Serving the Quantitative Finance Community

 
User avatar
pcaspers
Topic Author
Posts: 712
Joined: June 6th, 2005, 9:49 am

Carr / Madan: A note on sufficient conditions for no arbitrage

December 14th, 2020, 8:40 pm

I have a question on the paper "A note on sufficient conditions for no arbitrage" by Carr / Madan:

They say \sum_{i=1}^\infty q_{i,j} = 1. To me it seems this sums equals Q_{1,j}  = S_0 - C_{1,j} / K_1 != 1. Do we have add an additional strike, possibly K = K_0 = 0 and attach the probability 1 - Q_{1,j} to it to complete the probability distribution?

Less important, do we have to put an additional condition on the sequence of strikes to ensure (C_{i,j} - C_{i+1,j}) / (K_{i+1}-K_i) tends to zero when i goes to infinity, like "there is an epsilon > 0 s.t. K_{i+1}-K_i > epsilon for all i". 
 
User avatar
Cuchulainn
Posts: 64681
Joined: July 16th, 2004, 7:38 am
Location: Drosophila melanogaster
Contact:

Re: Carr / Madan: A note on sufficient conditions for no arbitrage

December 14th, 2020, 9:34 pm

Howdy Peter,
Here is article by Carr and Itkin, might be something to be gleaned?

https://engineering.nyu.edu/sites/defau ... ammaMo.pdf

And closer, MSc of Lykke Rasmussan, section 6 (BTW having difficulty reading your LATEX  :o). She has 3 conditions for no-arbitrage.
I think Joerg knows thesis.
Attachments
Mar2012_MasterThesis_LykkeRasmussen.pdf
(6.03 MiB) Downloaded 54 times
Last edited by Cuchulainn on December 14th, 2020, 9:52 pm, edited 1 time in total.
"Compatibility means deliberately repeating other people's mistakes."
David Wheeler

http://www.datasimfinancial.com
http://www.datasim.nl
 
User avatar
Cuchulainn
Posts: 64681
Joined: July 16th, 2004, 7:38 am
Location: Drosophila melanogaster
Contact:

Re: Carr / Madan: A note on sufficient conditions for no arbitrage

December 14th, 2020, 9:49 pm

And I quote

[$]\frac{\partial C}{\partial K}  \le 0[$]

[$]\frac{\partial^2 C}{\partial K^2} \ge 0[$]

[$]\frac{\partial C}{\partial \tau}  \ge 0[$]

Does this make sense?
"Compatibility means deliberately repeating other people's mistakes."
David Wheeler

http://www.datasimfinancial.com
http://www.datasim.nl
 
User avatar
Alan
Posts: 10652
Joined: December 19th, 2001, 4:01 am
Location: California
Contact:

Re: Carr / Madan: A note on sufficient conditions for no arbitrage

December 15th, 2020, 2:28 pm

I have a question on the paper "A note on sufficient conditions for no arbitrage" by Carr / Madan:

They say \sum_{i=1}^\infty q_{i,j} = 1. To me it seems this sums equals Q_{1,j}  = S_0 - C_{1,j} / K_1 != 1. Do we have add an additional strike, possibly K = K_0 = 0 and attach the probability 1 - Q_{1,j} to it to complete the probability distribution?

Less important, do we have to put an additional condition on the sequence of strikes to ensure (C_{i,j} - C_{i+1,j}) / (K_{i+1}-K_i) tends to zero when i goes to infinity, like "there is an epsilon > 0 s.t. K_{i+1}-K_i > epsilon for all i". 

I agree with your sum and agree that an additional (unstated) condition seems to be needed. Since they are doing a lattice version of the continuum problem, let's first look at that. Fixing the maturity T, the call value is [$]C(K)[$].  Breeden and Litzenberger say that [$]q(K) \equiv C_{KK}(K)[$] can be interpreted, under some conditions, as the pdf of finding the stock price at strike K on maturity. What are the conditions? If the stock price cannot actually reach 0, then for mass preservation we need

[$] 1 = \int_{0^+}^{\infty} C_{KK} \, dK =  C_K |^{K=\infty}_{K=0^+} [$]    

The upper end is not problematic: since C(K) is decreasing to 0, [$]C_K[$] must be decreasing to 0 as well.

