increased trade between levered intermediaries and other investors. We

find no evidence of increased transaction volume as prices of securitizations declined throughout late

36

2007 and 2008. Instead, turnover of nontraditional securitizations experienced large declines during

the crisis. Turnover of GSE MBS, and to a lesser extent of traditional consumer ABS, increased; and

to the extent that mutual funds suffered outflows, they met redemptions through sales of GSE MBS.

Our results therefore challenge the importance of fire sales during the crisis, or at least the canonical

models of fire sales, and suggest that the collapse of the secondary market may be better described as

a buyers’ strike.

We hope that future research builds on the stylized facts documented in our paper to shed

further light on the rise and fall of securitization.

37

References Acharya, Viral V., Thomas Cooley, Matthew Richardson, and Ingo Walter, 2010, Manufacturing Tail Risk: A Perspective on the Financial Crisis of 2007-09, Foundations and Trends in Finance 4, 247-325. Acharya, Viral V., Thomas Cooley, Matthew Richardson, and Ingo Walter, 2010, Market Failures and Regulatory Failures: Lessons from Past and Present Crises, Masahiro Kawai and Eswar Prasad, ed.: Financial Sector Regulation and Reforms in Emerging Markets (Brookings Institution Press). Acharya, Viral V., Marco Pagano, and Paolo Volpin, 2013, Seeking Alpha: Excess Risk Taking and Competition for Managerial Talent, NBER Working Paper 18891. Acharya, Viral V, Philipp Schnabl, and Gustavo A Suarez, 2012, Securitization without risk transfer, Journal of Financial Economics 107, 515-536. Adrian, Tobias, and Hyun Song Shin, 2013, Procyclical Leverage and Value at Risk, Review of Financial Studies, forthcoming. Ashcraft, Adam B., and Til Schuermann, 2008, Understanding the Securitization of Subprime Mortgage Credit, Foundations and Trends in Finance 2, 191-309. Ashcraft, Adam B., Paul Goldsmith-Pinkham, and James Vickery, 2010, MBS ratings and the mortgage credit boom, Federal Reserve Bank of New York Staff Report 449. Bär, Michaela, Conrad S. Ciccotello, and Stefan Ruenzi, 2010, Risk management and team-managed mutual funds, Journal of Risk Management in Financial Institutions 4, 57–73. Barnett-Hart, Anna Katherine, 2009, The Story of the CDO Market Meltdown: An Empirical Analysis, Harvard College Undergraduate Thesis. Becker, Bo, and Victoria Ivashina, 2012, Reaching for Yield in the Bond Market, Working Paper. Beltratti, Andrea, and Rene M. Stulz, 2012, The credit crisis around the globe: Why did some banks perform better? Journal of Financial Economics 105, 1–17. Bernanke, Ben S., 2005, Monetary Policy in a World of Mobile Capital, Cato Journal 25, 1-12. Bernanke, Ben S., Carol Bertaut, Laurie Pounder DeMarco, and Steven Kamin, 2011, International Capital Flows and the Returns to Safe Assets in the United States, 2003-2007, Board of Governors of the Federal Reserve System International Finance Discussion Paper 1014. Black, Fisher, 1986, Noise, The Journal of Finance 41, 529–543. Bliss, Richard T, Mark E Potter, and Christopher Schwarz, 2008, Performance Characteristics of Individually-Managed versus Team-Managed Mutual Funds, Journal of Portfolio Management 34, 110-119.

38

Brunnermeier, Markus, and Yuliy Sannikov, 2013, A Macroeconomic Model with a Financial Sector, American Economic Review, in press. Brunnermeier, Markus, and Lasse Pedersen, 2009, Market Liquidity and Funding Liquidity, Review of Financial Studies 22, 2201–2238. Caballero, Ricardo, 2010, The “Other” Imbalance and the Financial Crisis, NBER Working Paper 15636. Caballero, Ricardo and Emmanuel Farhi, 2013, A Model of the Safe Asset Mechanism (SAM): Safety Traps and Economic Policy, Working Paper. Chernenko, Sergey, and Aditya Vikram Sunderam, 2013, Frictions in Shadow Banking: Evidence from the Lending Behavior of Money Market Mutual Funds, Working Paper. Chevalier, Judith, and Glenn Ellison, 1997, Risk Taking by Mutual Funds as a Response to Incentives, Journal of Political Economy 105, 1167–1200. Chevalier, Judith, and Glenn Ellison, 1999, Career concerns of mutual fund managers, The Quarterly Journal of Economics 114, 389–432. Cordell, Larry, Yilin Huang, and Meredith Williams, 2012, Collateral damage: Sizing and assessing the subprime CDO crisis, Federal Reserve Bank of Philadelphia Working Paper 11-30. Coval, Joshua, Standard & Poor's, 2007, Asset fire sales (and purchases) in equity markets, Journal of Financial Economics 86, 479–512. Coval, Joshua, Jakub Jurek, and Erik Stafford, 2009, The Economics of Structured Finance, Journal of Economic Perspectives 23: 3-25. Dang, Tri Vi, Gary B. Gorton, and Bengt Holmstrom, 2011, Ignorance, Debt and Financial Crises, Working Paper. Dávila, Eduardo, 2011, Dissecting Fire Sales Externalities, Working Paper. De Long, J. Bradford, Andrei Shleifer, Lawrence H. Summers, and Robert J. Waldmann, 1990, Noise Trader Risk in Financial Markets, The Journal of Political Economy 98, 703-738. Ellul, Andrew, and Vijay Yerramilli, 2012, Stronger Risk Controls, Lower Risk: Evidence from U.S. Bank Holding Companies, Journal of Finance, Forthcoming. Erel, Isil, Taylor Nadauld, Rene M. Stulz, 2013, Why Did U.S. Banks Invest in Highly-Rated Securitization Tranches? Working Paper. Fabozzi, Frank J., 2005, The Structured Finance Market: An Investor’s Perspective, Financial Analysts Journal, 27-40.

