Dynamic Debt Runs and Financial Fragility - Evidence from the 2007 ABCP Crisis
Debt runs played a central role in the financial crisis of 2007-2008, reigniting the debate about their causes and how they can be prevented. This research uses the 2007 asset-backed commercial paper (ABCP) crisis as a basis to study the determinants of debt runs.
Debt runs played a central role in the financial crisis of 2007-2008. Investors ran on asset-backed commercial paper (ABCP) starting in July, 2007; on repo in September, 2007; and on money market mutual funds in September, 2008. Investors also ran on large banks such as Northern Rock (September, 2007) and Bear Stearns (March, 2008).
These runs reignited the debate about their causes and how they can be prevented. This research contributes to the debate by measuring the sensitivity of runs to several contributing factors, including maturity mismatch, leverage, asset volatility and liquidity, and the strength of credit guarantees. The results help answer four questions that are vital to policy makers, regulators, bankers, and investors:
- How fragile are financial intermediaries?
- How can we design financial intermediaries ex anteto control the risk of future runs?
- What are the warning signs that a run is imminent?
- Which interventions best prevent runs ex post once conditions have started deteriorating?
We estimate a dynamic model of debt runs using data from the 2007 crisis in asset-backed commercial paper. The model allows yields to change over time, which introduces dilution risk. The conduit must offer higher yields to induce rollover if conditions worsen, which dilutes the claims of other lenders. Introducing dilution risk into the model can make runs up to 11 times more likely. Our model of fundamental-driven runs fits several features of the data, including the dramatic increase in yields on ABCP leading up to runs, the high probability of recovery once a run starts, the positive relation between yields and the probability of future runs, the overall level of volatility in ABCP yields, and the positive relation between yield volatility and the yield level. The model fits much better in the subsample of conduits with the weakest credit guarantees. We find that runs are very sensitive to conduit leverage and expected asset liquidation costs. Runs are much less sensitive to the degree of maturity mismatch, the strength of credit guarantees, and the asset's volatility and growth rate. These sensitivities are useful inputs to regulators and banks attempting to control the risk of runs. Our analysis can be extended and improved in three main directions. To keep the estimation tractable, we assume that yields cannot exceed an exogenous cap. Second, we have taken debt maturity as given. Brunnermeier and Ohmke (2012) show that rollover risk can produce a rat race in which debt maturity unravels to its shortest possible value. Empirically, we see that the maturity rat race occurs before yields adjust. The average maturity drops but then levels off around 25 weeks before runs occur, after which the yields start adjusting upwards. Why maturities and yields adjust sequentially instead of simultaneously is an important question that we also leave for future research. Finally, the dynamic debt runs framework, and the estimation method we propose here, can be used to study the runs on money market funds in late 2008, potentially shedding more light on their causes.