RDP 2013-05: Liquidity Shocks and the US Housing Credit Crisis of 2007–2008 3. Literature Review

In examining the relationship between bank liquidity and lending, this paper relates to several branches of the macroeconomic literature. The theoretical literature provides a framework in which banks' financing conditions can affect overall lending due to credit market imperfections (e.g. Bernanke and Blinder 1988; Holmstrom and Tirole 1997; Stein 1998). But empirical studies face a challenge in tracing the channels through which credit supply shocks are transmitted. Traditionally, empirical research has relied on either time series or cross-sectional variation in the balance sheet positions of banks to identify the effect of bank financing conditions on lending.^[3] For example, Peek and Rosengren (1995) use US bank-level data to document a positive relationship between bank capital and credit growth during the 1990–1991 recession. However, the evidence is not compelling, as banks that face more creditworthy borrowers are likely to experience fewer loan losses. The lower losses could translate into higher levels of capital and may also encourage more lending.^[4] In other words, endogeneity could be a problem because differences in bank-level credit growth may reflect differences in the risk profile of borrowers or other demand conditions. Other studies that use instrumental variables (Paravisini 2008) or natural experiments (Peek and Rosengren 2000) generally provide more compelling evidence that liquidity supply shocks, which are exogenous to demand, affect lending. For instance, Peek and Rosengren (2000) demonstrate that US subsidiaries of Japanese banks were more likely than domestic US banks to cut credit to the US commercial real estate sector following a negative balance sheet shock to their Japanese parent. As the shock stemmed from overseas, it is likely to have been exogenous to demand conditions in the United States. However, it is still possible that the Japanese subsidiaries and the domestic US banks were lending to different pools of borrowers within the US commercial real estate sector, so that differences in demand conditions across banks could still have driven the results.

My paper belongs to a growing literature that uses loan-level information to identify the causal effect of credit supply shocks. The increasing availability of loan-level data has allowed researchers to implement more sophisticated identification strategies than empirical studies that rely on either aggregate or bank-level data. The seminal paper in this branch of the literature is Khwaja and Mian (2008). They examine the impact of liquidity shocks on bank lending by exploiting cross-bank liquidity variation induced by unanticipated nuclear tests in Pakistan in 1998. The nuclear tests caused the Pakistani Government (in anticipation of balance of payment problems) to restrict withdrawals of US dollar-denominated deposit accounts to local currency only, and at an unfavourable exchange rate. The collapse of the US dollar-denominated deposit market disproportionately affected banks that relied more on US dollar-denominated deposits for liquidity. They show that, for the same firm borrowing from two different banks, the bank exposed to the larger potential decline in liquidity was more likely to reduce lending. To the extent that the within borrower comparison fully absorbs borrower-specific changes in credit demand, the estimated difference in loan growth between banks can be attributed to differences in bank liquidity shocks. This within borrower identification scheme has now been adopted in a range of empirical studies that have access to loan-level information (e.g. Albertazzi and Marchetti 2010; Iyer et al 2010; Jimenez et al 2011; Cetorelli and Goldberg 2012; Schnabl 2012). To the best of my knowledge, this paper is the first to apply this within borrower identification strategy to household lending.

The US housing credit market provides a natural testing ground to examine the nature of credit supply shocks because it was the market at the epicentre of the global financial crisis. Most of the existing research on the current crisis has looked at the effect of changes in credit supply on the investment behaviour of large corporate borrowers (e.g. Duchin, Ozbas and Sensoy 2010; Ivashina and Scharfstein 2010; Campello et al 2012). But this is unlikely to be the primary channel through which the financial crisis affected the real economy. Instead, the prolonged period of weak economic conditions in the US economy was more likely to be due to developments in residential mortgage finance.

There are a few other recent papers that also treat the shutdown of the securitisation market in 2007 as a negative liquidity shock and examine how this affected bank lending (e.g. Gozzi and Goetz 2010; Calem, Covas and Wu 2011; Dagher and Kazimov 2012). However, my paper covers a wider cross-section of lenders, a longer time series, and utilises loan-level information which allows me to control for variation in the distribution of borrowers across banks more effectively.

