measure based upon the difference. They do so by introducing a simple theoretical
framework that explains the effect of callability on the interest-rate factor and that by
construction, relative spreads should bring about a stronger yield spread - interest rate factor
relation,
Collin-Dufresne et al. (2001) identify more factors that may determine changes in credit
spreads. For example, Duffee (1998) omits important factors such as the Longstaff and
Schwartz (1995) asset factor in his regression analysis. Collin-Dufresne et al. (2001)
combine the explanatory variables offered by Longstaff and Schwartz and Duffee, in
addition to a convexity term and firm leverage variable. They conclude that credit spread
changes are primarily driven by local supply/demand shocks, which are independent of both
credit-risk factors and liquidity factors. Their regression model is so far the most
comprehensive in the extant literature and forms the basis for a number of recent models.
These accommodate time varying volatility and autocorrelation in spread changes through
GARCH and ARMA specifications and more recently through the incorporation of liquidity
based variables (Ericsson and Renault, forthcoming; Chen et al., forthcoming). Note that
autocorrelation in spread returns may be due to illiquidity of that particular bond in
secondary markets.
In this paper, we apply each of the above seminal regression models using Canadian,
investment-grade, corporate bond indices. The strength of our study lies in using Canadian
corporate bonds, which are devoid of the tax effect. Studies using U.S. bond data cannot
avoid these tax effects. Canada’s Doomsday call provision also allows identifying callable
and noncallable indices. The effects of taxes and the uniqueness of the Canadian Doomsday
call provision are discussed in turn.
3. The Uniqueness of Canadian Bond Data
3.1 Tax Effects
Canadian bond data enables controlling for tax effects arising from the different tax rates,