Local martingale explained

In mathematics, a local martingale is a type of stochastic process, satisfying the localized version of the martingale property. Every martingale is a local martingale; every bounded local martingale is a martingale; in particular, every local martingale that is bounded from below is a supermartingale, and every local martingale that is bounded from above is a submartingale; however, a local martingale is not in general a martingale, because its expectation can be distorted by large values of small probability. In particular, a driftless diffusion process is a local martingale, but not necessarily a martingale.

Local martingales are essential in stochastic analysis (see Itô calculus, semimartingale, and Girsanov theorem).

Definition

Let

(\Omega,F,P)

be a probability space; let

F*=\{Ft\midt\geq0\}

be a filtration of

F

; let

X\colon[0,infty) x \OmegaS

be an

F*

-adapted stochastic process on the set

S

. Then

X

is called an

F*

-local martingale if there exists a sequence of

F*

-stopping times

\tauk\colon\Omega\to[0,infty)

such that

\tauk

are almost surely increasing:

P\left\{\tauk<\tauk+1\right\}=1

;

\tauk

diverge almost surely:

P\left\{\limk\toinfty\tauk=infty\right\}=1

;

F*

-martingale for every

k

.

Examples

Example 1

Let Wt be the Wiener process and T = min the time of first hit of -1. The stopped process Wmin is a martingale. Its expectation is 0 at all times; nevertheless, its limit (as t → ∞) is equal to -1 almost surely (a kind of gambler's ruin). A time change leads to a process

\displaystyleXt=\begin{cases} Wmin\left(\tfrac{t{1-t},T\right)}&for0\let<1,\\ -1&for1\let<infty. \end{cases}

The process

Xt

is continuous almost surely; nevertheless, its expectation is discontinuous,

\displaystyle\operatorname{E}Xt=\begin{cases} 0&for0\let<1,\\ -1&for1\let<infty. \end{cases}

This process is not a martingale. However, it is a local martingale. A localizing sequence may be chosen as

\tauk=min\{t:Xt=k\}

if there is such t, otherwise

\tauk=k

. This sequence diverges almost surely, since

\tauk=k

for all k large enough (namely, for all k that exceed the maximal value of the process X). The process stopped at τk is a martingale.[1]

Example 2

Let Wt be the Wiener process and ƒ a measurable function such that

\operatorname{E}|f(W1)|<infty.

Then the following process is a martingale:

Xt=\operatorname{E}(f(W1)\midFt)=\begin{cases} f1-t(Wt)&for0\let<1,\\ f(W1)&for1\let<infty; \end{cases}

where

fs(x)=\operatorname{E}f(x+Ws)=\intf(x+y)

1{\sqrt{2\pi
s}}
-y2/(2s)
e

dy.

The Dirac delta function

\delta

(strictly speaking, not a function), being used in place of

f,

leads to a process defined informally as

Yt=\operatorname{E}(\delta(W1)\midFt)

and formally as

Yt=\begin{cases} \delta1-t(Wt)&for0\let<1,\\ 0&for1\let<infty, \end{cases}

where

\deltas(x)=

1{\sqrt{2\pi
s}}
-x2/(2s)
e

.

The process

Yt

is continuous almost surely (since

W1\ne0

almost surely), nevertheless, its expectation is discontinuous,

\operatorname{E}Yt=\begin{cases} 1/\sqrt{2\pi}&for0\let<1,\\ 0&for1\let<infty. \end{cases}

This process is not a martingale. However, it is a local martingale. A localizing sequence may be chosen as

\tauk=min\{t:Yt=k\}.

Example 3

Let

Zt

be the complex-valued Wiener process, and

Xt=ln|Zt-1|.

The process

Xt

is continuous almost surely (since

Zt

does not hit 1, almost surely), and is a local martingale, since the function

u\mapstoln|u-1|

is harmonic (on the complex plane without the point 1). A localizing sequence may be chosen as

\tauk=min\{t:Xt=-k\}.

Nevertheless, the expectation of this process is non-constant; moreover,

\operatorname{E}Xt\toinfty

  as

t\toinfty,

which can be deduced from the fact that the mean value of

ln|u-1|

over the circle

|u|=r

tends to infinity as

r\toinfty

. (In fact, it is equal to

lnr

for r ≥ 1 but to 0 for r ≤ 1).

Martingales via local martingales

Let

Mt

be a local martingale. In order to prove that it is a martingale it is sufficient to prove that
\tauk
M
t

\toMt

in L1 (as

k\toinfty

) for every t, that is,

\operatorname{E}|

\tauk
M
t

-Mt|\to0;

here
\tauk
M
t

=

M
t\wedge\tauk

is the stopped process. The given relation

\tauk\toinfty

implies that
\tauk
M
t

\toMt

almost surely. The dominated convergence theorem ensures the convergence in L1 provided that

style(*)\operatorname{E}\supk|

\tauk
M
t

|<infty

   for every t.Thus, Condition (*) is sufficient for a local martingale

Mt

being a martingale. A stronger condition

style(**)\operatorname{E}\sups\in[0,t]|Ms|<infty

   for every tis also sufficient.

Caution. The weaker condition

style\sups\in[0,t]\operatorname{E}|Ms|<infty

   for every tis not sufficient. Moreover, the condition

style\supt\in[0,infty)\operatorname{E}

|Mt|
e

<infty

is still not sufficient; for a counterexample see Example 3 above.

A special case:

styleMt=f(t,Wt),

where

Wt

is the Wiener process, and

f:[0,infty) x R\toR

is twice continuously differentiable. The process

Mt

is a local martingale if and only if f satisfies the PDE

(

\partial
\partialt

+

12
\partial2
\partialx2

)f(t,x)=0.

However, this PDE itself does not ensure that

Mt

is a martingale. In order to apply (**) the following condition on f is sufficient: for every

\varepsilon>0

and t there exists

C=C(\varepsilon,t)

such that

style|f(s,x)|\leC

\varepsilonx2
e

for all

s\in[0,t]

and

x\inR.

Technical details

  1. For the times before 1 it is a martingale since a stopped Brownian motion is. After the instant 1 it is constant. It remains to check it at the instant 1. By the bounded convergence theorem the expectation at 1 is the limit of the expectation at (n-1)/n (as n tends to infinity), and the latter does not depend on n. The same argument applies to the conditional expectation.

References

. Bernt Øksendal . Stochastic Differential Equations: An Introduction with Applications . Sixth . Springer . Berlin . 2003 . 3-540-04758-1.