But, at the lower end, we need the condition that 

(*)  [$]C_K(0^+) = -1[$].

In the Carr-Madan lattice version this would be the condition that

[$] 1 = \frac{C_0 - C_1}{K_1 - K_0} = \frac{S_0 - C_1}{K_1}[$].

So you need to assume that [$]C_1 = S_0 - K_1[$] as the lattice version of (*). 
 
User avatar
Cuchulainn
Posts: 64681
Joined: July 16th, 2004, 7:38 am
Location: Drosophila melanogaster
Contact:

Re: Carr / Madan: A note on sufficient conditions for no arbitrage

December 15th, 2020, 9:16 pm

Alan, If you differentiate Dupire PDE twice with respect to K you get a PDE with zero BC and Dirac payoff, so it's a pdf, yes?. We discussed for BS PDE in same vein. (but transform to (0,1)).a pdf is always >= 0 and that can be proved by PDE theory.

Is Dupire PDE just a common-or-garden PDE at the end of the day?
"Compatibility means deliberately repeating other people's mistakes."
David Wheeler

http://www.datasimfinancial.com
http://www.datasim.nl
 
User avatar
Alan
Posts: 10652
Joined: December 19th, 2001, 4:01 am
Location: California
Contact:

Re: Carr / Madan: A note on sufficient conditions for no arbitrage

December 15th, 2020, 10:44 pm

Dupire/Gyongy/Local vol model is somewhat specialized theory, in the sense that it requires a diffusion process (or at least an Ito process) for the asset price. But, yes, 2 [$]K[$]-derivatives should be a PDE for [$]q(K,T)[$] if you know the local vol function. And (usually) 0's at the spatial boundaries, as you say.

Breeden-Litzenberger is more general with minimal assumptions. Basically, just need existence of a (norm-preserving) [$]q(K,T)[$] at a given option expiration [$]T[$]. Doesn't require any type of dynamics: diffusions or whatever. The particle can just magically appear at expiration.
 
User avatar
pcaspers
Topic Author
Posts: 712
Joined: June 6th, 2005, 9:49 am

Re: Carr / Madan: A note on sufficient conditions for no arbitrage

December 16th, 2020, 11:28 am

In the Carr-Madan lattice version this would be the condition that

[$] 1 = \frac{C_0 - C_1}{K_1 - K_0} = \frac{S_0 - C_1}{K_1}[$].

So you need to assume that [$]C_1 = S_0 - K_1[$] as the lattice version of (*). 
Thanks. I am a bit reluctant to make this assumption, since [$]K_1 > 0[$], so this would only hold if the stock has zero volatility?
 
User avatar
pcaspers
Topic Author
Posts: 712
Joined: June 6th, 2005, 9:49 am

Re: Carr / Madan: A note on sufficient conditions for no arbitrage

December 16th, 2020, 11:36 am

And I quote

[$]\frac{\partial C}{\partial K}  \le 0[$]

[$]\frac{\partial^2 C}{\partial K^2} \ge 0[$]

[$]\frac{\partial C}{\partial \tau}  \ge 0[$]

Does this make sense?
It does. I am setting up an arbitrage-checker on a discrete grid though and want some theoretical foundation for that.
 
User avatar
Alan
Posts: 10652
Joined: December 19th, 2001, 4:01 am
Location: California
Contact:

Re: Carr / Madan: A note on sufficient conditions for no arbitrage

December 16th, 2020, 5:19 pm

In the Carr-Madan lattice version this would be the condition that

[$] 1 = \frac{C_0 - C_1}{K_1 - K_0} = \frac{S_0 - C_1}{K_1}[$].

So you need to assume that [$]C_1 = S_0 - K_1[$] as the lattice version of (*). 
Thanks. I am a bit reluctant to make this assumption, since [$]K_1 > 0[$], so this would only hold if the stock has zero volatility?
The continuum limit version would be something like 
(**) [$]C(K) = S_0 - K + o(K), \quad \mbox{as} \quad K \rightarrow 0[$].

