Skip to content
All library documents

Solving the Hull–White Bond Pricing Equation with Boundary Conditions

Article Quant Q&A · Author: rupert

Summary

The answer explains how to derive the time-dependent bond price function in the Hull–White extended Vasicek model. Instead of trying to obtain the target result by manipulating a later equation alone, it solves the first-order equation for the bond-price function with respect to time, then uses the boundary condition that the bond price is zero at maturity.

It introduces primitive functions for the time-varying mean-reversion coefficient and integrates the resulting equation, leaving a maturity-dependent term that the boundary condition determines. An alternate derivation rewrites the later relation in terms of the derivative of the maturity derivative, integrates with respect to maturity, and again applies the boundary condition. The answer checks that the resulting form satisfies the target relation. This is an algebraic explanation of the cited model equations; it does not discuss calibration, market data, or the model’s pricing accuracy.

Key ideas

  • The bond-price equation can be solved as a first-order linear differential equation in time.
  • The zero price at maturity determines the otherwise arbitrary maturity-dependent term.
  • The derivation uses primitive functions to express the solution for time-varying coefficients.
  • A second route integrates the maturity derivative after using the original differential equation.

Tags

Full text
# Hull White Extended Vasicek model


# Hull White Extended Vasicek model












I am trying to understand a formula in the landmark paper by Hull&White "Pricing Interest Rate Derivative Securities" (1990). I cannot see how rearranging (11) and applying boundary conditions results in (13) on page 6 (as below).

Please could you give me a pointer and apologies if its my rudimentary calculus that is at fault.

## Answer by ir7 (score 1, accepted)

https://quant.stackexchange.com/a/57688

You are right, equation (11) is derived mechanically from (7) (by taking the derivative wrt to $T$ and then combining is with (7)), and somehow they think that (13) can be obtained from (11) without remembering (7). Maybe by smartly integrating (note for example that $B_tB_T - BB_{tT}$ is the numerator of the derivative wrt to $t$ of fraction $B/B_T$) and using the boundary condition (I couldn't figure it out).

Of course, what we can do is solve the first-order linear equation in $t$ (7)

$$ B_t = a(t)B-1. $$

With the usual primitive functions:

$$ \alpha'(t) = a(t), \; \; \beta'(t) = -{\rm e}^{-\alpha(t)}, $$

the general solution to equation (7) is

$$ B(t,T) = c(T){\rm e}^{\alpha(t)} + {\rm e}^{\alpha(t)}\beta(t), $$

with $c(T)$ arbitrary function of $T$.

As $B(T,T)=0$, we must have:

$$c (T)= -\beta(T).$$

So:

$$ B(t,T) = -{\rm e}^{\alpha(t)} \left(\beta(T) - \beta(t)\right).$$

We can then easily check that this solution respects (13):

$$ B(0,T) = -{\rm e}^{\alpha(0)} \left(\beta(T) - \beta(0)\right) $$

$$ B(0,t) = -{\rm e}^{\alpha(0)} \left(\beta(t) - \beta(0)\right) $$

$$\partial B(0,t)/\partial t = -{\rm e}^{\alpha(0)}\beta'(t) = {\rm e}^{\alpha(0)} {\rm e}^{-\alpha(t)}$$

Edit: Note that (11) can be written as:

$$ (B_T)_t =\frac{1-B_t}{B}B_T $$ which is equivalent to $$ (\ln B_T)_t = \frac{1-B_t}{B}. $$

At this point we need to remember from (7) that the right hand side is a function of $t$ only, $a(t)$, otherwise it's getting cumbersome to progress from here. The solution is $$ B_T = {\rm e}^{\alpha (t) + \gamma (T)} $$ for $ \gamma (T)$ an arbitrary function of $T$. Integrating wrt to $T$, we get:

$$ B(t,T) = {\rm e}^{\alpha (t)} (\Gamma (T) + \eta (t)) $$ for $ \eta (t)$ an arbitrary function of $t$ and $\Gamma^\prime = {\rm e}^{\gamma}$.

Boundary condition $B(T,T)=0$ then forces:

$$\Gamma(T) = -\eta(T). $$

So,

$$B(t,T) = -{\rm e}^{\alpha(t)} \left(\eta(T) - \eta(t)\right).$$

One more time, noting that

$$ B_t = -{\rm e}^{\alpha(t)}a(t)\eta(T) + {\rm e}^{\alpha(t)}a(t) \eta(t) + {\rm e}^{\alpha(t)} \eta^\prime (t),$$

(7) then implies:

$$\eta(t) = \beta(t). $$

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.