Anchoring on Credit Spreads

This paper documents that the path of credit spreads since a firm’s last loan influences the level at which it can currently borrow. If spreads have moved in the firm’s favor (i.e., declined), it is charged a higher interest rate than justified by current fundamentals, and if spreads have moved to its detriment, it is charged a lower rate. We evaluate several possible explanations for this finding, and conclude that anchoring (Tversky and Kahneman [1974]) to past deal terms is most plausible. JEL classification: G30, G32 ∗Dougal is at the LeBow College of Business at Drexel University, Engelberg and Parsons are the Rady School of Management at the University of California, San Diego, and Van Wesep is at the Vanderbilt University Owen Graduate School of Management. We thank Aydoğan Altı, Malcolm Baker, Adolfo DeMotta, Jan Ericsson, Dirk Hackbarth, Jay Hartzell, David Hirshleifer, Victoria Ivashina (discussant), Anil Shivdasani, Sheridan Titman, and Siew Hong Teoh. We also thank seminar participants at UNC, McGill, DePaul, University of Michigan, UC Irvine, SFS Cavalcade, University of Washington, Florida State University, 2012 AFA Meeting, University of Oregon, Brigham Young University, Tilburg University, Erasmus University, Duisenberg School of Finance, Indiana University, and the 2011 NBER Corporate and Behavioral Programs in Chicago for helpful feedback. Virtually all borrowers pay a spread above the risk free rate, a premium that compensates lenders for expected losses in default, illiquidity, and other considerations. Calculating the appropriate spread requires answering a number of questions. How likely is the firm to default, and over what horizon? If it does default, how big are the losses? Is default more likely to occur in bad economic times, or is it largely dependent on firm-specific factors? How easy will it be to sell the firm’s bonds or bank notes, and will this be more difficult during certain economic states? The common element in these questions is their perspective: they are all forecasts. Retrospective information is only relevant to the extent that it improves the lender’s estimate of forward looking variables. Data that are purely historical – in the sense that they do not provide information about the firm’s creditworthiness – should not affect spreads. This paper provides evidence that non-informative historical signals do, in fact, influence borrowing costs, suggesting that observed credit spreads likely depart from a fully rational benchmark spread. What do we mean by non-informative, historical signals? Suppose that two neighbors living across the street from one another both want to refinance their home loans, and that prevailing rates on 30-year mortgages are currently 6% on average. Neighbor 1 originated his mortgage five years ago, when average rates were 8%, and neighbor 2 originated her mortgage ten years ago, when average rates were 4%. It would be surprising if this differential timing in their prior loan originations translated to a difference in borrowing costs on new loans today: we would not expect for neighbor 1 to pay a higher rate today than neighbor 2, given that the only (observable) distinction is when each happened to have last borrowed. And yet, this is precisely what we find in the syndicated loan market. On average, when a firm borrows money from a bank, the level of aggregate credit spreads when the firm last borrowed correlates with the spread it receives on a new loan. The pattern is consistently one of stickiness, whereby firms that last borrowed when spreads were high pay a premium, and firms that last borrowed when spreads were low receive a discount. To give a specific example, suppose that a BBB-rated firm took out a 5-year term loan in the year 2000, and then another in 2003. Suppose also that the average BBB credit spread rose by 50 basis points over these three years. We find that, relative to other BBB borrowers in 2003, the firm will receive a discount of approximately 6 basis points, negating 12% of the change in aggregate spreads. In order to highlight how bizarre this finding is, in the preceding example we excluded all firm-specific information. However, prior aggregate spreads must ultimately translate to current borrowing costs through some firm-specific variable, and the most likely candidate is the firm’s actual spread at the time it last borrowed. When we extend the analysis to