In this paper we compare counterfactual corporate bond issuing dates to actual issuing dates in order to test the ability of firms to time the credit market. The 50 most active bond issuing financial firms and the 50 most active industrial firms are studied using one week, one month, and one quarter windows. The ability to time firm-specific CDS prices is studied from January 2002 - October 2009. The ability to time the risk-free rate (10 year US government bond) is studied from January 1988 - October 2009. We find that: firms do not successfully time the risk-free rate or the credit spreads. There is no evidence of CDS timing ability over one week or one month, but there is some borderline evidence at one quarter. For a typical bond issue, the firm loses about 1% of the face value of the bond relative to a 1 month window, due to their inability to time the market. If the firms could improve their market timing, they could save many hundreds of millions of dollars. Since there is a degree of statistical predictability in the data, we find it surprising that these firms are not able to do a better job of timing the credit market.

This paper investigates a two-factor affine model for the credit spreads on corporate bonds. The Þrst factor can be interpreted as the level of the spread, and the second factor is the volatility of the spread. The riskless interest rate is modeled using a standard two-factor affine model, thus leading to a four-factor model for corporate yields. This approach allows us to model the volatility of corporate credit spreads as stochastic, and also allows us to capture higher moments of credit spreads. We use an extended Kalman Þlter approach to estimate our model on corporate bond prices for 108 Þrms. The model is found to be successful at Þtting actual corporate bond credit spreads, resulting in a signiÞcantly lower root mean square error (RMSE) than a standard alternative model in both in-sample and out-of-sample analyses. In addition,key properties of actual credit spreads are better captured by the model.