5.4. Yield Spreads and Real Interest Rates
Fridson, Garman and Wu (1997) suggest that an empirical investigation of the default
risk - riskless rate relation should be based on the real interest rate, rather than nominal rates
that are the standard in yield spread studies. They argue that if the company’s revenues are
fully adjusted with respect to inflation, its effective cost of capital is related to the real
riskless rate rather than the nominal rate. As the real riskless rate increases, the firm’s cost of
short-term debt increases as well, and its available cash flow decreases. This will hurt the
firm's ability to issue new long-term debt, and expected default rates will rise.
In agreement with their hypothesis, Fridson, Garman and Wu (1997) find a significantly
positive relationship between default rates and real riskless rates. In the context of the
current paper, this result implies a positive yield spread - real riskless rate relation when
callability is controlled for. In this section, we first examine the relationship between yield
spreads and the inflation rate and real riskless rate. Then we test the most comprehensive of
the above models for yield spreads, namely the Collin-Dufresne et al. (2001) model, using
real, rather than nominal, Government of Canada yield. Both tests show that the yield spread
of callable bonds is negatively related to the nominal riskless rate. However, when the real
riskless rate is used instead, this significant negative correlation disappears. Since the
decision to call is based on the level of nominal rates rather than real rates, the yield spread
attributed to callability should be negatively related to nominal riskless rate and unrelated to
the real rate. This result presents further evidence for our conclusion that call risk dominates
the observed negative yield spread - nominal riskless rate relation.
5.5. Inflation, Real Interest Rate and Yield Spreads
Following Fridson, Garman and Wu’s (1997) prediction Equation (5) was also estimated
with the results reported in Table 6, Again, in some cases, we find that our SCM data set is
characterized by the autoregressive nature of the OLS residuals of regression model (5). As
23