Asked in
Niels Bohr
Male Models

Why are the martingale models less restrictive than the random walk model?


User Avatar
Wiki User
May 03, 2009 1:22AM

The error term in a random walk is assumed to be iid (often white-noise), but the error in a martingale doesn't have to be. If the error is AR(1) however, then the process can't be martingale, as the error in last period is known, and so the current period error is not mean zero anymore. But the error may have second order serial correlation (like an ARCH process), and still be a martingale. The error in a random walk however must be independent of the prior error (at all orders).