Survival analysis of callable bonds

We present an empirical analysis of callable bonds. We use a hazard rate model for an issuer's propensity to call a bond to examine covariates relating to redemption behavior. We take the unique approach of separating issuers who redeem despite the call option never being in-the-money from those who call after the option has gone in-the-money. This methodology illuminates different factors driving conditional and unconditional call timing.; The contribution to the literature is threefold. From a corporate finance perspective, we examine how a manager's knowledge of the firm's current and future financial position impacts his bond redemption decisions. From a valuation perspective, we use a semi-parametric hazard rate approach incorporating a broader range of systematic and non-systematic factors with which to analyze the timing of exercise over a longer date range than in previous work. Methodologically, accounting for the evolution of covariate values through time is well-suited to analyzing a time-dependent instrument like a call option.; We make two unique findings regarding bonds that are called after going in-the-money. First, recent rating transitions are a statistically significant factor relating to the timing of the call. Second, the intensity with which the bonds are in-the-money, as measured by the number of months the market price exceeds the call price as a percentage of the total number of months from the first time in-the-money, also influences the timing of the call. In-the-money call timing is also related to opportunity cost, future growth prospects, and interest rate expectations. For bonds called when out-of-the-money, industry classification is a significant factor in call behavior, as is bond age.; Extending the analysis, we fit parametric distributions to the call hazard function. We conjecture that the lognormal distribution may approximate expected calling behavior for the in-the-money cohort; the Weibull distribution may estimate expected behavior for out-of-the-money callers in the industrial and finance sectors; and the log-logistic distribution may estimate the hazard function for out-of-the-money callers in the utility sector. Comparing the findings from the parametric and semi-parametric analyses shows that the results are robust to assumption of functional form.