I know that an $I(1)$ process becomes stationary after differencing once. However, I somehow always equated that to its being a random walk because say having a unit root process like \begin{eqnarray} y_{t} = y_{t-1} + \epsilon_t \end{eqnarray} is clearly $I(1)$ and becomes converted to an $MA_\infty$ $y_t = \sum_{s=0}^t \epsilon_{t-s}$, which strikes me as a random walk as discussed here for example.

Now, is there a difference between a random walk a process $I(1)$?


If I understand it correctly, a random walk is a special case of an $I(1)$ process.

A random walk can be written as


as you have done it. The increments $\varepsilon_t$ have to be independent of each other over time, e.g. $\varepsilon_t \sim i.i.N(0,\sigma^2$).

In case of a $I(1)$, the increments need not be independent; they have to be stationary, e.g. $\varepsilon_t$ follows ARMA($p,q$).

Hence, random walk is a special case of an $I(1)$ process. For example, a random walk is an ARIMA($0,1,0$); meanwhile, an $I(1$) process can be ARIMA($p,1,q$) with $p,q \geqslant 0$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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