Martingale Transform: A Key to Predictable Bonds

Martingale Transform: A Key to Predictable Bonds

Finance Published: October 13, 2001
CQUALMS

The Martingale Transform

In a martingale setting, the martingale transform is a crucial concept that helps us analyze the behavior of random variables under certain conditions.

Let's start with a simple example. Consider a martingale relative to a filtration (Fn)n≥0, denoted by Zn,n ≥ 0. The martingale transform Y · Zn is defined as follows:

(Y · Zn) = Z0 + n Xj=1 Yj(Zj −Zj−1)

To show that if the random variables Yn are bounded, then the martingale transform (Y · Zn)n is a martingale relative to (Fn)n≥0, we need to establish some conditions.

Firstly, suppose the random variables Yn are bounded. This means there exists a constant M such that |Yj| ≤ M for all j ≥ 1. Now, let's consider the martingale transform:

(Y · Zn) = Z0 + n Xj=1 Yj(Zj −Zj−1)

Since the random variables Yn are bounded, we can substitute the upper bound M into the expression above:

(Y · Zn) ≤ (Z0 + nM)

Now, let's consider a stopping time τ. Suppose τ is a stopping time such that Zn∧τ = 0 for some n ≥ 1. We need to show that the sequence Yn is predictable.

To do this, we can use the Optional Stopping Formula:

E(Zτ∧n) = EZt∧n

where Ez is the expectation of Zt over the filtration (Fn) up to time τ. Since Zn∧τ = 0 for some n ≥ 1, we have E(Zτ∧n) = 0.

Now, let's consider the martingale transform:

(Y · Zn) = Yn+1 - Yn

Using the Optional Stopping Formula again, we get:

E((Y · Zn)∧(τ ∧ n)) = E(Yn+1 - Yn | τ ∧ n)

Since Zn∧τ = 0 for some n ≥ 1, we have E((Y · Zn)∧(τ ∧ n)) = 0.

Now, let's use the Optional Stopping Formula to establish the martingale property of the sequence (Zt) in general:

E(Zt∧n) = EZt∧n

Since Yj is predictable and bounded, we can conclude that the sequence Zt is a martingale relative to the usual filtration.

Bonds: A Portfolio Perspective

Now let's consider a specific case: bonds. A bond with maturity M is a contract that pays the owner $1 at the maturity date M. We want to find a formula for the price (in dollars) of one maturity-M bond at time t ≤ M.

Using the martingale transform, we can write:

Bt,M = Bt,M(ω)

= E[Bt,M|Fn]

Since Fn is a filtration, we can rewrite this as:

Bt,M = E[Bt,M|Z0 + 1Xj=1 Yj(Zj −Zj−1)]

Using the martingale transform definition again, we get:

Bt,M = Z0 + 1E[∑(Yj(Zj −Zj−1))|Fn]

Now, let's consider a coupon-bearing bond. The price at time t is Ct dollars. We want to find a formula for the price at time t = 0 of such a contract.

Using the martingale transform, we can write:

Bt,TQ0+1Xj=1 Yj(Zj −Zj−1)

= Z0 + TqE[∑(Yj(Zj −Zj−1))|Fn]

Now, let's use the Optional Stopping Formula to establish the martingale property of the sequence Sn:

Sn = ∑(Bt,TQ0+1Xj=1 Yj(Zj −Zj−1))

E(Sn) = E[E(Sn|Ft)]

Using the Optional Stopping Formula again, we get:

E(Sn) = EZT∧∞

Now, let's use the martingale transform to establish the martingale property of Zt:

Zt = 1 Bt,TQ0+1Xj=1 Bj,j+1

Using the Optional Stopping Formula again, we get:

EZt = Ez∞

Comparing these two expressions for E(Sn) and EZT∧∞, we can conclude that Sn is a martingale relative to (Fn)≥0.

Gambler’s Ruin: Revisited

The game of gambler's ruin is a classic example of an infinite-horizon problem. Suppose the coin is unfair with probability p > 1/2 and the probability of winning is q = 1 −p < 1/2. We want to determine the probability that Slim exhausts his initial fortune B before Fats exhausts his initial fortune A.

Let's define Xi as the cumulative change in Fats' fortune after n plays. The martingale transform Zn := q p Sn is a martingale relative to (Fn)≥0.

To show that Zn is a martingale, we need to establish some conditions.

Firstly, suppose the initial fortunes are B and A, respectively. We want to show that the sequence Sn is predictable.

Using the martingale transform definition again, we get:

Sn = 1 Bn+1 + q p (Bn−1 - B)

Now, let's consider a stopping time τ such that Sn = +B or Sn = −A. We need to show that the sequence Zn is predictable.

To do this, we can use the Optional Stopping Formula:

E(Zτ∧n) = EZt∧n

Since Zn is a martingale, we have E(Zτ∧n) = 1 for any n ≥ 0.

Now, let's consider the martingale transform:

Zn := q p Sn

Using the Optional Stopping Formula again, we get:

E((Zn∧(τ ∧ n))|Fn) = Ezt∧n

Since Zn is a martingale, we have E((Zn∧(τ ∧ n))|Fn) = 1 for any n ≥ 0.

Now, let's use the Optional Stopping Formula to establish the martingale property of Sn:

Sn = ∑(Bt,Tq p  Yj(Zj −Zj−1))

E(Sn) = E[E(Sn|Ft)]

Using the Optional Stopping Formula again, we get:

E(Sn) = EzT∧∞

Now, let's use the martingale transform to establish the martingale property of Zn:

Zn := q p Sn

Using the Optional Stopping Formula again, we get:

EZn = Ez∞

Comparing these two expressions for E(Sn) and EZT∧∞, we can conclude that Sn is a martingale relative to (Fn)≥0.

Conclusion

In conclusion, we have analyzed several key concepts in the context of MathFinance 345/Stat390. We explored the martingale transform, its implications on portfolios, bonds, and gambler's ruin. Our analysis highlights the importance of understanding these concepts to make informed decisions in finance.

← Back to Research & Insights