39

Fahlenbrach, Rüdiger, and René M. Stulz, 2011, Bank CEO Incentives and the Credit Crisis, Journal of Financial Economics 99, 11-26. Geanakopolos, John, 2009, The Leverage Cycle, in D. Acemoglu, K. Rogoff, and M. Woodford, eds.: NBER Macroeconomics Annual 2009, pp. 1-65 (University of Chicago Press). Geanakopolos, John, 2010, Incorporating Financial Features into Macroeconomics: Discussion, in Macroeconomic Challenges: The Decade Ahead. Jackson Hole, Federal Reserve Bank of Kansas City Economic Policy Symposium, 2010. Gennaioli, Nicola, Andrei Shleifer, and Robert Vishny, 2012, Journal of Financial Economics, Journal of Financial Economics 104, 452–468. Gerardi, Kristopher, Andreas Lehnert, Shane M. Sherlund, and Paul Willen, 2008, Making Sense of the Subprime Crisis, Brookings Papers on Economic Activity 39(2), 69-159. Gorton, Gary B., 2008, The Panic of 2007, in Maintaining Stability in a Changing Financial System, Proceedings of the 2008 Jackson Hole Conference, Federal Reserve Bank of Kansas City, 2008. Gorton, Gary B., and Andrew Metrick, 2010, Regulating the Shadow Banking System, Brookings Papers on Economic Activity 2010, 261-312. Gorton, Gary B., and Andrew Metrick, Haircuts, Federal Reserve Bank of St. Louis Review 92(6). Greenwood, Robin, and Stefan Nagel, 2009, Inexperienced investors and bubbles, Journal of Financial Economics 93, 239–258. Hanson, Samuel Gregory, and Aditya Vikram Sunderam, 2013, Are There Too Many Safe Securities? Securitization and the Incentives for Information Production, Journal of Financial Economics 108, 565–584. Harrison, J Michael, and David M. Kreps, 1978, Speculative Investor Behavior in a Stock Market with Heterogeneous Expectations, The Quarterly Journal of Economics 92(2), 323-36. Hong, Harrison, and Jeremy C. Stein, 2003, Differences of Opinion, Short-Sales Constraints and Market Crashes, Review of Financial Studies 16, 487-525. Iannotta, Giuliano, and George Pennacchi, 2012, Bank Regulation, Credit Ratings, and Systematic Risk, Working Paper. Isenberg, Daniel J., 1986, Group polarization: A critical review and meta-analysis, Journal of Personal and Social Psychology 50(6), 1141-1151. Jensen, Michael, and William H. Meckling, 1976, Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure, Journal of Financial Economics 3, 305-360. Kacperczyk, Marcin, and Philipp Schnabl, 2013, How safe are money market funds? Quarterly Journal of Economics 128, 1073–1122.

40

Keys, Benjamin J., Tanmoy K. Mukherjee, Amit Seru, and Vikrant Vig, 2009, Did Securitization Lead to Lax Screening? Evidence from Subprime Loans, Quarterly Journal of Economics 125, 307-362. Kiyotaki, Nobuhiro, and John Moore, 1997, Credit Cycles, Journal of Political Economy 105, 211-248. Landier, Augustin, David Sraer, and David Thesmar, 2011. The Risk-Shifting Hypothesis, IDEI Working Paper 699. Lorenzoni, G., 2008, Inefficient Credit Booms, Review of Economic Studies 75, 809-833. Merrill, Craig, Taylor Nadauld, Rene M Stulz, and Shane M Sherlund, 2013, Why did financial institutions sell RMBS at fire sale prices during the financial crisis? Charles A. Dice Working Paper 2013-2. Milbradt, Konstantin, 2012, Level 3 Assets: Booking Profits, Concealing Losses, Review of Financial Studies 25, 55-95. Miller, Edward M., 1977, Risk, Uncertainty, and Divergence of Opinion, The Journal of Finance 32, 1151-1168. Myers, David G, and Helmut Lamm, 1976, The Group Polarization Phenomenon, Psychological Bulletin 83, 602–627. Morris, Stephen, and Hyun Song Shin, 2012, Contagious Adverse Selection, American Economic Journal: Macroeconomics 4, 1-21. Rajan, Uday, Amir Seru, and Vikrant Vig, 2012, The Failure of Models that Predict Failure: Distance, Incentives and Defaults, Journal of Financial Economics, accepted. Scheinkman, Jose, and Wei Xiong, 2003, Overconfidence and speculative bubbles, Journal of Political Economy 111, 1183–1219. Shin, H. “Global Banking Glut and Loan Risk Premium”, Mundell-Fleming Lecture, IMF Economic Review 60 (2), 155-192, 2012 Shleifer, Andrei, and Robert Vishny, 1992, Liquidation Values and Debt Capacity: A Market Equilibrium Approach, Journal of Finance 47, 1343-1366. Shleifer, Andrei and Robert Vishny, 2011, Fire Sales in Finance and Macroeconomics, Journal of Economic Perspectives 25, 29-48.

Figure 1Issuance of Traditional and Nontraditional Securitizations

This figure shows quarterly issuance of traditional and nontraditional securitizations in SDC data.Traditional securitizations include CMBS, prime RMBS, consumer ABS, and other ABS. Nontra-ditional securitizations include nonprime RMBS and CDOs.

(a) Traditional vs. Nontraditional

0

100

200

300

400

Pro

ceed

s ($

bill

ion)

1988

q4

1990

q4

1992

q4

1994

q4

1996

q4

1998

q4

2000

q4

2002

q4

2004

q4

2006

q4

2008

q4

Traditional Non-traditional

Traditional vs. Non-Traditional Securitization

(b) Traditional

0

20

40

60

80

100

Pro

ceed

s ($

bill

ion)

1998

q4

1999

q4

2000

q4

2001

q4

2002

q4

2003

q4

2004

q4

2005

q4

2006

q4

2007

q4

2008

q4

CMBS Prime RMBSConsumer ABS Other ABS

Traditional Securitization

(c) Nontraditional

0

100

200

300

Pro

ceed

s ($

bill

ion)

1998

q4

1999

q4

2000

q4

2001

q4

2002

q4

2003

q4

2004

q4

2005

q4

2006

q4

2007

q4

2008

q4

Nonprime RMBS CDO

Nontraditional Securitization

(d) CDO

0

20

40

60

80

100

Pro

ceed

s ($

bill

ion)

1998

q4

1999

q4

2000

q4

2001

q4

2002

q4

2003

q4

2004

q4

2005

q4

2006

q4

2007

q4

2008

q4CBO CLO ABS CDO

Collateralized Debt Obligations

42

Figure 2Credit Spreads on Securitizations and Corporate Bonds, 2003-2007

This figure plots the credit spreads on newly issued securitizations versus the spreads on corpo-rate bonds. Each quarter we compute the value-weighted average credit spread on securitizations,based on the underlying collateral. We compute this average for traditional consumer asset-backedsecurities backed by credit card loans, automotive loans, student loans, and other personal loans(“Consumer ABS”); private residential mortgage backed securities (RMBS) backed by subprime andAlt-A mortgages (“Non-Prime RMBS”); commercial mortgage backed securities (CMBS); and col-lateralized debt obligations (“CDOs”). We compute value-weighted averages of new-issue spreads,restricting attention to the spreads on floating rate notes indexed to 1-, 3-, and 6-month LIBORto avoid benchmarking issues. In Panel A, we plot the credit spreads on newly issued AAA-ratedsecuritizations. For reference, we plot the average secondary spreads over LIBOR (based on in-terest rate swaps) on 3-year AAA-rated and BBB-rated corporate bonds from Barclays. In PanelB, we plot the credit spreads on newly issued BBB-rated securitizations alongside the spreads onBBB-rated, BB-rated and B-rated corporate bonds.

