Butterfly Spread Convexity, Strict Positivity, and No-Arbitrage Conditions
Summary
The discussion examines whether a long call butterfly must have a strictly positive premium under no-arbitrage. A butterfly’s strike-price second difference is nonnegative when call prices are convex in strike. In a smooth model, this difference approximates the second strike derivative, which the Breeden–Litzenberger relation connects to the discounted risk-neutral density. The accepted answer argues for strict positivity when the strike is in the underlying’s reachable domain and the density there is positive.
A second response stresses that nonnegative butterfly prices alone do not characterize absence of arbitrage. It outlines further requirements on call spreads, including nonnegative prices, an upper bound, and a restriction on when their prices can be zero; it also gives an example where a positive butterfly coexists with an arbitrage. The strict-positivity argument needs care: reachability alone does not ensure a positive density at every point, and density-based reasoning assumes appropriate regularity. Thus the document offers useful convexity intuition but not a universal strict-positivity proof.
Key ideas
- Call-price convexity in strike implies nonnegative butterfly premiums under standard assumptions.
- A small-strike-spacing butterfly approximates the second derivative of call price with respect to strike.
- The Breeden–Litzenberger relation links that derivative to the discounted risk-neutral density when regularity conditions hold.
- Strict positivity requires more than reachability if the density may vanish at the strike.
- Nonnegative butterfly premiums alone are insufficient for no-arbitrage; call-spread conditions also matter.
Tags
Full text
# How to prove no-arbitrage when a long butterfly is strictly positive?
# How to prove no-arbitrage when a long butterfly is strictly positive?
I want to prove why there are no arbitrage opportunities when a long butterfly is strictly positive. I know there is a similar topic out there, but it seems it doesn't solve my question: Prove that the butterfly condition is always greater than zero.
By call-put parity, I know
> $C(T,K+∆K)-2C(T,K)+C(T,K-∆K) \geq 0$
is valid, but how to prove strict positivity? I know it makes sense for a long butterfly to be strictly positive in no arbitrage condition. But I just don't know how to get it mathematically. Hope I can get some idea here.
## Answer by Daneel Olivaw (score 5, accepted)
https://quant.stackexchange.com/a/61034
I am not sure why the question you link to does not provide an answer. I’ll try to answer it but it is really similar to what has already been said there. Bottom line is: if the value $K$ is reachable by the underlying asset $S$, that is $K$ belongs to the domain of process $S$, then the butterfly should be strictly positive.
First note that the butterfly is actually an approximation of the second derivative w.r.t. to strike: $$\lim_{h \rightarrow 0}\frac{C(t,K+h)-2C(t,K)+C(t,K-h)}{h^2}=\frac{\partial^2C}{\partial K^2}(t,K)$$ where obviously $h^2>0$. Yet by the Breeden-Litzenberger formula, we know that: $$\frac{\partial^2C}{\partial K^2}(T,K)=e^{-rt}q(t,K)\geq 0$$ where $q$ is the risk-neutral density of the underlying $S$ and $r$ the risk-free rate. You now see that if $K$ is a value which $S$ can reach, that is $K$ belongs to the domain of $S$, then the density of $S$ at $K$ must be strictly positive, that is: $$C(t,K+h)-2C(t,K)+C(t,K-h)\approx h^2e^{-rt}q(t,K)>0$$
Going further, let us introduce the Dirac delta function $\delta$, which is characterized by the following property for any real-valued function $f$: $$\int_{-\infty}^{+\infty}\delta(x)f(x)dx=f(0)$$ Hence the density can be expressed as: $$q(t,K)=\int_{-\infty}^{+\infty}\delta(s-K)q(t,s)ds=E^Q\left(\delta(S_t-K)\right)$$ That is, the risk-neutral density corresponds to the price of a payoff which is non-negative everywhere and strictly positive for one state the world, i.e. if $S_t=K$ $-$ informally the payoff would be infinite if $S_t=K$, see the definition of the Dirac delta. Hence to avoid arbitrage the price of this claim, $e^{-rt}q(t,K)$, must be strictly positive.
## Answer by Cettt (score 4)
https://quant.stackexchange.com/a/61040
as it was stated correctly in the question all long butterfly options have to have a non-negative premium in order for No-arbitrage to hold. So we can say that:
> No-Arbitrage holds implies All Butterfly spreads have a non-negative premium.
However, the reverse is not true. Just because all butterfly spreads have non-negative premiums does not mean that there is No-Arbitrage. There is a well-known paper by Davis and Hobson. They state necessary and sufficient conditions for option prices to fulfill No-Arbitrage. Their notation is a bit hard to get used to. So here is quick (and a bit sloppy) summary of theorem 3.1.
There is no Arbitrage in the market if and only if
- All Butterfly spreads have a non-negative premium
- All Call-spread (i.e Long Option with Strike K1 and Short Option with Strike K2>K1) have a non-negative premium.
- Call spreads are not too expensive (so there is a upper bound on call spread premiums).
- The premium of a call spread can only be zero if both call options have zero premium.
The authors distinguish between model independent arbitrage and weak-arbitrage. They show that if conditions 1., 2. or 3. fail there is model independent arbitrage. If condition 4. fails there is weak arbitrage.
Coming back to you question, here is an easy example of why positive butterfly premiums do not guarantee No-Arbitrage:
Let $K_1 = 1, K_2 = 2, K_3 = 3$ be three strikes and $p_1 = 1, p_2 = 4, p_3 = 8$ be the premiums where $p_i$ is the premium for the call option with strike $K_i$. Also, we assume that the risk-free rate is equal to zero. The corresponding butterfly option $$ \frac 12 C_1 - C_2 + \frac 12 C_3 $$ has a premium of $$ p_{\text{butterfly}} = \frac 12 p_1 - p_2 + \frac 12 p_3 = \frac 12 > 0. $$ But there is an arbitrage opportunity. Simply going long in $C_1$ and short in $C_2$ results in an arbitrage opportunity.
I hope that this was helpfull.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.