Likely the precise nature of the sub-leading terms depends on whether or not  the stock price cannot reach the origin. It's probably been done carefully in somebody's paper; you might look for discussions of the Breeden-Litzenberger formula generalized to models that allow bankruptcy.  Another thing to check is the behavior implied by Roger Lee's no-arb condition for the implied vol at extreme strikes.

If you don't like my relation on the lattice, an alternative might be: 
require only that [$]C_1 \ge S_0 - K_1[$] and [$]C_1 > C_2[$] and interpret any "missing mass" as the probability of finding the stock at the origin (or at least below [$]K_1[$]). This was more or less your original idea of how to handle it.

As a practical matter, if your arbitrage checker uses for [$]K_1[$] some smallest quotable strike, then there's certainly a non-zero probability of finding the stock price below that value.  
 
User avatar
pcaspers
Topic Author
Posts: 712
Joined: June 6th, 2005, 9:49 am

Re: Carr / Madan: A note on sufficient conditions for no arbitrage

December 17th, 2020, 8:11 am

That makes sense. Actually, in the paper, they start without making any model assumptions, they just impose some no-arbitrage conditions on a discrete set of observed option prices. From that, they construct a discrete probability distribution for [$]S_{T_j}[$] at each maturity [$]T_j[$] compatible with the observed prices. I think it's okay to have [$]P(S_{T_j} = 0) > 0[$] in this model. In a second step they argue that under an additional calendar spread arbitrage condition there exists a discrete martingale generating the observed call prices, thereby proving that the call price matrix is free of (static) arbitrage.

For the arbitrage-checker all that does not really matter, the arbitrage conditions in the paper include the point [$]K_0=0, C_0=S_0[$] in a reasonable way. I was just a little confused about the subsequent derivation of aribtrage-freeness.
 
User avatar
Cuchulainn
Posts: 64681
Joined: July 16th, 2004, 7:38 am
Location: Drosophila melanogaster
Contact:

Re: Carr / Madan: A note on sufficient conditions for no arbitrage

December 21st, 2020, 3:34 pm

Dupire/Gyongy/Local vol model is somewhat specialized theory, in the sense that it requires a diffusion process (or at least an Ito process) for the asset price. But, yes, 2 [$]K[$]-derivatives should be a PDE for [$]q(K,T)[$] if you know the local vol function. And (usually) 0's at the spatial boundaries, as you say.

Breeden-Litzenberger is more general with minimal assumptions. Basically, just need existence of a (norm-preserving) [$]q(K,T)[$] at a given option expiration [$]T[$]. Doesn't require any type of dynamics: diffusions or whatever. The particle can just magically appear at expiration.
I went through the steps in deriving Breeden-Litzenberger, starting from the expectation for call C. Let's instead take the PDE for C and differentiate wrt K and KK to produce 2 pdes with a Heaviside and Dirac, respectively. I have done it for 'normal' delta and gamma by approximating it by fdm. 

Some remarks and questions

1. I don't need to solve FPE to get the transition probability function. I could do it but haven't got around to it yet.
2. My approach looks like  the PDE version of the Derman/Kani probabilistic argument. I would say that they are essentially the same: SDE->Ito->PDE.
3. Computing theta using my PDE approach is not clea. [$]T[$] is nowhere to be found in pde nor payoff. Unless we view [$]\sigma(T)[$] as holding [$]T[$].
4. I'm not sure when/how to pull [$]\sigma^{2}(K,T)[$] out of the hat as it were.
5. I'm curious as to why Carr/Madan start with discrete formulation.

The Continuous Sensitivity Equation (CSE) is discussed here.

https://www.datasim.nl/application/file ... hesis_.pdf

Dupire PDE on interval (0,1)
https://onlinelibrary.wiley.com/doi/epd ... wilm.10014
"Compatibility means deliberately repeating other people's mistakes."
David Wheeler

http://www.datasimfinancial.com
http://www.datasim.nl
 
User avatar
pcaspers
Topic Author
Posts: 712
Joined: June 6th, 2005, 9:49 am

Re: Carr / Madan: A note on sufficient conditions for no arbitrage

December 22nd, 2020, 4:01 pm

On 5 I believe it's because they have a specific goal in mind, namely showing that the conditions they state are sufficient for no-arbitrage. The discrete model they construct is the most parsimonious and natural way to do that.