Kolmogorov’s martingale inequality
Theorem (Kolmogorov’s martingale inequality).
Let , for , be a submartingalewith continuous sample paths.Then for any constant ,
(The notation means , the positive part of .)
Notice the analogy with Markov’s inequality
.Of course, the conclusion
is much stronger than Markov’s inequality,as the probabilistic bound applies to an uncountable numberof random variables
. The continuity and submartingale hypothesesare used to establish the stronger bound.
Proof.
Let be a partition of the interval .Let
and split into disjoint parts ,defined by
Also let be the filtration under which is a submartingale.
Then
definition of | ||||
is submartingale | ||||
is -measurable | ||||
iterated expectation | ||||
monotonicity. |
Since the sample paths are continuous by hypothesis,the event
can be expressed as an countably infinite intersection
of events of the form with finer and finer partitions of the time interval .By taking limits, it followshas the same bound as the probabilities .∎
Corollary.
Let , for , be a square-integrable martingalepossessing continuous sample paths, whoseunconditional mean is .For any constant ,
Proof.
Apply Kolmogorov’s martingale inequality to ,which is a submartingale by Jensen’s inequality.∎