I am trying to learn by myself stochastic calculus, and I have a question regarding semi-martingale stochastic process and geometric Brownian motion (GBM).

We know that a stochastic process $S_t$ is said to follow a GBM if it satisfies the following stochastic differential equation (SDE): $\textrm{d}S_t = \mu_t S_t \textrm{d}t+\sigma S_t \textrm{d}W_t$ where $W_{t}$ is a Brownian motion

I was reading a paper where the authors assumed that under the null hypothesis, the real-valued logarithmic efficient price process denoted by $p_t^e$ is given by the Brownian semi-martingale: $\textrm{d}p_t^e = \mu_t \textrm{d}t+\sigma \textrm{d}W_t$

My question is, what is the difference between a GBM and a semi-martingale stochastic process? Are they the same stochastic process?


1 Answer 1


I will provide a heuristic explanation.

A local martingale $(M_t)_t$ (mart) is a stochastic process whose expected increments are zero that is $E(\mathrm{d}M_t)=0$. Under suitable technical conditions, a local martingale is also a (true) martingale that is the current value is the best estimate for its future value: $E(M_t)=M_0$. The difference between the two cases is roughly that a local martingale might have paths under which it experiences explosive behavior.

A process $(A_t)_t$ with locally bounded variation (BV) is roughly speaking a process which is "relatively smooth" in the sense that it never experiences an arbitrarily large jump over a very small time interval. This process is also required to be right-continuous with left limits and adapted.

A semimartingale $(S_t)_t$ is a stochastic process allowing the decomposition: $$S\equiv M+A$$ i.e. it can be written as the sum of a martingale and a bounded variation process. Geometric Brownian Motion (GBM) is a semimartingale: $$\textrm{d}S_t=\underbrace{\mu S_t\textrm{d}t}_{\text{BV}}+\underbrace{\sigma S_t\textrm{d}W_t}_{\text{mart}}$$ Indeed on every infinitesimal interval $\textrm{d}t$ the GBM (1) increases/decreases by a proportional rate $\mu$, the amount being smaller the smaller the interval or the value $S_t$, nothing jumpy going on here, and (2) a random amount which depends on the Brownian Motion $W$ whose expected increment is zero: $E(\textrm{d}W_t)=0$.

Semimartingales are important in the theory of stochastic process because they are nice to integrate. Intuitively, they are made up of an "unbiased" stochastic term with no upward/downward trend, and a trend component which is restricted to bounded variation to avoid being explosive - and thus being integrable.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.