4

I'm reading Steven Shreve's Stochastic Calculus for Finance, in pp 274, there's such a PDE to be solved:

$$f_t(t,r)+(a(t)-b(t)r)f_r(t,r)+\frac12 \sigma^2(t)f_{rr}(t,r)=rf(t,r)$$

with terminal condition $$f(T,r)=1$$ for all $r$.

Then, it jumped to:

We initially guess and subsequently verify that the soltion has the form

$$f(t,r)=e^{-rC(t,T)-A(t,T)}$$

Of course the guess is correct and the solution is found:

$$C(t,T)=\int_t^T e^{-\int_t^s b(v) dv} ds$$ $$A(t,T)=\int_t^T \left(a(s)C(s,T)-\frac12 \sigma^2(s)C^2(s,T)\right)ds$$

But I'm a bit lost here -- how did it make the first guess that the solution is of the form $$f(t,r)=e^{-rC(t,T)-A(t,T)}$$ ? Is this something usual in PDE solutions? or is there some well-known facts in PDE that suggest such a form?

athos
  • 5,177
  • Yes it is usual – Snoop Dec 22 '21 at 03:20
  • 1
    @Snoop well it might be guessed and such guess might sounds natural to the expert in the area , so could you pls elaborate on what background would induce such educated guess? – athos Dec 22 '21 at 10:51

1 Answers1

2

There are clues in the form of the equation, though a finance intuition might be a better guide. Taking $a$, $b$, and $\sigma$ all zero suggests solutions like $e^{rt}$. Putting those terms back then you see that the plain $t$ won't do, so perhaps $e^{r c(t)}$ will be a possible place to begin looking. The other clue is the common practice of looking for product solutions like $g(r)h(t)$, or maybe in this case $e^{r c(t)} h(t)$, which seems to have worked.

It is important to realize that finding a formula that works does not prove that you have found THE solution. You have to make a uniqueness argument to assert that.

Bob Terrell
  • 3,812