Arbitrage-Free Cash Flows (Part of Proof, Geometric Problem)

134 Views Asked by At

(I've seen following notation, used in this question, on YouTube videos and in several undergradute text books, so I will not explain it. It is pretty standard!)

Fix a set of times $0=t_0<\cdots<t_n$ and consider a market consisting of $m$ cash flows: $$\{(c_{k,0},t_0),(c_{k,1},t_1),\ldots,(c_{k,n},t_n)\}$$ for $k=1,\ldots,m$. Since times are fixed represent the cash flows more compactly as $m$ elements $\mathbf c_1,\ldots,\mathbf c_m\in\mathbb R^{n+1}$. Let $\mathbb C$ be a linear subspace of $\mathbb R^{n+1}$ and let $\mathbf c$ be any non-zero linear combination of $\mathbf c_1,\ldots,\mathbf c_m$ such that $\mathbf c\in\mathbb C$.

(i) $\nexists\ \mathbf c\in\mathbb C$ such that $\mathbf c\geq 0$.

(ii) $\exists\mathbf d\in\mathbb R^{n+1}$ with $\mathbf d>0$ satisying $\mathbf c^T\mathbf d=0$ for all $\mathbf c\in\mathbb C$.

(i) implies (ii).

To show (ii) implies (i) is easy. If $\mathbf d>0$ and $\mathbf c^T\mathbf d=0$ then some $\mathbf c\in\mathbb C$ has negative elements.

Problem: How to show (i) implies (ii)? I interpret (i) as there exists no linear combination $\mathbf c$ such that there is arbitrage. But why does $\mathbf c$ need to be restricted to $\mathbb C$? And what is the meaning of (ii)?

1

There are 1 best solutions below

0
On BEST ANSWER

Let's first start by answering a few of the interpretation questions. $\mathbf c_1,\ldots,\mathbf c_m\in\mathbb R^{n+1}$ are cash flows, thus any linear combination of those cash flows represents a certain portfolio. Hence, you are "restricted" to $\mathbb{C}$ because that's the only cash flows we seem to know about, and thus linear combinations of those are the only portfolios we can make.

Next, a claim is made about the nature of $\mathbb{C}$: if $\mathbb{C}$ satisfies condition (i), then it should also satisfy condition (ii) and visa versa.

This theorem is a variant on Farkas' lemma. The usual approach to prove this lemma is to rely on a hyperplane separation theorem. I will not prove the theorem, you can look up a proof for it, but I think it is intuitive enough. It essentially states that two disjoint convex sets can be separated by a hyperplane.

A more precise formulation we can use in the present context:

Let $\mathbb{V}$ and $\mathbb{W}$ be two disjoint non-empty convex subsets of $\mathbb{R}^{n+1}$, then there exists an $\mathbf a \in \mathbb{R}^{n+1}$ with $\mathbf a \neq \mathbf 0$ and a $b \in \mathbb{R}$ such that $\mathbf a^T\mathbf x \leq b , \; \forall \mathbf x \in \mathbb{V}$ and $\mathbf a^T\mathbf x \geq b , \; \forall \mathbf x \in \mathbb{W}$.

To apply this theorem to prove that (i) implies (ii), we'll set $\mathbb{V}=\mathbb{C}$ and $\mathbb{W}=\{\mathbf x \in \mathbb{R}^{n+1}\setminus\{\mathbf 0\}| \mathbf x \geq 0\}$, the cone of all vectors with positive components, null vector excluded. It is clear that those sets satisfy the required condition since by (i) there is no $\mathbf c$ in $\mathbb{C}$ that can also be an element of $\mathbb{W}$. Also a linear subspace is convex and non-empty and it is not hard to show that $\mathbb{W}$ is convex and non-empty as well.

Therefore $\exists \mathbf a \in \mathbb{R}^{n+1}$ with $\mathbf a \neq \mathbf 0$ and a $b \in \mathbb{R}$ such that $\mathbf a^T\mathbf x \leq b , \; \forall \mathbf x \in \mathbb{C}$ and $\mathbf a^T\mathbf x \geq b , \; \forall \mathbf x \in \mathbb{W}$.

Observe that the null vector $\mathbf 0$ is in $\mathbb{C}$ and therefore $\mathbf a^T\mathbf 0 = 0 \leq b$. Thus, we also have for the elements $\mathbf x \in \mathbb{W}$ that $ \mathbf a^T\mathbf x \geq b \geq 0$. In particular, $\mathbb{W}$ contains the standardbasis $\{\mathbb{e}_k\}_{k \in \{0,\ldots,n\}}$. Plugging those in the last inequality shows that all the components of $\mathbf a$ are positive or $\mathbf a \geq 0$.

But this now implies that $b=0$. Indeed, if $b>0$ then the equation $\mathbf a^T\mathbf x = b$ represents a hyperplane cutting off the coordinate axes in positive values, but this would contradict the theorem that the hyperplane separates $\mathbb{W}$ from $\mathbb{C}$.

Therefore we boiled it down to $\exists \mathbf a \in \mathbb{R}^{n+1}$ with $\mathbf a \neq \mathbf 0$ and $\mathbf a \geq 0$ such that $\mathbf a^T\mathbf x \leq 0 , \; \forall \mathbf x \in \mathbb{C}$ and $\mathbf a^T\mathbf x \geq 0 , \; \forall \mathbf x \in \mathbb{W}$.

Finally, suppose $\exists \mathbf c \in \mathbb{C}$ such that $\mathbf a^T\mathbf c < 0$. But since $\mathbb{C}$ is a linear subspace, $-\mathbf c$ is also an element of $\mathbb{C}$ and $\mathbf a^T(-\mathbf c) \leq 0$. But simply multiplying the equation for $\mathbf c$ by $(-1)$ leads to a contradiction. Therefore $\forall \mathbf x \in \mathbb{C}: \mathbf a^T\mathbf x = 0$.

So, we are done with the proof, except for one detail: we need $\mathbf a >0$. But we have $\mathbf a \geq 0$. I'll leave that for you to handle.

So, we see that (i) implies that the linear subspace $\mathbb{C}$ is a subspace of a hyperplane that does not intersect with the cone $\mathbb{W}$. That's the geometrical interpretation for (ii).