(a) AAA

0

20

40

60

80

100

bps

2003q22003q4

2004q22004q4

2005q22005q4

2006q22006q4

2007q22007q4

AAA Consumer ABS AAA CMBSAAA Non-Prime RMBS AAA CDOsAAA Corporate BBB Corporate

(b) BBB

050

100150200250300350400450500

bps

2003q22003q4

2004q22004q4

2005q22005q4

2006q22006q4

2007q22007q4

BBB Consumer ABS BBB CMBSBBB Non-Prime RMBS BBB CDOsBBB Corporate BB CorporateB Corporate

43

Figure 3Ex-Ante Spreads versus Realized Loss Rates

This figure shows realized loss rates versus average spreads during the boom, 2003–2007. Realizedloss rates are from Moody’s (2012). Corporate spreads are from Bloomberg. For securitizations, wecompute value-weighted averages of new-issue spreads for securities in our eMAXX data, restrictingattention to the spreads on floating rate notes indexed to 1-, 3-, and 6-month LIBOR to avoidbenchmarking issues.

(a)

AAA Corp

orate

AA Corp

orate

A Corp

orate

BBB Corp

orate

BB Corp

orate

B Corp

orate

CCC Cor

pora

te

010

2030

40

Cum

ulat

ive

5 yr

Los

s

0 2 4 6 8Average Spread in Boom

Cumulative 5 yr Loss Fitted values

Corporate Spreads

(b)

AAA Corp

orate

BBB Corp

orate

BB Corp

orate

AAA Con

sumer

ABS

AAA CMBS

AAA Non

-prim

e RMBS

AAA CDO

05

1015

2025

Cum

ulat

ive

5 yr

Los

s

0 .5 1 1.5Average Spread in Boom

AAA Securitization vs. Corporate Spreads

(c)

AAA Corp

orate

AA Corp

orate

A Corp

orate

BBB Corp

orate

BB Corp

orate B C

orpora

te

BBB Con

sumer

ABS

BBB CMBS

BBB Non

-prim

e RMBS

BBB CDO

010

2030

4050

6070

Cum

ulat

ive

5 yr

Los

s

0 .5 1 1.5 2 2.5Average Spread in Boom

BBB Securitization vs. Corporate Spreads

44

Figure 4Rating Composition of Mutual Fund Portfolios

For each mutual fund-quarter observation we calculate the fraction of the portfolio invested intranches with di↵erent ratings. We then calculate the average across funds. We do this separatelyfor di↵erent types of securitizations: consumer ABS, CDO, CMBS, and nonprime RMBS. We useat-issuance ratings from Moody’s and restrict the sample to deals rated by Moody’s. NR meansthat the overall securitization deal was rated by Moody’s but the specific tranche was not.

(a) Consumer ABS

0

.2

.4

.6

.8

2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4

Aaa Aa A BaaBa B Caa NR

(b) CDO

0

.1

.2

.3

.4

.5

2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4

Aaa Aa A BaaBa B Caa NR

(c) CMBS

0

.2

.4

.6

.8

1

2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4

Aaa Aa A BaaBa B Caa NR

(d) Nonprime RMBS

0

.2

.4

.6

.8

2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4

Aaa Aa A BaaBa B Caa NR

45

Figure 5Rating Composition of Insurance Company Portfolios

For each insurance firm-quarter observation we calculate the fraction of the portfolio invested intranches with di↵erent ratings. We then calculate the average across insurance firms. We do thisseparately for di↵erent types of securitizations: consumer ABS, CDO, CMBS, and nonprime RMBS.We use at-issuance ratings from Moody’s and restrict the sample to deals rated by Moody’s. NRmeans that the overall securitization deal was rated by Moody’s but the specific tranche was not.

(a) Consumer ABS

0

.2

.4

.6

.8

1

2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4

Aaa Aa A BaaBa B Caa NR

(b) CDO

0

.2

.4

.6

2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4

Aaa Aa A BaaBa B Caa NR

(c) CMBS

0

.2

.4

.6

.8

1

2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4

Aaa Aa A BaaBa B Caa NR

(d) Nonprime RMBS

0

.2

.4

.6

.8

1

2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4

Aaa Aa A BaaBa B Caa NR

46

Figure 6Nontraditional Share of Variable Annuities versus Regular Mutual Funds

The sample consists of matched pairs of variable annuities and regular mutual funds that have thesame investment objectives and are managed by the same portfolio managers. We form the sampleof matched pairs as of 2003Q4.

80

100

120

140

160

# Fu

nd P

airs

.02

.03

.04

.05

.06

.07

Non

tradi

tiona

l sha

re

2003q3 2004q3 2005q3 2006q3 2007q3 2008q3 2009q3 2010q3

Variable Annuity Funds Regular Mutual Funds# Fund Pairs

47

Figure 7Mutual Fund Manager Experience and Nontraditional Share

This figure shows the average nontraditional share of mutual funds managed by experienced versusinexperienced portfolio managers. We split mutual funds into two groups based on the medianvalue of manager experience in Panel A and of manager age in Panel B. Manager characteristicsare measured as of 2004Q4. For team-managed funds we take the average of individual managers.

(a) Experience

.02

.03

.04

.05

.06

.07

.08

Non

tradi

tiona

l sha

re

2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4

Inexperienced Experienced

(b) Age

.02

.03

.04

.05

.06

.07

.08

Nont

radi

tiona

l sha

re2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4

Young Old

48

Figure 8Experience and Nontraditional Share in Individually- versus Team-Managed FundsThis figure shows the average nontraditional share of mutual funds managed by experienced versusinexperienced individual managers and teams. We separately split individually- and team-managedfunds into experienced and inexperienced ones based on their median value of experience. Man-ager characteristics are measured as of 2004Q4. For team-managed funds we take the average ofindividual managers.

.02

.04

.06

.08

.1

Non

tradi

tiona

l sha

re

2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4

Inexperienced individual Experienced individualInexperienced team Experienced team

49

Figure 9Insurance Company Characteristics and Nontraditional Share

This figure shows the average nontraditional share of insurance companies split based on di↵erentcharacteristics. Rating is the 2002 financial strength rating from AM Best.

(a) Type

0.000

0.020

0.040

0.060

0.080

2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4

Life P&C

(b) Rating

0.030

0.040

0.050

0.060

0.070

0.080

2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4

Weak rating Strong rating

(c) Capital

0.030

0.040

0.050

0.060

0.070

2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4

Weak capital Strong capital

(d) Mutual

0.030

0.040

0.050

0.060

2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4

Non mutual Mutual

(e) Public

0.035

0.040

0.045

0.050

0.055

0.060

2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4

Private Public

(f) External

.03

.04

.05

.06

2003q3 2004q3 2005q3 2006q3 2007q3 2008q3 2009q3 2010q3

Internal External

50

Figure 10Insurance Company Portfolio Turnover

Turnover is the absolute value of transactions, buys and sells, normalized by average holdings duringquarters t � 1 and t. Transactions in private securitizations and in GSE MBS are from eMAXX.Transactions in corporate bonds are from Mergent FISD.

