Securitized banking and the run on repo

Abstract

The panic of 2007–2008 was a run on the sale and repurchase market (the repo market), which is a very large, short-term market that provides financing for a wide range of securitization activities and financial institutions. Repo transactions are collateralized, frequently with securitized bonds. We refer to the combination of securitization plus repo finance as “securitized banking” and argue that these activities were at the nexus of the crisis. We use a novel data set that includes credit spreads for hundreds of securitized bonds to trace the path of the crisis from subprime-housing related assets into markets that had no connection to housing. We find that changes in the LIB-OIS spread, a proxy for counterparty risk, were strongly correlated with changes in credit spreads and repo rates for securitized bonds. These changes implied higher uncertainty about bank solvency and lower values for repo collateral. Concerns about the liquidity of markets for the bonds used as collateral led to increases in repo haircuts, that is the amount of collateral required for any given transaction. With declining asset values and increasing haircuts, the US banking system was effectively insolvent for the first time since the Great Depression.

Introduction

The 2007–2008 financial crisis was a system wide bank run. What makes this bank run special is that it did not occur in the traditional-banking system, but instead took place in the “securitized-banking” system. A traditional-banking run is driven by the withdrawal of deposits, a securitized-banking run is driven by the withdrawal of repurchase (repo) agreements. Hence, we describe the crisis as a run on repo. The purpose of this paper is to propose a mechanism for this new kind of bank run and to provide supporting evidence for this mechanism through analysis of two novel data sets.

Traditional banking is the business of making and holding loans, with insured demand deposits as the main source of funds. Securitized banking is the business of packaging and reselling loans, with repo agreements as the main source of funds. Securitized-banking activities were central to the operations of firms formerly known as investment banks (e.g. Bear Stearns, Lehman Brothers, Morgan Stanley, and Merrill Lynch), but they also play a role at commercial banks, as a supplement to traditional-banking activities of firms such as Citigroup, J.P. Morgan, and Bank of America.¹

We argue that the financial crisis that began in August 2007 was a systemic event, to meaning that the banking sector became insolvent, in the sense that it could not pay off its debt. What happened is analogous to the banking panics of the 19th century in which depositors en masse went to their banks seeking to withdraw cash in exchange for demand and savings deposits. The banking system could not honor these demands because the cash had been lent out and the loans were illiquid so they suspended convertibility and relied on clearinghouses to issue certificates as makeshift currency.² Evidence of the insolvency of the banking system (i.e., that the system cannot pay off the demand deposits, as demanded by depositors) in these earlier episodes is the discount on these certificates. We argue that the current crisis is similar in that contagion led to withdrawals in the form of unprecedented high repo haircuts and even the cessation of repo lending on many forms of collateral. Evidence of insolvency in 2008 is the bankruptcy or forced rescue of several large firms, with other (even larger) firms requiring government support to stay in business.

To perform our analysis, we use two novel data sets, one with information on 392 securitized bonds and related assets, including many classes of asset-backed securities (ABSs), collateralized debt obligations (CDOs), and credit default swaps (CDSs) and the other repo rates and repo haircuts.³ Using these data, we are able to provide a new perspective on the contagion in this crisis. In our exposition, we use this term “contagion” specifically to mean the spread of the crisis from subprime housing assets to non-subprime assets that have no direct connection to the housing market. In fact, we argue that to explain the crisis requires explaining why the spreads on non-subprime related asset classes rose dramatically.

To provide background for our analysis, we illustrate the differences between traditional banking and securitized banking in Fig. 1, Fig. 2. Fig. 1 provides the classic picture of the financial intermediation of mortgages by the traditional-banking system. In Step A, depositors transfer money to the bank, in return for a checking or savings account from which withdrawals can be made at any time. In Step B, the bank loans these funds to a borrower, who promises to repay through a mortgage on the property. The bank then holds this mortgage on its balance sheet, along with other non-mortgage loans made to retail and commercial borrowers.

Traditional-banking runs, for the most part, were ended in United States after the Great Depression, owing to a combination of influences, including enhanced discount window lending by the Federal Reserve and the introduction of deposit insurance. Deposit insurance removes any incentive for insured depositors to withdraw their funds, but larger insured banks cannot offer insured depositors to non retail depositors (including sovereign wealth funds, mutual funds, and cash-rich companies). One solution to this problem is the securitized-banking system illustrated in Fig. 2, which takes large deposits from investors (Step 1) and then intermediates these deposits to mortgage borrowers (Steps 2 and 3) and other debtors.

