# Forms of non-stationary process

Is that true that non-stationary processes only includes trends, cycles, random walks, and combinations of the three? If yes, has this statement been proofed? If not, what are the other forms of non-stationary?

• Do you have a reference for that statement? Dec 4, 2016 at 21:50
• No. This statement is only my guess after reading the webpage investopedia.com/articles/trading/07/stationary.asp Dec 4, 2016 at 22:51

No. There are an infinite number of types of non-stationarity which you can easily dream up. For example, changing variance, changing skewness, level shifts, seasonality in the 6th moment, etc.

Also, cycles can be stationary. Yule (1927) studied cycles in sunspots using stationary AR(2) models, for example.

• Sorry professor I was too exited... But how can I build models, which are specified to stationary processes, only after refusing hypothesis like unit root? As having no unit root, or etc., doesn't rule out the possibility of non-stationary. Is there any reading materials you would kindly recommend me? Dec 4, 2016 at 23:13
• You need to consider a specific class of stationary processes, such as ARMA processes. Read any good book on time series analysis such as Brockwell and Davis: amzn.com/0387953515/?tag=otexts-20 Dec 4, 2016 at 23:17

Is that true that non-stationary processes only includes trends, cycles, random walks, and combinations of the three?

Unless you define one of the three types or combination broader than I do, this would, exclude for example:

• Any superposition of a non-stationary process and a stationary process.

• Any process with a chaotic driver. Such a driver cannot be decomposed into periodic processes.

• If we are talking about discrete-time processes (as implied by cycle): Any process with a quasiperiodic driver.

• Any transient dynamics, i.e., a process that settles on (or asymptotically converges to) some stationary process after a while, e.g., a driven pendulum that takes some time to synchronise with its driver.