(a) ABS

.05

.1

.15

.2

.25

.3

2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4

Agencies Corporates ABS

(b) CDO

0

.1

.2

.3

2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4

Agencies Corporates ABS

(c) CMBS

.05

.1

.15

.2

.25

.3

2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4

Agencies Corporates ABS

(d) Nonprime RMBS

.05

.1

.15

.2

.25

.3

2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4

Agencies Corporates ABS

51

Figure 11Mutual Fund Portfolio Turnover

Turnover is the absolute value of transactions, buys and sells, normalized by average holdings duringquarters t� 1 and t. Sales and purchases are calculated as the change in par holdings adjusted forthe change in security’s outstanding amount using outstanding factors from Bloomberg. Interest-and principal-only tranches are excluded using the following screens: a) zero original balance fromany of the three rating agencies; b) tranche name; c) fund’s holdings of a security are greater than150% of outstanding balance; d) security weight in the fund’s portfolio is greater than 10%; e)outstanding factor is less than 0.01 or greater than 2; f) fund’s holdings of the security are greaterthan $100 million. The sample of funds consists of bond and hybrid funds matched to CRSP.

0

.2

.4

.6

.8

2003q4 2004q4 2005q4 2006q4 2007q4 2008q4 2009q4 2010q4

GSE MBS Consumer ABSCDO CMBSNonprime RMBS

52

Table 1Outstanding asset-backed securities and holdings of insurers and mutual funds by underlying collateral type, 2003-2010

This table reports estimates of the outstanding balance of US asset-backed securities by underlying collateral type. The sourcesfor estimated outstanding are listed in the footnotes. For each collateral type, we compute aggregate par holdings by insurers and mutualfunds using our eMaxx holdings data. For the market, % refers to the share of a given collateral type in the aggregate outstandingamount of all private credit securities. For insurance and mutual funds, % refers to the shares of a given collateral type in the aggregateportfolios of insurance firms and mutual funds.

All private credit securitiesa Traditional consumer ABSb CMBSc

Mutual Mutual MutualMarket Insurance Funds Market Insurance Funds Market Insurance Funds$ % $ % $ % $ % $ % $ % $ % $ % $ %

2003Q2 6,776 100.0% 1,755 25.9% 504 7.4% 701 10.3% 93 5.3% 22 4.4% 313 4.6% 108 6.1% 10 2.0%2003Q4 7,178 100.0% 1,829 25.5% 531 7.4% 721 10.0% 100 5.5% 20 3.8% 343 4.8% 119 6.5% 11 2.0%2004Q2 7,524 100.0% 1,927 25.6% 560 7.4% 726 9.7% 101 5.3% 25 4.4% 368 4.9% 130 6.7% 11 2.0%2004Q4 8,071 100.0% 2,002 24.8% 594 7.4% 729 9.0% 98 4.9% 24 4.1% 394 4.9% 145 7.2% 13 2.1%2005Q2 8,541 100.0% 2,060 24.1% 631 7.4% 762 8.9% 90 4.4% 24 3.8% 439 5.1% 153 7.4% 15 2.3%2005Q4 8,852 100.0% 2,088 23.6% 660 7.5% 781 8.8% 89 4.3% 28 4.3% 506 5.7% 168 8.1% 21 3.2%2006Q2 9,415 100.0% 2,127 22.6% 706 7.5% 794 8.4% 84 4.0% 23 3.3% 553 5.9% 185 8.7% 19 2.7%2006Q4 10,103 100.0% 2,097 20.8% 764 7.6% 836 8.3% 76 3.6% 24 3.1% 614 6.1% 189 9.0% 24 3.2%2007Q2 10,879 100.0% 2,132 19.6% 837 7.7% 870 8.0% 64 3.0% 27 3.2% 697 6.4% 195 9.1% 38 4.5%2007Q4 11,577 100.0% 2,145 18.5% 886 7.7% 894 7.7% 68 3.2% 26 2.9% 758 6.5% 216 10.1% 41 4.6%2008Q2 11,698 100.0% 2,143 18.3% 958 8.2% 894 7.6% 85 4.0% 34 3.6% 741 6.3% 222 10.4% 48 5.0%2008Q4 11,170 100.0% 2,085 18.7% 956 8.6% 835 7.5% 75 3.6% 27 2.8% 718 6.4% 223 10.7% 50 5.3%2009Q2 11,676 100.0% 2,146 18.4% 1,035 8.9% 808 6.9% 68 3.2% 26 2.5% 703 6.0% 221 10.3% 48 4.7%2009Q4 11,654 100.0% 2,226 19.1% 1,121 9.6% 798 6.8% 70 3.1% 23 2.1% 671 5.8% 227 10.2% 52 4.6%2010Q2 11,432 100.0% 2,288 20.0% 1,181 10.3% 744 6.5% 75 3.3% 32 2.7% 647 5.7% 220 9.6% 58 4.9%2010Q4 11,816 100.0% 2,353 19.9% 1,243 10.5% 690 5.8% 68 2.9% 28 2.3% 617 5.2% 208 8.8% 59 4.8%

a Data on the outstanding quantity of all private credit securities, which include corporate bonds, foreign bonds, and privately issued ABS is fromtable L.212 ‘Corporate and Foreign Bonds’ of the Federal Reserve’s Financial Accounts of the United States (i.e. the Z1 release which was formerlyknown as the Flow of Funds Accounts). Data on the holding of insurers (which includes both P&C and life insurers) and mutual funds (which excludesmoney market mutual funds) is also from Table L.212.

b Data on the outstanding quantity of traditional consumer ABS, which consists of ABS collateralized by credit card debt, automotive loans, studentloans, manufactured housing and equipment loans, is from SIFMA athttp://www.sifma.org/uploadedFiles/Research/Statistics/StatisticsFiles/SF-US-ABS-SIFMA.xls.

c Data on the outstanding quantity of CMBS, which consists of MBS collateralized by commercial mortgages and multifamily residential mortgages,is from tables L.219 ‘Multifamily Residential Mortgages’ and Table L.220 ‘Commercial Mortgages’ of the Federal Reserve’s Financial Accounts of theUnited States.

53

Table 2Outstanding asset-backed securities and holdings of insurers and mutual funds by underlying collateral type, 2003-2010

This table reports estimates of the outstanding balance of US asset-backed securities by underlying collateral type. The sourcesfor estimated outstanding are listed in the footnotes. For each collateral type, we compute aggregate par holdings by insurers and mutualfunds using our eMaxx holdings data. For the market, % refers to the share of a given collateral type in the aggregate outstandingamount of all private credit securities. For insurance and mutual funds, % refers to the shares of a given collateral type in the aggregateportfolios of insurance firms and mutual funds.