Step 1 in Fig. 2 is an analogue to Step A from Fig. 1, with one important difference. In the traditional-banking system shown in Fig. 1, the deposits are insured by the government. To achieve similar protection in Step 1 of Fig. 2, the investor receives collateral from the bank. In practice, this deposit-collateral transaction takes the form of a repo agreement: the investor buys some asset (i.e., the collateral) from the bank for $X, and the bank agrees to repurchase the same asset some time later (perhaps the next day) for $Y. The percentage (Y−X)/X is the repo rate, and is analogous to the interest rate on a bank deposit. Typically, the total amount of the deposit will be some amount less than the value of the underlying asset, with the difference called a haircut. For example, if an asset has a market value of $100 and a bank sells it for $80 with an agreement to repurchase it for $88, then we would say that the repo rate is 10% [=(88−80)/80] and the haircut is 20% (100−80/100). If the bank defaults on the promise to repurchase the collateral, then the investor has the right to terminate the agreement and keep or sell the collateral.

Turning next to the lower right corner of Fig. 2, we show how the second part of the intermediation differs from traditional banking. In Fig. 1, the bank did the work of underwriting the loan itself. In Fig. 2, the bank outsources this function to a direct lender. Such lenders grew to prominence in the most recent housing boom, with a specialization of underwriting loans to be held for only a short time before being sold to banks. Much has been written about potential conflicts in this separation of the loan decision from the source of finance, but that is not our topic here. In principle, there is no reason that this separation must necessarily lead to poor underwriting, and in any event such problems do not imply anything about contagion or systemic events.

Another key component of securitized banking is in the securitization itself: the intermediation activities that transfer most of the mortgage loans to outside investors in Step 4. We discuss this step in detail in Section 2. For our purposes here, the key idea is that the outputs of this securitization are often used as collateral in Step 1, so that securitized banking is a cycle that requires all steps to keep running. In this paper, we show how this cycle broke down in the crisis.

Fig. 3 summarizes the relations between the main elements of traditional and securitized banking. The familiar elements of traditional banking are reserves, deposit insurance, interest rates on deposits, and the holding of loans on balance sheet. Bank solvency is promoted by requiring a fraction of deposits to be held in reserve, and in emergencies these reserves can be replenished by borrowing from the central bank. The analogue in securitized banking is the repo haircut, which forces banks to keep some fraction of their assets in reserve when they borrow money through repo markets. Deposit insurance is a promise made by the government to pay depositors in the event of default. The analogue in securitized banking is collateral. A bank in need of cash can raise deposit rates to attract it; the analogues for securitized banking are the repo rates. Finally, the cash raised in traditional banking is lent out, with the resulting loans held on the balance sheet. In securitized banking, funds are lent only temporarily, with loans repackaged and resold as securitized bonds. Some of these bonds are also used as collateral to raise more funds, which completes the cycle.

The run on repo is depicted in Fig. 4, which plots a haircut index from 2007 to 2008. The details of this index are explained in Section 4; for now, just think of the index as an average haircut for collateral used in repo transactions, not including US treasury securities. This index rises from zero in early 2007 to nearly 50% at the peak of the crisis in late 2008. During this time period, several classes of assets were stopped entirely from being used as collateral, an unprecedented event that is equivalent to a haircut of 100%.

To see how the increase in haircuts can drive the banking system to insolvency, take as a benchmark a repo market size of, for example, $10 trillion. With zero haircuts, this is the amount of financing that banks can achieve in the repo markets. When the weighted-average haircut reaches, say, 20%, then banks have a shortage of $2 trillion. In the crisis, some of this amount was raised early on by issuing new securities. But, this fell far short of what was needed. Furthermore, selling the underlying collateral drives asset prices down, which then reinforces the cycle: lower prices, less collateral, more concerns about solvency, and ever increasing haircuts.

In addition to repo, other short-term debt experienced runs. There were runs, in particular, on asset-backed commercial paper programs and structured investment vehicles. Papers that consider the runs on asset-backed commercial paper programs during the crisis include Covitz, Liang, and Suarez (2009) and Carey, Correa, and Kotter (2009). Aside from asset-backed commercial paper, the commercial paper of financial firms, the predominant issuers of corporate short-term commercial paper also saw withdrawals when investors refused to reinvest as the paper came due; see Kacperczyk and Schnabl (2010).⁴ Also important was the run on money market funds following the failure of Lehman Brothers, as explained by The Investment Company Institute (2009). In summary, all short-term debt markets were vulnerable during the crisis. In this paper, we focus on the repo market because of its large size. Also, repo is secured by collateral that can be “rehypothecated,” which means that a depositor of cash in the bank takes physical possession of bond collateral and then can reuse that collateral. So, the collateral has a money multiplier. When haircuts rise, the money multiplier works in reverse, causing a massive deleveraging process. This does not happen for unsecured short-term debt.

