This dissertation empirically examines refunding decisions employed by issuers of tax-exempt bonds. Callable bonds contain embedded call options by virtue of provisions in bond indentures that permit the issuing firm to buy back the bond at a predetermined strike price. Such an embedded American call option has two components to its value, the intrinsic value and the time value. The issuer can realize at least as much as the intrinsic value by exercising immediately, when the option is in-the-money. Usually it is optimal for the holder of an in-the money American option to wait rather than exercise immediately, because the option has time value. It is rational for the holder to exercise the option when the total value of the option is no more than the intrinsic value. Option pricing theory can be used to identify two sub-optimal refunding strategies: those that refund too early, and those that refund too late. In such cases the holder incurs losses.

I analyze the refunding decisions for two different samples of tax-exempt bonds issued between 1986 and 1993: the first consists of 2,620 bonds that are called, and the second contains 23,976 bonds that are never called. The generalized Vasicek (1977) model in the Heath, Jarrow, and Morton (1992) framework is used to construct binomial trees for interest rates, bond prices, and call option prices. The option pricing lattice is then used to compute the loss in value from sub-optimal refunding strategies, refunding efficiency, and months from optimal time for bonds in these two samples.

Results suggest that sub-optimal refunding decisions cause losses to the issuers, which are present across bond and issuer characteristics. For the pooled sample of 26,596 bonds, the loss in value from sub-optimal refunding decisions totaled $7.2 billion, amounting to a loss of about 1.75% of total principal amount. Results indicate that issuers either wait too long to refund or never refund and cannot realize the present value saving of switching a high coupon bond with a low coupon bond, over a longer period of time. These results critically depend on the assumptions of underlying term structure model and are sensitive to model calibrated parameter values.