All Private RMBSa Subprime RMBSb CDOsc

Mutual Mutual MutualMarket Insurance Funds Market Insurance Funds Market Insurance Funds$ % $ % $ % $ % $ % $ % $ % $ % $ %

2003Q2 297 4.4% 29 1.6% 6 1.3% 563 8.3% 73 4.2% 8 1.6% 279 4.1% 21 1.2% 4 0.8%2003Q4 276 3.8% 22 1.2% 5 0.9% 601 8.4% 56 3.1% 8 1.5% 302 4.2% 20 1.1% 5 0.9%2004Q2 314 4.2% 29 1.5% 10 1.7% 765 10.2% 65 3.4% 10 1.8% 347 4.6% 21 1.1% 6 1.0%2004Q4 351 4.3% 51 2.5% 10 1.7% 929 11.5% 83 4.2% 14 2.4% 391 4.8% 23 1.1% 5 0.9%2005Q2 388 4.5% 55 2.7% 11 1.7% 1,139 13.3% 96 4.7% 16 2.5% 454 5.3% 21 1.0% 3 0.5%2005Q4 425 4.8% 62 3.0% 13 2.0% 1,349 15.2% 109 5.2% 18 2.7% 517 5.8% 24 1.2% 2 0.4%2006Q2 462 4.9% 69 3.2% 12 1.7% 1,579 16.8% 117 5.5% 14 2.0% 627 6.7% 24 1.1% 2 0.2%2006Q4 499 4.9% 74 3.5% 15 1.9% 1,808 17.9% 122 5.8% 16 2.1% 795 7.9% 23 1.1% 2 0.2%2007Q2 525 4.8% 73 3.4% 19 2.2% 1,852 17.0% 115 5.4% 19 2.3% 961 8.8% 21 1.0% 2 0.3%2007Q4 552 4.8% 84 3.9% 25 2.8% 1,895 16.4% 130 6.1% 17 1.9% 1,005 8.7% 21 1.0% 2 0.3%2008Q2 518 4.4% 83 3.9% 23 2.4% 1,729 14.8% 138 6.4% 18 1.9% 1,004 8.6% 27 1.3% 4 0.4%2008Q4 484 4.3% 82 3.9% 22 2.3% 1,563 14.0% 136 6.5% 16 1.7% 1,018 9.1% 27 1.3% 2 0.2%2009Q2 436 3.7% 74 3.4% 18 1.8% 1,437 12.3% 126 5.9% 13 1.3% 987 8.5% 26 1.2% 2 0.2%2009Q4 389 3.3% 67 3.0% 16 1.5% 1,311 11.2% 125 5.6% 11 1.0% 936 8.0% 26 1.2% 1 0.1%2010Q2 350 3.1% 62 2.7% 20 1.7% 1,221 10.7% 119 5.2% 13 1.1% 872 7.6% 25 1.1% 2 0.1%2010Q4 311 2.6% 54 2.3% 22 1.8% 1,131 9.6% 107 4.6% 14 1.1% 838 7.1% 24 1.0% 2 0.1%

a Data on the outstanding quantity of private RMBS, which consists of private MBS collateralized by first lien single-family residential mortgages, isfrom table L.218 ‘Home Mortgages’ of Federal Reserve’s Financial Accounts of the United States.

b Data on the outstanding quantity of subprime RMBS, a subset of all private RMBS, is from SIFMA athttp://www.sifma.org/uploadedFiles/Research/Statistics/StatisticsFiles/SF-US-ABS-SIFMA.xls.Specifically, we use the series for ‘home equity ABS’—market-speak for subprime RMBS.

c Data on the outstanding quantity of CBOs, CDOs, and CLOs is from SIFMA athttp://www.sifma.org/uploadedFiles/Research/Statistics/StatisticsFiles/SF-Global-CDO-SIFMA.xls.SIFMA provides data on global CBOs, CDOs, and CLOs outstandings. Based on SIFMA’s issuance data by currency, we assume that 75% ofoutstanding CDOs and CLOs are USD-denominated.

54

Table 3Variance Decomposition of Mutual Funds’ Nontraditional Share

This table reports the R

2 from the regressions of mutual funds’ nontraditional share on variousfixed e↵ects. The sample of funds consists of bond and hybrid mutual funds. The sample periodis 2003Q1–2010Q4.

Objective ⇥ FamilyDate Objective Date Fund All 5 Funds 10 Funds

Adjusted R2 0.017 0.145 0.175 0.623 0.312 0.231 0.116N 26550 26550 26550 26550 24331 18454 11762

55

Table 4Performance-Flow Relationship in Variable Annuity versus Regular Mutual Funds

This table reports the results of the performance-flow regressions for variable annuity versus regularmutual funds. The sample consists of all bond and hybrid funds in CRSP Mutual Fund Database.The sample period is January 2009–June 2013. Fund flows are winsorized at the 1st and 99thpercentiles in columns 1–2 and at the 5th and 95th percentiles in columns 3–4. Objective-monthdate fixed e↵ects are included in all specifications. Standard errors are adjusted for clustering byfund in columns 2–4. *, **, and *** indicate statistical significance at 10%, 5%, and 1%.

(1) (2) (3) (4)Variable annuity �0.003⇤⇤⇤ �0.003⇤⇤ �0.001 �0.000

(0.001) (0.002) (0.001) (0.001)Rett�1⇥ Regular 0.192⇤⇤⇤ 0.192⇤⇤⇤ 0.121⇤⇤⇤ 0.111⇤⇤⇤

(0.025) (0.034) (0.016) (0.018)Rett�1⇥ Variable annuity 0.057⇤ 0.057 0.057⇤⇤⇤ 0.044⇤⇤

(0.031) (0.036) (0.016) (0.018)Log(TNAt�1) �0.001⇤⇤⇤

(0.000)Rett�1⇥ Log(TNAt�1) 0.005⇤⇤⇤

(0.002)Constant 0.018⇤⇤⇤ 0.018⇤⇤⇤ 0.009⇤⇤⇤ 0.012⇤⇤⇤

(0.000) (0.001) (0.000) (0.001)N 323208 323208 323208 323208Adjusted R

2 0.016 0.016 0.036 0.038

56

Table 5Summary Statistics: Mutual Fund Managers

This table reports summary statistics for mutual fund managers in our data. The unit of observationis fund-quarter. The sample of funds consists of bond and hybrid mutual funds. The sampleperiod is 2003Q1–2010Q4. Fund manager characteristics are measured as of December 31, 2004.

N Mean Median SD Min MaxTeam managed 23377 0.7 1.0 0.5 0.0 1.0Number of fund managers 23377 2.7 2.0 2.5 0.0 21.0Age 15379 45.2 44.0 7.1 29.0 71.0Experience 22832 8.5 7.9 4.5 0.0 29.1

57

Table 6Nontraditional Share of Mutual Funds

This table reports the results of the regressions of mutual funds’ nontraditional share on fund andportfolio manager characteristics. Experienced is a binary variable equal to 1 for funds above themedian of experience. Manager characteristics are measured as of 2004Q4. Retail share is the shareof retail share classes in fund assets. Family retail share is the share of retail share classes in fundfamily assets. Family ETF, equity, and taxable shares are the shares of ETFs, equity funds, andtaxable bond funds in fund family assets. The omitted category is municipal funds. Objective fixede↵ects are included. Standard errors are adjusted for clustering by fund. *, **, and *** indicatestatistical significance at 10%, 5%, and 1%.