This paper and those mentioned above are part of a rapidly growing literature that tries to empirically show what happened during the crisis. Aside from runs the financial crisis is complicated in many other dimensions. Studies have been made of the breakdown of various arbitrage relations, perhaps due to counterparty risk and attendant funding problems, e.g., Coffey, Hrung, and Sarkar (2009), Gorton (2010), Baba and Packer (2009), Stanton and Wallace (2009), Fontana (2009), and Fender and Scheicher (2009). Other research looks at counterparty risk and liquidity, e.g., Arora, Gandhi, and Longstaff (2009), Schwarz (2009), and Singh and Aitken (2009). Other papers discuss the international dimensions of the crisis. and compare the crisis to previous crises, e.g., Eichengreen, Mody, Nedeljkovic, and Sarno (2009) and Reinhart and Rogoff (2008). Ivashina and Scharfstein (2008) look at bank lending during the crisis. The real effects of the crisis are also important to consider, as do e.g., Almeida, Campello, and Laranjeira (2009) and Campello, Giabona, Graham, and Harvey (2009). Many other papers look at subprime mortgages, rating agencies, auction rate securities, short selling prohibitions, and so on, so the above list is far from being complete.⁵

The remainder of the paper is organized as follows. In Section 2, we provide institutional background for our analysis, with a discussion of the growth of securitized banking, using subprime mortgages as the case study. We use this case study to provide more detail for Step 4 in Fig. 2 and to explain the mechanics of securitization and the repo market.

In Section 3, we introduce and explain the two main state variables used in the paper: the ABX index—which proxies for fundamentals in the subprime mortgage market and the LIB OIS, which is the spread between the LIBOR (London Inter Bank Offered Rate, for unsecured interbank borrowing) and the rate on an overnight interest swap (OIS) a proxy for the risk-free rate. In our analysis the LIB-OIS spread acts primarily as a proxy for counterparty risk in the banking system. We then plot these state variables for 2007 and 2008 and review the timeline for the crisis. The ABX data show that the deterioration of the subprime market began in early 2007. As is now well known, this deterioration had a direct impact on banks, which had many of these securitized assets and pre-securitized mortgages on their balance sheets. This real deterioration in bank balance sheets became apparent in the interbank markets in mid-2007, as evidenced by an upward spike in the LIB-OIS in August. This state variable remained in a historically high but narrow range until September 2008, when the events at Fannie Mae (Federal National Mortgage Association), Freddie Mac (Federal Home Loan Mortgage Corporation), Lehman, and AIG (American International Group Inc.) led to a rapid deterioration in interbank markets and increase in the LIB-OIS spread that persisted until the end of 2008.

We posit that the increased risk at banks had several interrelated effects, all of which centered on the securitized assets used as collateral in the repo market. We provide evidence for these effects, using a data set with information on securitized bonds, credit default swaps, and other assets used in repo transactions. These data were created by large financial institutions and are used for trading and portfolio valuation by a wide range of market participants. Section 3 provides summary statistics on these data and illustrates how some of these assets co-moved with the ABX and the LIB-OIS.

Section 5 gives the main empirical results of the paper. Without a structural model of repo markets, we are only able to talk about co-movement of spreads on various assets, and thus we use the language of correlation instead of causation in our empirical analysis. Section 5.1 explains our methodology and presents results for a few representative asset classes. Section 5.2 uses the full set of asset classes to demonstrate that it was the interbank markets (LIB-OIS), and not the subprime housing market (ABX), that was correlated with increases in the spreads on non-subprime securitized assets and related derivatives. These increased spreads are equivalent to a price decrease, which represents a fall in the value of collateral used in repo transactions. Then, as lenders began to fear for the stability of the banks and the possibility that they might need to seize and sell collateral, the borrowers were forced to raise repo rates and haircuts. Both of these increases occurred in the crisis. In 5.3 Repo spreads, 5.4 Repo haircuts, we find that these increases were correlated with changes in the LIB-OIS (for repo rates) and changes in the (expected future) volatility of the underlying collateral (for repo haircuts), consistent with the model of Dang, Gorton, and Holmström (2011). It is the rise in haircuts that constitutes the run on repo. An increase in a haircut is tantamount to a withdrawal from the bank, forcing deleveraging on a large scale. Section 5.5 uses data from Schwarz (2009) to confirm that the LIB-OIS relations found for credit spreads and repo rates are primarily driven by counterparty risk.

Section 6 reviews our arguments and concludes the paper. Appendix A defines some of the paper's terminology that could be unfamiliar for some readers and also includes descriptions for each of the asset classes of securitized bonds that are used in our empirical analysis. Appendix B gives more detail on the data construction.