It is important to have a model that predicts periodic behavior, but this sort of prediction is moot if the data don't support such a hypothesis. More often than not, any given data, say a measurement m(t) as a function of time t, will not exhibit true periodicity. There are many possible reasons for this. For example, even if the phenomenon in question is truly periodic, any measurement has some uncertainty in it, and this uncertainty will tend to obscure the periodicity in the data function m(t).

Often, the phenomenon in question is not truly periodic but is the result of many (sometimes a lot and sometimes a few) confluent factors, where each factor is itself periodic, but where the different factors have different periods. For example, if you were studying the behavior of horseshoe crabs on the New England coast, you would have to consider effects that vary on a daily cycle (for example, light and dark), a monthly cycle (for example, tides), and a yearly cycle (for example, the seasons). Thus, you would expect any measured function of time m(t) for horseshoe crabs to be a sum of (at least) three functions, m(t) = md(t) + mm(t) + my(t), which are each periodic, but with periods equal to 24 hours, 29 days, and 365 days, respectively.