2004 2005 2006 2007 2008 2009Experienced �0.006⇤ �0.009⇤⇤ �0.020⇤⇤⇤ �0.034⇤⇤⇤ �0.023⇤⇤⇤ �0.001

(0.004) (0.004) (0.006) (0.007) (0.007) (0.006)Log(TNA) 0.002 0.003 0.005⇤ 0.005⇤ �0.001 �0.003

(0.002) (0.002) (0.003) (0.003) (0.003) (0.002)Log(Family TNA) 0.002 0.001 �0.000 �0.002 �0.001 �0.001

(0.001) (0.001) (0.002) (0.002) (0.002) (0.002)Log(Fund age) �0.002 �0.010⇤⇤ �0.011⇤ �0.013⇤ 0.003 0.006

(0.003) (0.004) (0.007) (0.008) (0.006) (0.005)Retail share 0.007 0.019⇤⇤ 0.020⇤ 0.010 �0.009 0.009

(0.007) (0.008) (0.010) (0.010) (0.010) (0.014)Family retail share 0.003 �0.019⇤ �0.035⇤⇤ �0.006 0.018 �0.021

(0.008) (0.010) (0.014) (0.015) (0.016) (0.022)Family ETF share 0.005 0.018 0.017 0.049 �0.068⇤⇤ �0.096⇤⇤

(0.024) (0.034) (0.035) (0.045) (0.029) (0.038)Family equity share 0.003 �0.003 0.001 0.019 0.019 0.026

(0.012) (0.027) (0.047) (0.028) (0.036) (0.032)Family taxable share 0.031⇤⇤ 0.026 0.033 0.087⇤⇤ 0.097⇤⇤ 0.054

(0.014) (0.028) (0.048) (0.036) (0.044) (0.034)Constant �0.006 0.042 0.062 0.055⇤ 0.024 0.023

(0.014) (0.032) (0.054) (0.031) (0.035) (0.034)N 3102 2969 2915 2935 2888 2706Adjusted R

2 0.287 0.303 0.272 0.185 0.177 0.118

58

Table 7Manager Experience and Performance-Flow Relationship

This table reports the results of monthly performance-flow regressions for bond and hybrid mutualfunds in our sample. The sample period is 2003–2010. Fund manager characteristics are measuredas of December 31, 2004. Fund flows are winsorized at the 1st and 99th percentiles. Objective-datefixed e↵ects are included. Heteroskedasticity robust standard errors are reported. *, **, and ***indicate statistical significance at 10%, 5%, and 1%.

(1) (2) (3) (4)Returnt�1 0.368⇤⇤⇤ 0.379⇤⇤⇤ 0.367⇤⇤⇤ 0.362⇤⇤⇤

(0.037) (0.043) (0.037) (0.040)Log(TNA)t�1 �0.002⇤⇤⇤ �0.001⇤⇤⇤ �0.002⇤⇤⇤ �0.002⇤⇤⇤

(0.000) (0.000) (0.000) (0.000)Returnt�1⇥ Log(TNA)t�1 �0.016⇤⇤⇤ �0.015⇤⇤⇤ �0.016⇤⇤⇤ �0.016⇤⇤⇤

(0.004) (0.004) (0.004) (0.005)Old 0.001⇤⇤

(0.000)Returnt�1⇥ Old 0.005

(0.017)Experienced 0.001⇤

(0.000)Returnt�1⇥ Experienced 0.010

(0.015)Team managed �0.001⇤⇤⇤

(0.000)Returnt�1⇥ Team managed 0.010

(0.015)Constant 0.012⇤⇤⇤ 0.010⇤⇤⇤ 0.012⇤⇤⇤ 0.014⇤⇤⇤

(0.001) (0.001) (0.001) (0.001)N 131465 88402 131465 134186Adjusted R

2 0.056 0.052 0.056 0.054

59

Table 8Nonlinearity of Experience

This table reports the results of the regressions of mutual funds’ nontraditional share on portfoliomanager experience defined using di↵erent cuto↵s. Manager characteristics are measured as of2004Q4. The sample period is 2007Q1–2007Q4. Same controls as in Table 6 are included but notreported. Objective fixed e↵ects are included. Standard errors are adjusted for clustering by fund.*, **, and *** indicate statistical significance at 10%, 5%, and 1%.

Started managing mutual funds before January 12001 2000 1999 1998 1997 1996 1995

Experienced �0.000 �0.002 �0.005 �0.014⇤ �0.032⇤⇤⇤ �0.023⇤⇤⇤ �0.027⇤⇤⇤

(0.011) (0.009) (0.008) (0.007) (0.007) (0.007) (0.006)N 2935 2935 2935 2935 2935 2935 2935Adjusted R

2 0.164 0.164 0.164 0.167 0.182 0.173 0.176

60

Table 9Teams, Experience, and Nontraditional Share

This table reports the results of the regressions of mutual funds’ nontraditional share on fund andportfolio manager characteristics. The sample period is 2007Q1–2007Q4. Experienced is a binaryvariable equal to 1 for funds above the median of experience. Manager characteristics are measuredas of 2006Q4. Retail share is the share of retail share classes in fund assets. Family retail share

is the share of retail share classes in fund family assets. Family ETF, equity, and taxable sharesare the shares of ETFs, equity funds, and taxable bond funds in fund family assets. The omittedcategory is municipal funds. Objective fixed e↵ects are included. Standard errors are adjusted forclustering by fund. *, **, and *** indicate statistical significance at 10%, 5%, and 1%.

(1) (2) (3) (4) (5) (6)Log(Experience) �0.004 �0.009

(0.004) (0.010)Log(Manager age) �0.068 �0.053

(0.041) (0.050)Experienced �0.021⇤⇤⇤ �0.022⇤⇤

(0.007) (0.011)Experience percentile �0.027⇤⇤⇤

(0.010)Team 0.027⇤⇤

(0.013)Experienced ⇥ Team 0.005

(0.015)Log(TNA) 0.006⇤⇤ 0.006 0.007 0.007⇤⇤ 0.007⇤⇤ 0.007⇤⇤

(0.003) (0.004) (0.004) (0.003) (0.003) (0.003)Log(Family TNA) �0.003 �0.005 �0.004 �0.003 �0.003 �0.002

(0.003) (0.004) (0.004) (0.003) (0.003) (0.003)Log(Fund age) �0.012⇤ �0.017 �0.016 �0.012⇤ �0.012⇤ �0.011

(0.007) (0.010) (0.010) (0.007) (0.007) (0.007)Retail share 0.005 0.004 0.004 0.003 0.004 0.004

(0.010) (0.013) (0.013) (0.010) (0.010) (0.010)Family retail share �0.007 0.004 0.004 �0.005 �0.005 �0.002

(0.015) (0.020) (0.020) (0.015) (0.015) (0.015)Family ETF share 0.081⇤ �0.944 �0.836 0.086⇤ 0.083⇤ 0.091⇤

(0.046) (0.722) (0.714) (0.047) (0.047) (0.047)Family equity share �0.022 �0.029 �0.026 �0.021 �0.020 �0.014

(0.038) (0.054) (0.054) (0.038) (0.038) (0.037)Family taxable share 0.045 0.062 0.066 0.047 0.048 0.055

(0.039) (0.051) (0.052) (0.039) (0.038) (0.038)Constant 0.087⇤ 0.365⇤⇤ 0.322 0.084⇤ 0.085⇤ 0.050

(0.050) (0.183) (0.203) (0.049) (0.049) (0.050)N 3224 2187 2187 3224 3224 3224Adjusted R

2 0.163 0.181 0.182 0.170 0.166 0.182

61

Table 10Variance Decomposition of Insurance Companies’ Nontraditional Share

This table reports the R2 from the regressions of insurance companies’ nontraditional share on date,firm, and financial group fixed e↵ects. Financial group fixed e↵ect refers to the ultimate parentaccording to AM Best.

