Content area
Full Text
L ike all investors, investors in the fixed-income market are demanding to be compensated for the risks they are bearing. When investing in bonds issued by corporations, the main risk is that the corporation may not be able to fulfill its obligations at some point in time during the maturity of the bond (a default situation). Compensation for this type of risk is shown in the extra yield the investor receives compared with owning an otherwise equal default-free, or risk-free, asset. This extra yield is often referred to as a bond's credit spread .
In determining the theoretical size of the credit spread, researchers have traditionally leaned on several structural models. The most commonly used is the Merton [1974] model for estimating a model implied credit spread. However, most previous papers on this topic show that credit spreads implied by such a structural model are too low compared with the actual credit spreads observed in the market. 1 This has given rise to the so-calledCredit Spread Puzzle .
A recent paper by Feldhütter and Schaefer [2013] (hereafter called FS) points to two biases from which many of the previous papers suffer. These biases make earlier conclusions about the existence of the Credit Spread Puzzle somewhat misleading.
The first bias that FS point out comes from the fact that many previous papers useaverage firm variables (both cross-sectional and time series) to calculate anaverage model implied spread, which they then compare with anaverage actual spread. This is usually done per rating category. FS point out that this may lead to a convexity bias, as the model implied spreads that are based on average firm-specific variables typically are lower than the average spread for a group.
The second bias that FS note arises from the fact that some papers use an implied measure for the asset volatility when calculating the default probability. They imply out the asset volatility such that the model implied default probability equals the actual historical default probability. The assumption here is that the historical default probabilities are a good proxy for the expected (future) default probabilities. FS use a simulation technique to show that realized default probabilities are not a good proxy for expected default probabilities.
FS control for...