Gozzi and Goetz (2010) focus on small banks that lend within their own local markets while I examine the behaviour of all lenders, regardless of size, location or geographic reach. Restricting the sample to small local banks is likely to bias the causal effect of bank liquidity shocks for two reasons. First, if affected borrowers are able to switch to large banks when small banks cut off their funding then restricting the sample to only small banks may overstate the true aggregate effect of the shock. Second, if liquidity-constrained banks are relatively more likely to cut lending to non-local borrowers, then the focus on local lending could understate the true effect of a credit supply shock.

Calem et al (2011) rely on bank-level variation in funding liquidity and lending and hence only control for unobservable variation in the characteristics of each bank's average borrower. In contrast, I control for changes in the distribution of each bank's (unobservable) borrower characteristics. This will be important if the effect of the supply shock varies across different borrowers within a bank's loan portfolio.

Dagher and Kazimov (2012) also treat the shutdown of the securitisation market as an exogenous negative liquidity shock, but use each bank's share of non-deposit funding, rather than the share of securitisation funding, to identify the treatment group of mortgage lenders. Their identification strategy also assumes that loan applicants that reside within the same metropolitan statistical area share similar characteristics, whereas my strategy assumes that applicants residing in the same Census tract are similar, which is more likely to be true given that tracts are defined based on residents sharing similar characteristics. And, unlike my study, Dagher and Kazimov do not consider which borrowers were most affected by the credit supply shock.^[5]

The housing market is also a potentially interesting area in which to identify any home bias in lending because the location of the asset (the home) is a fundamental determinant of its price (and hence its collateral risk). Geographic location is therefore potentially a significant determinant of credit risk in home mortgage lending. There is an extensive literature identifying a home bias in the global allocation of capital (Coeurdacier and Rey 2011) but, to the best of my knowledge, there is little research on home bias in residential mortgage lending.

Moreover, only recently has evidence emerged that this home bias increases when economic conditions worsen (i.e. that there is a flight to home effect). For example, in response to an adverse shock to financing conditions, international banks in the syndicated lending market shifted their lending activity towards their home country, regardless of the perceived risk of the borrowers (Giannetti and Laeven 2012; De Haas and Van Horen 2012). Broadly speaking, there are two possible explanations for a home bias in credit markets – information asymmetries and behavioural biases. If lenders cannot observe borrower risk perfectly and it is costly to collect information on the creditworthiness of borrowers then lenders may be better informed about local borrowers than non-local borrowers. Under this explanation, geographic distance is a proxy for credit risk and, similar to the flight to quality, lenders will re-balance their portfolios towards local borrowers when economic and financial conditions deteriorate. Alternatively, certain lenders may specialise in lending to distant borrowers and have more sophisticated loan screening and monitoring technologies than local lenders. In this case, local lenders would not have an informational advantage, so any home bias may be better explained by a behavioural bias towards familiar assets rather than by information asymmetries.

Footnotes

There is a close analog to the literature on the credit channel of monetary policy. The credit channel of monetary policy can be divided into two channels – the ‘bank lending channel’ and the ‘borrower balance sheet channel’. The bank lending channel measures the effect of monetary policy shocks on the real economy through their effect on the balance sheets of lenders. In contrast, the borrower balance sheet channel measures the effect of monetary policy shocks on the real economy through their effect on borrower balance sheets. The effect of liquidity shocks on lending is sometimes loosely referred to as a ‘bank lending channel’ in the literature, despite the fact that neither monetary policy nor the real economy are considered. [3]

Lower loan losses will also indirectly boost the level of lending, as the existing stock of loans will not be dragged down by loans that are written off. [4]

According to US flow of funds data, the outstanding value of RMBS (US$6.4 trillion) was twice as large as the outstanding value of non-deposit liabilities owed by private depository institutions (US$3.2 trillion) at the end of 2006. Moreover, the flow of funds indicates that the stock of non-deposit funding continued to grow during the crisis period, while data provided by the Securities Industry and Financial Markets Association (SIFMA) suggest that RMBS issuance fell by about US$1 trillion between 2006 and 2008. This suggests that the shock to the RMBS market was more likely to have had a significant impact on US bank liquidity, and hence mortgage lending, than any shock to non-deposit funding. [5]