Date Firm GroupAdjusted R2 0.002 0.582 0.458N 77272 77272 77272

62

Table 11Nontraditional Share and Capital Ratio

This table reports the results of the regressions of insurance companies’ nontraditional share. Thechange in nontraditional share is calculated relative to 2004. Size is the log of the par value ofthe bond portfolio. Standard errors are adjusted for clustering by firm. *, **, and *** indicatestatistical significance at 10%, 5%, and 1%.

2004 2005 2006 2007 2008 2009Panel A: Nontraditional Share

Capital/Assets2003 �0.071⇤⇤⇤ �0.098⇤⇤⇤ �0.119⇤⇤⇤ �0.122⇤⇤⇤ �0.117⇤⇤⇤ �0.099⇤⇤⇤

(0.017) (0.019) (0.021) (0.025) (0.027) (0.031)Size 0.002 0.003⇤ 0.003 0.004⇤ 0.005⇤⇤ 0.003

(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)Constant 0.024 0.020 0.033 0.027 0.021 0.033

(0.019) (0.018) (0.020) (0.023) (0.023) (0.025)N 967 891 834 742 732 732Adjusted R

2 0.144 0.217 0.205 0.210 0.189 0.111Panel B: Change in Nontraditional Share

Capital/Assets2003 �0.025⇤⇤⇤ �0.039⇤⇤⇤ �0.045⇤⇤⇤ �0.037⇤ �0.020(0.009) (0.015) (0.016) (0.020) (0.026)

Size 0.003⇤⇤ 0.003 0.004⇤ 0.005⇤⇤ 0.003⇤

(0.001) (0.002) (0.002) (0.002) (0.002)Constant �0.018 �0.009 �0.016 �0.022 �0.011

(0.013) (0.019) (0.021) (0.021) (0.021)N 886 824 733 721 712Adjusted R

2 0.130 0.082 0.110 0.086 0.027

63

Table 12Nontraditional Share and Financial Strength Rating

This table reports the results of the regressions of insurance companies’ nontraditional share. Thechange in nontraditional share is calculated relative to 2004. Strong rating is a binary variableequal to one for firms whose 2003 AM Best financial strength rating is above the median. Size isthe log of the par value of the bond portfolio. Standard errors are adjusted for clustering by firm.*, **, and *** indicate statistical significance at 10%, 5%, and 1%.

2004 2005 2006 2007 2008 2009Panel A: Nontraditional Share

Strong rating 0.003 �0.012 �0.018⇤ �0.023⇤⇤ �0.027⇤⇤ �0.021⇤

(0.008) (0.009) (0.010) (0.011) (0.012) (0.012)Size 0.004⇤⇤ 0.007⇤⇤⇤ 0.008⇤⇤⇤ 0.009⇤⇤⇤ 0.010⇤⇤⇤ 0.008⇤⇤⇤

(0.002) (0.002) (0.002) (0.003) (0.003) (0.003)Constant �0.014 �0.032⇤ �0.030 �0.038 �0.041⇤ �0.026

(0.016) (0.019) (0.021) (0.023) (0.024) (0.025)N 831 768 726 652 639 633Adjusted R

2 0.064 0.131 0.127 0.160 0.173 0.101Panel B: Change in Nontraditional Share

Strong rating �0.013⇤⇤ �0.019⇤⇤ �0.025⇤⇤⇤ �0.029⇤⇤⇤ �0.023⇤⇤

(0.006) (0.008) (0.008) (0.009) (0.010)Size 0.004⇤⇤⇤ 0.005⇤⇤ 0.006⇤⇤⇤ 0.007⇤⇤⇤ 0.005⇤⇤

(0.002) (0.002) (0.002) (0.002) (0.002)Constant �0.030⇤⇤ �0.026 �0.035⇤ �0.038⇤⇤ �0.023

(0.013) (0.017) (0.019) (0.019) (0.021)N 766 722 650 635 621Adjusted R

2 0.231 0.160 0.237 0.230 0.117

64

Table 13Nontraditional Share and Mutual Organization Form

This table reports the results of the regressions of insurance companies’ nontraditional share. Thechange in nontraditional share is calculated relative to 2004. Mutual insurance firms are owned bypolicyholders. Size is the log of the par value of the bond portfolio. Standard errors are adjustedfor clustering by firm. *, **, and *** indicate statistical significance at 10%, 5%, and 1%.

2004 2005 2006 2007 2008 2009Panel A: Nontraditional Share

Mutual �0.001 �0.003 �0.004 �0.010 �0.009 �0.003(0.011) (0.017) (0.020) (0.019) (0.021) (0.026)

Size 0.005⇤⇤⇤ 0.007⇤⇤⇤ 0.007⇤⇤⇤ 0.008⇤⇤⇤ 0.009⇤⇤⇤ 0.007⇤⇤⇤

(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)Constant �0.016 �0.034⇤ �0.033 �0.040⇤ �0.042⇤⇤ �0.026

(0.016) (0.018) (0.021) (0.021) (0.021) (0.023)N 1045 1109 1093 997 1117 1251Adjusted R

2 0.070 0.106 0.088 0.105 0.103 0.058Panel B: Change in Nontraditional Share

Mutual 0.001 0.000 �0.005 �0.005 0.000(0.009) (0.012) (0.012) (0.014) (0.019)

Size 0.004⇤⇤ 0.004⇤⇤ 0.006⇤⇤ 0.006⇤⇤⇤ 0.004⇤

(0.002) (0.002) (0.002) (0.002) (0.002)Constant �0.031⇤⇤ �0.032 �0.042⇤ �0.043⇤⇤ �0.023

(0.015) (0.021) (0.022) (0.021) (0.023)N 954 880 778 766 757Adjusted R

2 0.094 0.055 0.084 0.073 0.024

65

Table 14Nontraditional Share and External Managers

This table reports the results of the regressions of insurance companies’ nontraditional share. Thechange in nontraditional share is calculated relative to 2004. External manager is a binary variableequal to one for firms whose bond portfolios are managed by una�liated external managers. Aninsurance company and an asset management firm are considered to be una�liated if they havedi↵erent ultimate parents in Capital IQ. Size is the log of the par value of the bond portfolio.Standard errors are adjusted for clustering by firm. *, **, and *** indicate statistical significanceat 10%, 5%, and 1%.

