Dear Mark Joshi. I have a few questions concerning your recent article "Fourier Transforms, Option Pricing and Controls". Can I ask you them in this thread?

http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1941464

21 posts • Page **1** of **2** • **1**, 2

Dear Mark Joshi. I have a few questions concerning your recent article "Fourier Transforms, Option Pricing and Controls". Can I ask you them in this thread?

http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1941464

http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1941464

- stronzo
**Posts:**11**Joined:**Tue Feb 07, 2012 11:28 am

this is the place to ask.

- mj
- Site Admin
**Posts:**1380**Joined:**Fri Jul 27, 2007 7:21 am

Thank you very much. First, concerning (3.4), are you assuming that F_T(0) = 0 ?

- stronzo
**Posts:**11**Joined:**Tue Feb 07, 2012 11:28 am

E(S_T)= F_T (0)

at some points in the paper, we assume this is 1; see after (4.1)

at some points in the paper, we assume this is 1; see after (4.1)

- mj
- Site Admin
**Posts:**1380**Joined:**Fri Jul 27, 2007 7:21 am

Sorry, I wanted to write "=1" instead of "=0".

Yes, but it seems to me that you are assuming here also ((3.4) comes before (4.1)) that F_T(0)=1. Aren't you?

Yes, but it seems to me that you are assuming here also ((3.4) comes before (4.1)) that F_T(0)=1. Aren't you?

- stronzo
**Posts:**11**Joined:**Tue Feb 07, 2012 11:28 am

why do you think that?

- mj
- Site Admin
**Posts:**1380**Joined:**Fri Jul 27, 2007 7:21 am

In fact, in all your section 3.3 page 6, you are assuming that F_T(0)=1. Otherwise, there is necessary a term missing in F_T(0) coming from (3.1).

But it's ok, you can do it by (4.1).

In all your section 3.3, you are looking for \sigma such that \hat{\phi}(z)=0 for z\in\Gamma_{\eta} such that Re(z)=0. However, why is it important to have \hat{\phi}(z)=0 at this point ? This property does not imply that e^{-izl}\hat{\phi} is less oscillating on \Gamma_{\eta}. I agree, that value is the more natural value. But, have you tested numerically other values of \sigma?

But it's ok, you can do it by (4.1).

In all your section 3.3, you are looking for \sigma such that \hat{\phi}(z)=0 for z\in\Gamma_{\eta} such that Re(z)=0. However, why is it important to have \hat{\phi}(z)=0 at this point ? This property does not imply that e^{-izl}\hat{\phi} is less oscillating on \Gamma_{\eta}. I agree, that value is the more natural value. But, have you tested numerically other values of \sigma?

- stronzo
**Posts:**11**Joined:**Tue Feb 07, 2012 11:28 am

well the objective is to make the thing as small as possible and making it vanish there is one way to do so. Making it zero there makes the real part vanish to second order.

All the tests we did indicated that the total modulus of the integrand was minimized this way.

All the tests we did indicated that the total modulus of the integrand was minimized this way.

- mj
- Site Admin
**Posts:**1380**Joined:**Fri Jul 27, 2007 7:21 am

Page 8, I read : "In some cases, if we restrict the logarithm naively to its principal branch then the characteristic function will become discontinuous, leading to erroneous option prices" and then: "even though we compute the complex logarithm and power using their principal branch, there will not be discontinuity problems in evaluating the characteristic functions." What should I understand ? Do you use or not the principal branch to evaluate complex logarithm?

In Formula (4.5) for d(\xi), do you evaluate \sqrt(a+i b) as \exp((1/2)\ln(a+i b)) ? What is for you the principal branch of the complex logarithm ? For z=a+ib being a complex number which is not a negative real, is it the unique complex number of the forum x+iy with x:=\ln(|z|), y\in ]-\pi,\pi[ and \exp(iy)=z/|z|?

Given (4.4) are we sure that when integrating (3.2) along \Gamma_{\eta} we will never have to consider complex logarithm of negative reals?

In Formula (4.5) for d(\xi), do you evaluate \sqrt(a+i b) as \exp((1/2)\ln(a+i b)) ? What is for you the principal branch of the complex logarithm ? For z=a+ib being a complex number which is not a negative real, is it the unique complex number of the forum x+iy with x:=\ln(|z|), y\in ]-\pi,\pi[ and \exp(iy)=z/|z|?

Given (4.4) are we sure that when integrating (3.2) along \Gamma_{\eta} we will never have to consider complex logarithm of negative reals?

- stronzo
**Posts:**11**Joined:**Tue Feb 07, 2012 11:28 am

you need to read the Lord--Kahl paper or Gatheral's book where these issues are discussed in depth.

- mj
- Site Admin
**Posts:**1380**Joined:**Fri Jul 27, 2007 7:21 am

All right. I'm going to read "Complex logarithms in Heston-like models" (the reference you gave in your bibliography). So does it mean that the answer is "Yes" to all of my questions above i.e., you use the principal branch of the complex logarithm (the one with the argument in ]-\pi,\pi[) and there is no problem?

- stronzo
**Posts:**11**Joined:**Tue Feb 07, 2012 11:28 am

basically yes as long as you define the thing you are taking the log of in the right way.

- mj
- Site Admin
**Posts:**1380**Joined:**Fri Jul 27, 2007 7:21 am

OK. Is your code available freely somewhere?

- stronzo
**Posts:**11**Joined:**Tue Feb 07, 2012 11:28 am

no...

There is some Fourier transform Heston code in QuantLib however.

There is some Fourier transform Heston code in QuantLib however.

- mj
- Site Admin
**Posts:**1380**Joined:**Fri Jul 27, 2007 7:21 am

But is it easy to use complex numbers in C++? Is there a function which returns the principal branch of the logarithm or at least the argument of a given complex number?

- stronzo
**Posts:**11**Joined:**Tue Feb 07, 2012 11:28 am

21 posts • Page **1** of **2** • **1**, 2

Return to General technical discussion

Users browsing this forum: No registered users and 1 guest