2004 2005 2006 2007 2008 2009Panel A: Nontraditional Share

External manager 0.007 �0.005 �0.010 �0.008 �0.009 �0.004(0.005) (0.007) (0.008) (0.009) (0.009) (0.010)

Size 0.002 0.006⇤⇤⇤ 0.007⇤⇤⇤ 0.007⇤⇤⇤ 0.008⇤⇤⇤ 0.007⇤⇤⇤

(0.001) (0.002) (0.002) (0.002) (0.002) (0.002)Constant �0.001 �0.060⇤⇤⇤ �0.063⇤⇤ �0.062⇤⇤ �0.066⇤⇤ �0.052⇤

(0.021) (0.023) (0.032) (0.027) (0.028) (0.030)N 2300 2265 2203 2088 2165 2219Adjusted R

2 0.021 0.094 0.085 0.091 0.083 0.043Panel B: Change in Nontraditional Share

External manager �0.004 �0.007 �0.005 �0.008 �0.004(0.005) (0.007) (0.007) (0.008) (0.009)

Size 0.003⇤⇤⇤ 0.005⇤⇤ 0.005⇤⇤ 0.005⇤⇤ 0.004⇤

(0.001) (0.002) (0.002) (0.002) (0.002)Constant �0.044⇤⇤ �0.055⇤ �0.054⇤ �0.054 �0.041

(0.020) (0.032) (0.029) (0.033) (0.034)N 2100 2007 1858 1855 1847Adjusted R

2 0.086 0.069 0.069 0.061 0.025

66

Table 15Cross-Sectional Determinants of Insurance Company Transactions

This table reports the results of the regressions of net purchases of nontraditional securitizations andof GSE MBS on various insurance company characteristics. Net purchases are scaled by the laggedpar value of the insurance firm’s fixed income portfolio. Crisis is 2007Q3–2009Q2. External manageris a binary variable equal to one for firms whose bond portfolios are managed by una�liated externalmanagers. An insurance company and an asset management firm are considered to be una�liatedif they have di↵erent ultimate parents in Capital IQ. Strong rating is a binary variable equal toone for firms whose 2003 AM Best financial strength rating is above the median. P&C is a binaryvariable equal to one for property and casualty insurance. The omitted caterogy is life insurance.Date fixed e↵ects are included. Standard errors are adjusted for clustering by firm. *, **, and ***indicate statistical significance at 10%, 5%, and 1%.

Net purchases ofNontraditional securitizations GSE MBS(1) (2) (3) (4) (5) (6) (7) (8)

Size 0.020⇤⇤ 0.025⇤⇤⇤ 0.025⇤⇤⇤ 0.016⇤ �0.426⇤⇤⇤�0.398⇤⇤⇤�0.406⇤⇤⇤�0.424⇤⇤⇤

(0.008) (0.008) (0.009) (0.009) (0.034) (0.032) (0.035) (0.036)Size ⇥ Crisis 0.017 0.027⇤⇤ 0.027⇤⇤ 0.020 �0.183⇤⇤⇤�0.123⇤⇤�0.169⇤⇤⇤�0.192⇤⇤⇤

(0.014) (0.013) (0.014) (0.016) (0.067) (0.060) (0.062) (0.068)External 0.025 0.281⇤

(0.039) (0.158)External ⇥ Crisis �0.142 �0.236

(0.088) (0.277)Strong capital 0.055 0.153

(0.036) (0.134)Strong capital ⇥ Crisis 0.025 �0.058

(0.074) (0.222)Strong rating 0.036 �0.178

(0.042) (0.155)Strong rating ⇥ Crisis �0.127⇤ 0.324

(0.068) (0.269)P&C �0.028 0.624⇤⇤⇤

(0.041) (0.151)P&C ⇥ Crisis �0.010 �0.093

(0.084) (0.283)Constant 0.110⇤⇤ 0.013 0.043 0.138⇤⇤ 5.107⇤⇤⇤ 4.778⇤⇤⇤ 5.005⇤⇤⇤ 4.973⇤⇤⇤

(0.052) (0.042) (0.046) (0.061) (0.287) (0.227) (0.237) (0.280)N 41956 39467 36063 44514 41956 39467 36063 44514Adjusted R

2 0.005 0.007 0.006 0.006 0.029 0.029 0.031 0.030

67

Table 16Cross-Sectional Determinants of Mutual Fund Transactions

This table reports the results of the regressions of net purchases of nontraditional securitizations andof GSE MBS on mutual fund flows. Net purchases of nontraditional securitizations are calculatedas the change in par holdings adjusted for the change in security’s outstanding amount usingoutstanding factors from Bloomberg. Net purchases of GSE MBS are calculated similarly exceptthat we use aggregate instead of security-specific prepayment information. Specifically we calculateaggregate prepayment speed as gross issuance of Fannie and Freddie MBS minus the change inoutstanding amount. Net purchases are scaled by the lagged par value of the mutual fund’s bondportfolio. Crisis is 2007Q3–2009Q2. Fund flows are scaled by lagged total net assets, and arewinsorized at the 1st and 99th percentiles. Inflows are equal to max(0,Fund flows). Outflows areequal to max(0,�Fund flows). Size is the log of the par value of the bond portfolio. Objective-datefixed e↵ects are included. Standard errors are adjusted for clustering by fund. *, **, and ***indicate statistical significance at 10%, 5%, and 1%.

Net purchases ofNontraditionalsecuritizations GSE MBS

(1) (2) (3) (4)Inflows 0.216⇤⇤⇤ 0.187⇤⇤⇤ 0.423⇤⇤⇤ 0.390⇤⇤⇤

(0.048) (0.046) (0.054) (0.054)Inflows ⇥ Crisis �0.076 �0.040 0.235⇤⇤ 0.245⇤⇤⇤

(0.072) (0.071) (0.094) (0.094)Outflows �0.032 �0.105 �0.194⇤⇤⇤ �0.263⇤⇤⇤

(0.071) (0.069) (0.063) (0.063)Outflows ⇥ Crisis 0.056 0.040 �0.283⇤ �0.273⇤

(0.097) (0.096) (0.150) (0.159)Size �0.005⇤⇤⇤ �0.006⇤⇤⇤

(0.001) (0.001)Size ⇥ Crisis 0.004⇤⇤⇤ 0.001

(0.001) (0.001)Nontraditional share 0.176⇤⇤⇤ 0.024⇤⇤

(0.026) (0.011)Nontraditional share ⇥ Crisis �0.104⇤⇤⇤ �0.005

(0.028) (0.024)Constant 0.014⇤⇤⇤ 0.056⇤⇤⇤ 0.022⇤⇤⇤ 0.088⇤⇤⇤

(0.001) (0.007) (0.001) (0.008)N 20076 20076 23537 23537Adjusted R

2 0.047 0.083 0.072 0.078

68